test: reject unresolved MSI authoring and invalid component identity

* fix(studio): simplify voice cloning setup

* docs: link cloning redesign changelog

* fix(studio): keep clone recording controls available

* fix(studio): lock clone capture transitions

.claude/skills/omnivoice/SKILL.md

@@ -5,6 +5,9 @@ description: "Local TTS, voice cloning, voice design, and video dubbing via the
 
 # VoiceStudio
 
+The canonical cross-agent package lives at `skills/omnivoice/SKILL.md`. This
+Claude-specific package retains the MCP lifecycle helpers and references.
+
 ## Overview
 
 Generate audio locally via the VoiceStudio MCP server. Tools: `generate_speech`, `list_voices`, `list_personalities`, `list_languages`, `check_health`. Resources: `voice://{id}`, `history://recent`.
@@ -166,4 +169,4 @@ The MCP server does not expose the dubbing endpoint. The full transcribe → tra
 
 Backend Swagger / OpenAPI: `http://127.0.0.1:3900/docs` (when backend is up).
 
-Upstream: github.com/debpalash/VoiceStudio — FSL-1.1-ALv2 (free for personal/internal/non-commercial; auto-converts to Apache-2.0 two years after each release).
+Upstream: github.com/debpalash/VoiceStudio. The app uses AGPL-3.0-only; optional engines and downloaded models retain their own licenses. See `LICENSE-NOTICE.md` in the repository.

.claude/skills/omnivoice/scripts/start-backend.sh

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 # Start the OmniVoice FastAPI backend on 127.0.0.1:3900, detached, idempotent.
-# Honors $OMNIVOICE_HOME (default ~/OmniVoice-Studio).
+# Honors $OMNIVOICE_HOME (default ~/VoiceStudio).
 #
 # Exit codes:
 #   0   success (already running, or freshly started + healthy within 60s)
@@ -11,7 +11,7 @@
 
 set -euo pipefail
 
-HOME_DIR="${OMNIVOICE_HOME:-$HOME/OmniVoice-Studio}"
+HOME_DIR="${OMNIVOICE_HOME:-$HOME/VoiceStudio}"
 URL="${OMNIVOICE_API_URL:-http://127.0.0.1:3900}"
 LOG="$HOME_DIR/backend.log"
 

.github/CONTRIBUTING.md

@@ -40,6 +40,7 @@ sudo apt-get install -y \
   libwebkit2gtk-4.1-dev libgtk-3-dev libpango1.0-dev libcairo2-dev \
   libsoup-3.0-dev libgdk-pixbuf-2.0-dev \
   libayatana-appindicator3-dev librsvg2-dev libssl-dev libxdo-dev \
+  gstreamer1.0-plugins-good \
   libasound2-dev build-essential curl wget file
 ```
 

.github/SECURITY.md

@@ -4,7 +4,7 @@
 
 | Version | Supported |
 |---------|-----------|
-| 0.3.x (latest release + `main` previews) | ✅ Current — all fixes land here |
+| 0.5.x (latest release + `main` previews) | ✅ Current — all fixes land here |
 | 0.2.7 | ⚠️ Legacy stable — security fixes only, upgrade recommended |
 | < 0.2.7 | ❌ No longer supported |
 

.github/workflows/build-omnivoice-tts.yml

@@ -51,6 +51,13 @@ jobs:
           - os: ubuntu-latest
             platform: linux-x86_64
             experimental: false
+          - os: ubuntu-24.04-arm
+            platform: linux-aarch64
+            # Apple Silicon under Asahi Linux. Experimental: the Vulkan
+            # (Honeykrisp GPU) build path is new and the hosted arm64
+            # runner has no GPU — it validates that the binary builds;
+            # on-host Vulkan acceleration is exercised by users.
+            experimental: true
           - os: windows-latest
             platform: windows-x86_64
             experimental: false
@@ -80,11 +87,18 @@ jobs:
       # Linux-only: upstream `buildcpu.sh` enables `-DGGML_BLAS=ON` which
       # requires a system BLAS implementation at cmake configure time.
       - name: Linux system deps (BLAS for ggml-blas backend)
-        if: matrix.platform == 'linux-x86_64'
+        if: startsWith(matrix.platform, 'linux')
         run: |
           sudo apt-get update
           sudo apt-get install -y libopenblas-dev pkg-config
 
+      # linux-aarch64: let the build script's Vulkan path (Honeykrisp GPU
+      # on Asahi) engage instead of silently falling back to CPU.
+      - name: Vulkan dev deps (linux-aarch64 GPU backend)
+        if: matrix.platform == 'linux-aarch64'
+        run: |
+          sudo apt-get install -y glslc libvulkan-dev spirv-headers
+
       - name: Build omnivoice-tts
         shell: bash
         # Pass values through env (quoted) rather than ${{ }} interpolation

.github/workflows/ci.yml

@@ -11,6 +11,11 @@ on:
   push:
     branches: [main]
   workflow_dispatch:
+    inputs:
+      windows_wix_diagnostic:
+        description: Run only the tiny nonpublishing Windows MSI authoring diagnostic
+        type: boolean
+        default: false
 
 permissions:
   contents: read
@@ -21,6 +26,7 @@ env:
 
 jobs:
   test:
+    if: ${{ !inputs.windows_wix_diagnostic }}
     name: Tests (backend + frontend)
     runs-on: ubuntu-22.04
     env:
@@ -189,6 +195,7 @@ jobs:
   # and `cargo test --lib` runs the shell's unit tests natively on each OS.
   # Full bundling stays in release.yml on tag push.
   tauri-cross-platform:
+    if: ${{ !inputs.windows_wix_diagnostic }}
     name: Tauri shell check (${{ matrix.label }})
     needs: test
     strategy:
@@ -289,6 +296,7 @@ jobs:
   # job above misses. Narrow scope (tests/smoke/ only) — full pytest stays
   # on Linux until Phase 1's INST-01 lands setuptools for WhisperX.
   smoke-matrix:
+    if: ${{ !inputs.windows_wix_diagnostic }}
     name: Smoke (${{ matrix.label }})
     needs: test
     strategy:
@@ -442,3 +450,24 @@ jobs:
         env:
           HF_HUB_OFFLINE: "1"
           HF_HUB_CACHE: ${{ runner.temp }}/worker-artifact-empty-hf-cache
+
+  windows-wix-diagnostic:
+    name: Windows MSI authoring (no publishing)
+    needs: test
+    if: ${{ !cancelled() && (inputs.windows_wix_diagnostic || needs.test.result == 'success') }}
+    runs-on: windows-2022
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@v4
+      - uses: oven-sh/setup-bun@v1
+      - name: Bundle canonical system and per-user templates with a tiny payload
+        shell: pwsh
+        run: ./scripts/diagnose-windows-wix.ps1
+      - name: Preserve verbose linker output and rendered authoring
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: windows-wix-diagnostic
+          path: wix-diagnostic-artifacts/
+          if-no-files-found: warn
+          retention-days: 3

.github/workflows/release.yml

@@ -148,15 +148,20 @@ jobs:
   preview-gate:
     name: Preview gate
     runs-on: ubuntu-22.04
+    permissions:
+      contents: read
     outputs:
       is_preview: ${{ steps.decide.outputs.is_preview }}
       proceed: ${{ steps.decide.outputs.proceed }}
+      stable_tag: ${{ steps.decide.outputs.stable_tag }}
     steps:
       - uses: actions/checkout@v4
         with:
           fetch-depth: 50
       - id: decide
         shell: bash
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         run: |
           set -euo pipefail
           event="${{ github.event_name }}"
@@ -171,6 +176,13 @@ jobs:
               exit 1
             fi
             echo "is_preview=true" >> "$GITHUB_OUTPUT"
+            # Resolve once before the matrix starts so every platform stamps
+            # against the same immutable Stable-channel snapshot.
+            STABLE_TAG=$(gh release view --repo "$GITHUB_REPOSITORY" --json tagName --jq .tagName)
+            [[ "$STABLE_TAG" =~ ^v[0-9]+\.[0-9]+\.[0-9]+$ ]] || {
+              echo "::error::latest stable release has an invalid tag"; exit 1;
+            }
+            echo "stable_tag=$STABLE_TAG" >> "$GITHUB_OUTPUT"
           else
             echo "is_preview=false" >> "$GITHUB_OUTPUT"
           fi
@@ -497,35 +509,22 @@ jobs:
             echo "APPLE_TEAM_ID=$TID"
           } >> "$GITHUB_ENV"
 
-      # Stamp each preview build with a unique, monotonically increasing semver
-      # PRERELEASE so the updater actually offers it (a rolling preview that
-      # always reported the static 0.3.0 never looked "newer", so no update was
-      # ever delivered). Ephemeral, CI-only — never committed. Tauri reads the
-      # bundle + updater version from tauri.conf.json, so rewriting it here
-      # stamps the artifacts + latest.json. Under the versioning hard rule
-      # (owner-set 2026-06-11) main is always last-release + 1, so BASE-N is a
-      # prerelease of the NEXT version and semver-sorts ABOVE the last stable
-      # (0.3.6-N > 0.3.5) — preview users naturally upgrade past stable, and
-      # the Windows MSI ProductVersion (which strips the prerelease → 0.3.6)
-      # is also correctly above the last stable.
+      # Stamp each preview with a numeric prerelease that is strictly above the
+      # latest stable release. Main may intentionally retain the released
+      # version while AUTO_VERSION_BUMP is disabled; in that case the helper
+      # advances the preview base by one patch so stable users can still opt in
+      # and receive it. The edit is ephemeral and never committed.
       - name: Stamp preview version
         if: needs.preview-gate.outputs.is_preview == 'true'
         shell: bash
+        env:
+          STABLE_TAG: ${{ needs.preview-gate.outputs.stable_tag }}
         run: |
           set -euo pipefail
-          # package.json is the single source of truth; tauri.conf.json reads its
-          # version from it ("version": "../package.json"), so stamping
-          # package.json restamps the whole bundle.
-          CONF=frontend/package.json
-          BASE=$(jq -r .version "$CONF")
-          # MSI/WiX requires the semver pre-release identifier to be numeric-only
-          # (and <= 65535). "preview.N" hard-fails the Windows bundler, so the
-          # preview stamp is BASE-N — still sorts below the stable BASE for the
-          # updater, still unique per run.
-          PREVIEW_VERSION="${BASE}-${{ github.run_number }}"
-          tmp=$(mktemp)
-          jq --arg v "$PREVIEW_VERSION" '.version = $v' "$CONF" > "$tmp"
-          mv "$tmp" "$CONF"
+          PREVIEW_VERSION=$(python scripts/stamp-preview-version.py \
+            --package-json frontend/package.json \
+            --stable-tag "$STABLE_TAG" \
+            --run-number "${{ github.run_number }}")
           echo "Stamped preview version: $PREVIEW_VERSION"
 
       # The rolling `preview` release is REUSED every night, and macOS updater
@@ -649,6 +648,47 @@ jobs:
           updaterJsonPreferNsis: false
           includeUpdaterJson: true
 
+      - name: Build per-user Windows MSI
+        if: runner.os == 'Windows'
+        shell: bash
+        working-directory: frontend
+        env:
+          TAURI_SIGNING_PRIVATE_KEY: ${{ secrets.TAURI_SIGNING_PRIVATE_KEY }}
+          TAURI_SIGNING_PRIVATE_KEY_PASSWORD: ${{ secrets.TAURI_SIGNING_PRIVATE_KEY_PASSWORD }}
+        run: |
+          set -euo pipefail
+          python ../scripts/render-per-user-wix.py \
+            --source src-tauri/wix/main.wxs \
+            --system-wxs src-tauri/target/${{ matrix.rust_target }}/release/wix/x64/main.wxs \
+            --output src-tauri/target/wix-per-user/main.wxs
+          bunx tauri build --target ${{ matrix.rust_target }} --bundles msi \
+            --config src-tauri/tauri.per-user.conf.json
+          DIR="src-tauri/target/${{ matrix.rust_target }}/release/bundle/msi"
+          while IFS= read -r artifact; do
+            safe=${artifact// (Current User)/_Current_User}
+            [ "$safe" = "$artifact" ] || mv "$artifact" "$safe"
+          done < <(find "$DIR" -maxdepth 1 -type f -name '*Current*User*.msi*')
+
+      - name: Publish per-user Windows updater channel
+        if: runner.os == 'Windows'
+        shell: bash
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          RELEASE_TAG: ${{ (needs.preview-gate.outputs.is_preview == 'true') && 'preview' || github.ref_name }}
+        run: |
+          set -euo pipefail
+          DIR="frontend/src-tauri/target/${{ matrix.rust_target }}/release/bundle/msi"
+          MSI=$(find "$DIR" -name '*Current*User*.msi' -type f | head -1)
+          [ -n "$MSI" ] || { echo "per-user MSI missing"; find "$DIR" -type f; exit 1; }
+          [ -f "$MSI.sig" ] || { echo "per-user MSI signature missing"; exit 1; }
+          VERSION=$(jq -r .version frontend/package.json)
+          python scripts/build_windows_user_manifest.py \
+            --repo "$GITHUB_REPOSITORY" --tag "$RELEASE_TAG" --version "$VERSION" \
+            --asset "$(basename "$MSI")" --signature-file "$MSI.sig" \
+            --output latest-user.json
+          gh release upload "$RELEASE_TAG" "$MSI" "$MSI.sig" latest-user.json \
+            --clobber --repo "$GITHUB_REPOSITORY"
+
       # ── Installer smoke (Phase 0 GATE-03) ─────────────────────────────
       # Structural verification of the installed/extracted bundle. The thin
       # uv-venv installer ships NO frozen backend binary (the venv is built on
@@ -717,8 +757,9 @@ jobs:
         shell: bash
         run: |
           set -euo pipefail
-          MSI=$(find frontend/src-tauri/target/${{ matrix.rust_target }}/release/bundle/msi -name "*.msi" | head -1)
+          MSI=$(find frontend/src-tauri/target/${{ matrix.rust_target }}/release/bundle/msi -name "*.msi" ! -name '*Current*User*' | head -1)
           echo "Smoke-testing MSI: $MSI"
+          powershell.exe -NoProfile -ExecutionPolicy Bypass -File scripts/verify-windows-msi.ps1 -MsiPath "$(cygpath -w "$MSI")"
           # /quiet = no UI, /norestart = don't reboot the runner if a dep asks
           msiexec.exe //i "$(cygpath -w "$MSI")" //quiet //norestart
           INSTALL="/c/Program Files/VoiceStudio"
@@ -731,6 +772,16 @@ jobs:
           find "$INSTALL" -type f -path '*backend*main.py' | grep -q . || fail "backend source main.py missing"
           echo "OK — MSI installed shell + uv + backend resources"
 
+      - name: Per-user installer smoke (Windows, non-admin account)
+        if: runner.os == 'Windows'
+        timeout-minutes: 8
+        shell: bash
+        run: |
+          set -euo pipefail
+          MSI=$(find frontend/src-tauri/target/${{ matrix.rust_target }}/release/bundle/msi -name '*Current*User*.msi' | head -1)
+          powershell.exe -NoProfile -ExecutionPolicy Bypass \
+            -File scripts/smoke-per-user-msi.ps1 -MsiPath "$(cygpath -w "$MSI")"
+
       # linuxdeploy re-links .DirIcon as an ABSOLUTE symlink into the build
       # machine AFTER tauri's files-map has placed the real icon bytes — the
       # exact bug #1518 guarded against, resurfacing on the first real tag

.gitignore

@@ -159,6 +159,13 @@ tests/probe/reports/
 # reports). Working notes for whoever is driving a change, not a repo artifact.
 /remote/
 
+# OmniVoice GGUF runtime build artifacts (scripts/build-omnivoice-tts.sh).
+# Only 0-byte placeholders of omnivoice-tts-* are tracked; real binaries,
+# the checksums manifest and the copied libggml shared libs ship via CI.
+bin/libggml*
+bin/checksums.sha256
+bin/omnivoice-tts-linux-aarch64
+
 # Dubbing-demo intermediates. The .mp4/.srt/manifest.json in this directory ARE
 # committed (they ship with the app); the per-language source WAVs are just the
 # inputs scripts/render_dub_demo_audio.py hands to scripts/build_dub_demo.sh.

CHANGELOG.md

@@ -10,6 +10,101 @@ the frozen-backend fallback mirror it for their toolchains.
 
 **Highlights**
 
+- Fix current-user Windows installer validation and nested resource cleanup (#730)
+
+- Voice cloning now starts with a clear upload-or-record choice, reveals recording and reference details only when needed, and keeps sampling controls under Production Overrides (#1817)
+
+### Changed
+
+### Added
+
+### Docs
+
+### Fixed
+
+
+## [0.5.2] — 2026-09-02
+
+**Highlights**
+
+- Show estimated and measured model, dependency, cache, and temporary disk costs in the engine catalogue (#1718)
+- Preview builds now stay newer than Stable even when automatic post-release version bumps are disabled (#1762)
+- CosyVoice setup guidance now separates downloaded model files from the runtime that makes the engine available (#1761)
+- MCP tools can now keep audio out of agent context by returning files and accepting base-path-confined file inputs (#1760) — thanks @agudmund!
+- Hear a dub line as you type it — an opt-in live preview streams TTS for the edited segment (#1769) — thanks @mvanhorn!
+- Studio gains a Convert method: re-say any clip in one of your saved voices, speech to speech, fully local (#1765) — thanks @mvanhorn!
+- Hardsub video export gains an opt-in karaoke word-highlight caption style (#1764) — thanks @mvanhorn!
+- The batch queue can now watch a folder: new videos dropped into it are dubbed automatically (#1768) — thanks @mvanhorn!
+- The audiobook player now shows the chapter text and highlights the word being narrated (#1766) — thanks @mvanhorn!
+- The dub editor gains a casting board: drag voice chips onto speakers, dropdowns stay in sync (#1767) — thanks @mvanhorn!
+
+### Changed
+
+- Voice Design simplified: the 12-row fine-grained block collapses to one summary line with a five-field editor, English accent and Chinese dialect merge into a single field, and the starting-point chips now show 5 with an overflow toggle (#1793)
+
+### Added
+
+- The audiobook result is now a synced-lyrics player: chapter text follows playback with the current word highlighted and click-to-seek, timed from the render's own chapter durations with a karaoke-style even split — no ASR pass, fully local (#1766) — thanks @mvanhorn!
+- The dub CAST strip expands into a project-level casting board: drag voice chips (clone profiles, design presets, Default) onto speaker rows — or pick from a keyboard listbox — writing the same per-speaker cast fields as the existing dropdowns (#1767) — thanks @mvanhorn!
+- Studio's new Convert method turns a dropped or recorded clip into an existing voice profile's voice, with optional source-duration matching (#1765) — thanks @mvanhorn!
+- Opt-in watch folder on the batch queue: pick a directory once and new videos are auto-enqueued with your last Add-to-queue settings, with pause/stop controls and copy-in-progress protection — files upload as bytes, paths never leave the app (#1768) — thanks @mvanhorn!
+- Hardsub export can now burn karaoke word-highlight captions: an opt-in Line | Karaoke control renders a word-timed ASS sweep from timings persisted at transcription, with an even-split fallback for older jobs and translated tracks, plus a `GET /dub/ass/{job_id}` sidecar (#1764) — thanks @mvanhorn!
+- Windows releases now include an independently updatable per-user MSI that installs and uninstalls without elevation (#1713)
+- Dub segments can now stream live TTS while you edit a translated line — opt-in toggle, existing `/ws/tts` socket, shared generation admission, exports still render at full quality (#1769) — thanks @mvanhorn!
+- Engine status and diagnostic bundles now record loaded execution provider, device, precision, fallback stage, accelerator identity, runtime versions, and parent-process memory visibility (#1717)
+
+### Docs
+
+- Local gigastt is now documented as a supported OpenAI-compatible ASR endpoint, with loopback privacy distinguished from remote servers (#1736) — thanks @ekhodzitsky!
+- The CosyVoice guide now states that packaged builds have no one-click runtime installer and records the exact readiness checks exposed by [Discussion 1631](https://github.com/debpalash/VoiceStudio/discussions/1631) (#1761)
+- A production private-API guide now covers pinned containers, root credentials, network isolation, streaming proxies, health checks, upgrades, and benchmark evidence (#1720)
+- RX 6700 XT/gfx1031 over WSL2 ROCDXG is now explicitly unverified until a published end-to-end GPU workload proves the mapped path (#1716)
+
+### Fixed
+
+- The generation compute-time budget is now a Settings control (Performance & Device) instead of an env-var-only setting the timeout error recommended with no UI path — the error copy points there too, and long CPU/MPS renders get an upfront heads-up before they start (#1787)
+- Windows: the backend can now start when the install path contains non-English characters (e.g. a CJK username) on a non-UTF-8 system code page — a new or broken Python environment now builds at an ASCII-safe path automatically (a healthy existing one is never relocated), and a specific error message names the cause and a working fix if the interpreter still crashes in `site` (#1783)
+- Exports and other native-picker actions no longer 403 with "Invalid or expired desktop authorization" when the desktop app and backend resolve different data directories, e.g. dev mode or a custom data folder (#1781)
+- Voice Design no longer lets you pick a Chinese dialect and an English accent together — the picker keeps them mutually exclusive instead of round-tripping a 400 (#1771)
+- The desktop app no longer attaches to an already-running backend on version string alone: it now verifies the backend's actual code fingerprint too, so an orphaned or manually started backend reporting the current version but running older code (e.g. a stale `destination_path` export 422) gets replaced instead of adopted (#1770)
+- Korean locale overhauled: 231 mistranslations corrected and all 493 missing keys translated (#1776) — thanks @j30231!
+- Japanese "Cleaning…" clone status now reads as denoising instead of housekeeping (#1775) — thanks @j30231!
+- The batch dubbing queue now has a UI entry point — a quiet link on the Dub landing (it was previously unreachable: the app switched on a mode nothing ever set) (#1768) — thanks @mvanhorn!
+- OpenAI-compatible ASR now requires HTTPS outside loopback and refuses redirects so audio stays on the configured origin (#1736)
+- Windows isolated engines now retain direct Job ownership without an extra Python supervisor process that can deadlock the child loader (#1734)
+- The setup splash now waits through the backend's full startup budget instead of reporting slow Windows CUDA initialization as stuck after two minutes (#1749)
+- Dubbing jobs can now reuse every source-language code produced by automatic ASR detection without a 400 error on the next upload (#1737)
+- Incomplete Sherpa-ONNX model snapshots now self-repair before recognizer startup instead of failing on a missing ONNX file (#1733)
+- OmniVoice subprocess startup now allows slow packaged Windows Python runtimes to signal readiness before termination (#1711)
+- SRT files selected during source analysis now wait for speaker cloning, then replace transcript text without losing voices (#1709)
+- Windows MSI deployments can now prohibit WebView2 bootstrap with `DISABLEWEBVIEW2BOOTSTRAP=1`, and `AUTOLAUNCHAPP=0` reliably suppresses first launch (#1714)
+- Subtitle rows now provide 100 ms timing steppers and flag adjacent overlaps without requiring precise timeline dragging (#1710)
+- Repair-sync failures now retain uv's final dependency error instead of reporting only an opaque exit status (#1705)
+- YouTube ingest now retries yt-dlp's transient “page needs to be reloaded” response (#1706)
+- Dictation model readiness now follows the live Hugging Face cache selected in Settings (#1707)
+- Dictation capture now queues native events whenever its webview listener unmounts or reloads instead of emitting them to nobody (#1707)
+- Desktop-contained backends now exit when their owning app disappears instead of surviving as stale port-3900 processes (#1707)
+
+## [0.5.1] — 2026-08-28
+
+**Highlights**
+
+- OmniVoice generation on Apple Silicon now runs in a crash-isolated child, so fatal MPS memory exits no longer take down the local backend (#1697, #1698) — thanks @ndntran14!
+- Model-load GPU exhaustion now returns a sanitized, actionable dubbing error, and readiness correctly attributes the shared model status to TTS (#1695)
+- Source-mode development now restarts an isolated backend crash without tearing down the UI, while repeated crash loops still stop loudly with diagnostics (#1690)
+- Dubbing playback now keeps an audible companion source when a WebView can render the preview picture but cannot decode its audio (#1692)
+- Model Catalogue engine rows now use the available desktop width and keep identity, runtime state, and actions from crowding one another (#1689)
+- VoiceStudio now acts as a local speech platform: other apps can trigger its native dictation or connect through versioned HTTP, WebSocket, JSON-RPC, CLI, and MCP transports (#1646)
+- A timed-out in-process dub transcription no longer starts a second WhisperX/CTranslate2 call over the abandoned native worker, preventing the overlapping access that preceded Windows `0xC0000005` exits (#1669)
+- Windows debugger termination code `0x40010004` is no longer misreported as a backend crash or charged against automatic restart recovery (#1663)
+- Studio now keeps one generation reservation across page changes, preventing a remount from stacking native jobs until the backend reports capacity busy or is killed under memory pressure (#1670)
+- Uploaded dubbing videos are normalized to browser-safe H.264/AAC before preview, preventing valid VP9, AV1, or Opus media from failing with “no supported sources” (#1644)
+- Dubbing now separates spoken and target languages, preserves translations through segment cleanup, and lets failed translations be retried or skipped without restarting the batch (#1654) — thanks @Number16BusShelter!
+- Importing replacement SRT subtitles now keeps each cue bound to the best-overlapping source speaker and clone instead of resetting every line to a random default voice (#1660) — thanks @invio-a11y!
+- Uploading a Dub preview no longer blocks every backend request while ffmpeg extracts its audio (#1667) — thanks @tfreyd!
+- Docker quick starts now require the administrator key needed through container NAT instead of starting a UI whose protected actions return 403 (#1651) — thanks @wd357dui!
+- WSL2 AMD containers now use the `/dev/dxg` ROCDXG bridge with actionable GPU diagnostics instead of silently falling back to CPU (#1655) — thanks @wd357dui!
+- Ad-hoc voice-clone references now stay alive until cancelled or timed-out GPU work actually stops reading them, so prompt caching can finish instead of failing on a deleted temp file (#1668) — thanks @tfreyd!
 - Dictation now stays bound to the app where it started and recovers locally from silent recognizer output (#1175)
 - The backend now answers within a second of launch and narrates its startup step by step (#1550)
 - Reporting a bug from an outdated build now offers the latest release first (#1547)
@@ -18,10 +113,15 @@ the frozen-backend fallback mirror it for their toolchains.
 - Dub subtitles can be retimed, inserted, and merged in either direction from the segment table (#1612) — thanks @invio-a11y!
 
 ### Changed
+- Model Catalogue now uses one breathable workspace canvas with simpler pane and engine-family navigation instead of nested cards and scroll regions (#1685)
+- Linux source launchers now catch missing libxdo and GStreamer audio plugins before they can cause a linker error or an aborted, blank WebKit renderer (#1680, #1682)
 - Dictation now carries one native output session from shortcut-down through final delivery, restores text, HTML, image, or file-list clipboards only when untouched, keeps Wayland copy-safe unless current-focus insertion is explicitly enabled, and retries silent Sherpa speech only through an already-installed local ASR model (#1175)
 - The backend binds its port immediately and reports startup progress live — `/health` answers 503-with-step and a new `/startup/progress` endpoint lists every step while PyTorch, API routes, and database migrations load in the background, so "starting at step X" is never mistakable for "dead"; the desktop splash narrates each step (#1550)
 
 ### Added
+- A bundled Rust loopback sidecar exposes dictation start/stop/toggle, focused-output sessions, discovery, and JSON-RPC; the backend adds versioned streaming events and a dependency-free CLI bridge for Herdr, coding agents, editors, desktop apps, and TUIs (#1646)
+- Headless NVIDIA and ROCm machines can now join as worker-only Docker Compose services with no published UI and durable protocol-v2 enrollment; update both machines together before reconnecting (#1638) — thanks @jkrogers9862!
+- Linux ARM64 (Asahi Apple Silicon) support for the OmniVoice GGUF engine — a `linux-aarch64` binary built with GGML Vulkan where the toolchain allows it, so Apple GPUs accelerate generation through the open-source Honeykrisp driver instead of falling back to CPU-only (#1641)
 - One-command install on every desktop OS: `curl -fsSL https://voicestudio.sh/install | sh` (macOS/Linux/WSL) or `irm https://voicestudio.sh/install | iex` (Windows) — the URL serves the right script per platform, and Windows gains a source installer (`scripts/install.ps1`) with a 3-OS CI smoke (#1626)
 - Per-line subtitle management in the dub table: a line's end time is editable alongside its start (typing a time and dragging its timeline edge now take the same path), lines merge with the previous row as well as the next (`Ctrl/Cmd+Shift+M`), and a new line can be inserted into the gap after any row (#1612) — thanks @invio-a11y!
 - CI now enforces performance regression budgets on the hot paths — operation-count tests pin streaming TTS to one synthesis per sentence and cached dub re-mixes to zero re-synthesis; fast-path guards cover zero re-decoding and ⌈N/W⌉ native batch calls when enabled (#1594)
@@ -35,6 +135,7 @@ the frozen-backend fallback mirror it for their toolchains.
 - Opt-in 24-layer PocketTTS checkpoints via `OMNIVOICE_POCKETTTS_24L` — better prosody for it/de/es/pt at roughly 2x render time (still faster than real-time); the fast 6-layer model stays the default (#1613) — thanks @paoloantinori!
 
 ### Docs
+- Supported-version and install guidance now identifies 0.5.1 as the stable desktop and container release (#1687)
 - The Docker Hub overview now shows the current engine-switching demo, Model Catalogue, and gallery voice workflow (#1593)
 - The Docker Hub overview and install guide now show the v0.5 tags and the built-in API-key/share-PIN security model instead of obsolete v0.4 and no-authentication guidance (#1592)
 - The READMEs now lead with download buttons and a three-step first-clone walkthrough, and a new benchmarks page anchors measured per-engine/per-device numbers on the in-repo harness (#1555)
@@ -42,6 +143,18 @@ the frozen-backend fallback mirror it for their toolchains.
 - The OmniVoice guide now covers combining style attributes with a reference clip (consistent instruct stabilizes cloning; the reference wins conflicts), inline pronunciation control (pinyin / CMU phonemes), and corrects the claim that the default engine can't do voice design — it can, from attributes (#1565)
 
 ### Fixed
+- Workspaces now measure their responsive width when the post-bootstrap shell actually mounts, so native UI scaling reflows Projects and History instead of crushing the Dubbing demo into unreadable columns (#1683)
+- Dubbing keeps the source-language selector visible after a local file is chosen, so ASR can be pinned before transcription starts (#1678) — thanks @Lonki-lomki-cloud!
+- First-run media-engine downloads become available to TTS immediately without a restart, and missing media-process failures now point to repair controls (#1677) — thanks @farhataligpt-dev!
+- Source installs on AMD GPUs honour `OMNIVOICE_TORCH_VARIANT=rocm`: `bun run desktop` now swaps in the ROCm torch wheel after `uv sync` and launches the backend without re-syncing, instead of silently reverting to the CPU-only CUDA build on every start (#1665) — thanks @uberclokr!
+- `bun run desktop` on a fresh clone no longer fails with "resource path `../../frontend/dist` doesn't exist" — the dev launcher creates the placeholder Tauri resource directory before compiling (#1664) — thanks @uberclokr!
+- macOS no longer loses TTS after the first request when Python lacks `os.waitid`; subprocess ownership now uses a safe `waitpid` fallback without risking reused process groups (#1656) — thanks @paoloantinori!
+- Desktop startup, Retry, reset, uninstall, shutdown, and crash recovery now share one backend lifecycle owner; quitting interrupts first-run installers and gracefully drains then force-cleans the full backend process tree, so overlaps cannot duplicate or orphan it (#1635) — thanks @Xohaibxobi!
+- Large Stories and Audiobook projects now persist in IndexedDB instead of overflowing the `omnivoice.app` localStorage envelope, with quota-safe migration and orderly exit/reload flushing (#1636) — thanks @leodzai!
+- OmniVoice and its crash-isolated subprocess now route to AMD ROCm GPUs instead of warning and falling back to CPU (#1629) — thanks @j4r3kb!
+- Dictation now cancels pending startup work, capture resources, sockets, and timers when the capture widget closes, preventing late work against a destroyed webview (#1645)
+- Streaming generation failures now show recognized recovery guidance and appear in Diagnostics instead of only returning a generic error (#1607)
+- The worker-capacity transport test no longer races its own setup: the 1-slot limit now goes through the enrollment handshake instead of mutating client config after connect, where the server's stream-open ConfigUpdate (carrying the registered capacity of 2) could overwrite it and fake an over-accept; failed CI twice on 2026-08-21 (#1630)
 - Moving words across a speaker boundary in a dub — merging two lines and splitting them again — no longer dubs the second half in the first speaker's voice; each half now keeps the speaker, voice, direction, gain, and language of whoever actually says it (#1612) — thanks @invio-a11y!
 - Dictation on a WebView that refuses a 16 kHz audio context (WKWebView) now low-passes before downsampling, so frequencies above 8 kHz stop folding into the speech the recognizer is fed (#1610)
 - A microphone context that cannot be resumed now reports a mic error instead of leaving the dictation pill on "Listening" while capturing nothing (#1610)

LICENSE-NOTICE.md

@@ -10,10 +10,10 @@ Copyright 2024-present Palash Debnath and VoiceStudio contributors.
 
 VoiceStudio is **free and open-source software, licensed under the GNU
 Affero General Public License, Version 3 (AGPL-3.0)**. You are free to use,
-copy, modify, and redistribute it — and that **includes commercial and internal
-business use**: run the app, use its outputs commercially, sell the audio you
-produce with it, provide professional/client services with it, and deploy it
-within your organization.
+copy, modify, and redistribute it. That **includes commercial and internal
+business use** of the application itself. Model weights, tokenizers, and other
+third-party assets retain their own terms; this application license does not
+grant or summarize rights under those separate terms.
 
 Because this is the **Affero** GPL, one additional obligation applies: if you
 modify VoiceStudio and make that modified version available to others over
@@ -41,6 +41,12 @@ is **separately licensed under Apache License 2.0** by its upstream authors and
 is not relicensed here. Apache License 2.0 is compatible with, and may be
 combined under, the GNU AGPL-3.0. See `pyproject.toml`.
 
+Downloaded model weights are not relicensed by VoiceStudio. The default
+`k2-fsa/OmniVoice` model card identifies its code as Apache-2.0 and pretrained
+weights as CC-BY-NC. Its `audio_tokenizer/LICENSE` contains separate Boson
+Higgs Audio 2 and Meta Llama community terms. A commercial license for
+VoiceStudio-owned code does not replace any of those terms.
+
 Third-party dependencies retain their own licenses. See `Cargo.lock`,
 `bun.lock`, and `uv.lock` for the resolved set.
 

README.md

@@ -1,24 +1,30 @@
 <div align="center">
-  <img src="docs/logo.png" alt="VoiceStudio logo" width="120" height="120" />
+  <p><img src="docs/logo.png" alt="VoiceStudio logo" width="120" height="120" /></p>
   <h1>VoiceStudio</h1>
+  <p>
+    <a href="https://trendshift.io/repositories/28176?utm_source=repository-badge&amp;utm_medium=badge&amp;utm_campaign=badge-repository-28176" target="_blank" rel="noopener noreferrer"><img src="https://trendshift.io/api/badge/repositories/28176" alt="VoiceStudio ranking on Trendshift" width="220" height="48" /></a>
+  </p>
   <p><sub>Previously OmniVoice-Studio</sub></p>
-  <h3>Local voice cloning, dubbing, dictation, and long-form audio.</h3>
-  <p>16 TTS engines · 11 ASR engines · 646-language catalogue · macOS, Windows, and Linux</p>
-  <p><strong>Local-first.</strong> No account, API key, subscription, or usage meter for the core workflow.</p>
+  <h3>Clone voices, dub video, dictate, and produce long-form audio on your own hardware.</h3>
+  <p>16 TTS engines · 11 ASR engines · 646-language catalogue · macOS, Windows, Linux, and Docker</p>
+  <p>No account, API key, subscription, or usage meter for the local workflow.</p>
 
   <p>
     <a href="#install">Install</a> ·
     <a href="#features">Features</a> ·
     <a href="#comparison">Compare</a> ·
     <a href="#requirements">Requirements</a> ·
+    <a href="#hardware-recommendations">Hardware</a> ·
     <a href="#engines">Engines</a> ·
     <a href="#architecture">Architecture</a> ·
     <a href="#api">API</a> ·
     <a href="#documentation">Docs</a> ·
+    <a href="#faq">FAQ</a> ·
     <a href="README_CN.md"><strong>简体中文</strong></a>
   </p>
 
   <p>
+    <a href="https://github.com/debpalash/VoiceStudio/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/debpalash/VoiceStudio/ci.yml?branch=main&style=flat-square&label=CI" alt="CI status" /></a>
     <a href="https://github.com/debpalash/VoiceStudio/stargazers"><img src="https://img.shields.io/github/stars/debpalash/VoiceStudio?style=flat-square&color=f59e0b" alt="GitHub stars" /></a>
     <a href="https://github.com/debpalash/VoiceStudio/releases"><img src="https://img.shields.io/github/downloads/debpalash/VoiceStudio/total?style=flat-square&color=8b5cf6&label=downloads" alt="Total downloads" /></a>
     <a href="https://github.com/debpalash/VoiceStudio/releases/latest"><img src="https://img.shields.io/github/v/release/debpalash/VoiceStudio?style=flat-square&color=10b981" alt="Latest release" /></a>
@@ -36,7 +42,7 @@
 </div>
 
 > [!WARNING]
-> **Active beta.** Use the [latest release](https://github.com/debpalash/VoiceStudio/releases/latest) for stable work or `main` for current fixes. Report problems through [GitHub Issues](https://github.com/debpalash/VoiceStudio/issues).
+> **Active beta.** Use the [latest release](https://github.com/debpalash/VoiceStudio/releases/latest) for stable work. `main` contains the newest fixes and may change between releases. Report problems through [GitHub Issues](https://github.com/debpalash/VoiceStudio/issues).
 
 ## At a glance
 
@@ -49,33 +55,55 @@
 | **Compute** | CUDA · Apple Silicon MPS/MLX · ROCm on Linux · CPU · optional remote workers |
 | **Interfaces** | Desktop app · local REST/SSE/WebSocket API · OpenAI-compatible audio API · MCP Server |
 | **Storage** | Voices, projects, settings, and outputs stay on the machine by default |
-| **License** | AGPL-3.0; optional engines keep their own model licenses |
+| **License** | AGPL-3.0 application; downloaded models keep their upstream terms |
 
 <a id="install"></a>
 
 ## Install
 
+Download a package from the [latest release](https://github.com/debpalash/VoiceStudio/releases/latest), then follow the platform guide.
+
 | Platform | Package | Guide |
 |---|---|---|
-| macOS 13.3+ | DMG, Apple Silicon | [Install on macOS](docs/install/macos.md) |
-| Windows 10/11 | MSI, x64 | [Install on Windows](docs/install/windows.md) |
+| macOS 13.3+ | Apple Silicon DMG | [Install on macOS](docs/install/macos.md) |
+| Windows 10/11 | x64 MSI; choose the current-user build when listed to install without admin access | [Install on Windows](docs/install/windows.md#install-pre-built-msi) |
 | Linux | AppImage, x86_64 with glibc 2.39+ | [Install on Linux](docs/install/linux.md) |
-| Docker | CUDA, ROCm, or CPU | [Run with Docker](docs/install/docker.md) |
+| Docker | CUDA, ROCm, CPU, and worker-only GPU profiles | [Run with Docker](docs/install/docker.md) |
 
-Download packages from the [latest release](https://github.com/debpalash/VoiceStudio/releases/latest). First launch creates a managed Python environment and downloads the default model. Later launches reuse both.
+First launch creates a managed Python environment and downloads the default model. Later launches reuse both.
 
 > [!NOTE]
-> On macOS, first launch needs a one-time right-click → **Open** approval. Intel Macs cannot run the local Python backend; use a [remote backend](docs/install/macos.md) instead.
+> On macOS, first launch needs a one-time right-click, then **Open** approval. Intel Macs cannot run the local Python backend; use a [remote backend](docs/install/macos.md) instead.
+
+### Quick Docker run
+
+```bash
+docker run -d -p 127.0.0.1:3900:3900 -v omnivoice-data:/app/omnivoice_data --name voicestudio palashdeb/omnivoice-studio:stable
+```
 
 ### First voice
 
 1. Launch VoiceStudio and open **Voice Cloning**.
-2. Add a clean voice sample. Three seconds works; 5–15 seconds usually gives a better prompt.
+2. Add a clean voice sample. Three seconds works; 5 to 15 seconds usually gives a better prompt.
 3. Enter text, choose a language, then select **Generate**.
 
+> [!TIP]
+> **Try without installing:** Run VoiceStudio in the cloud via the [Google Colab notebook](https://colab.research.google.com/github/debpalash/VoiceStudio/blob/main/notebooks/OmniVoice_Studio_Colab.ipynb). Explore audio quality comparisons in [benchmarks](docs/benchmarks.md) and prompt design tips in [expressive speech](docs/expressive-speech.md).
+
+### Audio samples
+
+Listen to sample outputs produced locally with VoiceStudio:
+
+| Workflow | Prompt / Reference Audio | Generated Audio |
+|---|---|---|
+| **Voice Cloning** | [demo_voice.wav](backend/assets/samples/demo_voice.wav) | [demo_clone_output.wav](backend/assets/samples/demo_clone_output.wav) |
+| **Voice Design** (US News Anchor) | *"Clear, authoritative American broadcast tone"* | [demo_voice_design_us_news_anchor.wav](backend/assets/samples/voice_design/demo_voice_design_us_news_anchor.wav) |
+| **Voice Design** (UK Audiobook) | *"Warm, expressive British storytelling voice"* | [demo_voice_design_audiobook_uk_narrator.wav](backend/assets/samples/voice_design/demo_voice_design_audiobook_uk_narrator.wav) |
+| **Video Dubbing** (Multilingual) | [source.src.wav](backend/assets/samples/demo/dubbing/source.src.wav) | [Spanish](backend/assets/samples/demo/dubbing/dubbed_es.src.wav) · [French](backend/assets/samples/demo/dubbing/dubbed_fr.src.wav) · [Japanese](backend/assets/samples/demo/dubbing/dubbed_ja.src.wav) · [Chinese](backend/assets/samples/demo/dubbing/dubbed_zh.src.wav) |
+
 ### Run from source
 
-Install the [development prerequisites](.github/CONTRIBUTING.md#development-setup), then:
+Install the [development prerequisites](.github/CONTRIBUTING.md#development-setup) (Node 20+/Bun and Python 3.11+), then:
 
 ```bash
 git clone https://github.com/debpalash/VoiceStudio.git
@@ -84,7 +112,7 @@ bun install
 bun run desktop
 ```
 
-Use `bun run dev` for the browser UI. See [Contributing](.github/CONTRIBUTING.md) for services, tests, and platform packages.
+The desktop launcher configures Python dependencies on first run via `uv` automatically. Use `bun run dev` for the browser UI. See [Contributing](.github/CONTRIBUTING.md) for services, tests, and platform packages.
 
 ### If setup fails
 
@@ -99,22 +127,22 @@ Use `bun run dev` for the browser UI. See [Contributing](.github/CONTRIBUTING.md
 
 | Area | Included |
 |---|---|
-| **Voice Cloning** | Zero-shot synthesis from a short reference clip |
-| **Voice Design** | Create a voice from age, accent, pitch, style, and delivery instructions |
-| **Video Dubbing** | Transcribe, translate, preserve speakers, synthesize, and export video |
+| **Voice Cloning** | Zero-shot synthesis from a short reference clip ([guide](docs/engines/README.md)) |
+| **Voice Design** | Create a voice from age, accent, pitch, style, and delivery instructions ([expressive speech](docs/expressive-speech.md)) |
+| **Video Dubbing** | Transcribe, translate, preserve speakers, synthesize, and export video ([export guide](docs/dubbing/export.md)) |
 | **Stories and audiobooks** | Multi-voice scripts · EPUB/PDF import · chapter rendering · `.m4b` export |
 | **[Dictation Widget](docs/features/dictation.md)** | System-wide shortcut, live transcription, optional local-LLM cleanup |
 | **Vocal Isolation** | Demucs speech/background separation |
-| **Speaker Diarization** | Pyannote and WhisperX speaker assignment |
-| **Batch Queue** | Queue large sets of audio and video jobs with per-job progress |
-| **Model Catalogue** | Install, remove, select, and route TTS, ASR, and LLM models |
-| **Remote Model Downloads** | Install models on enrolled remote workers with live progress |
-| **GPU Auto-Detect** | CUDA, MPS, ROCm, and CPU routing with per-engine checks |
+| **Speaker Diarization** | Pyannote and WhisperX speaker assignment ([guide](docs/features/diarization.md)) |
+| **Batch Queue** | Queue large sets of audio and video jobs with per-job progress, or watch a local folder for new videos |
+| **Model Catalogue** | Install, remove, select, and route TTS, ASR, and LLM models ([catalogue](docs/engines/README.md)) |
+| **Remote Model Downloads** | Install models on enrolled remote workers with live progress ([guide](docs/downloading-models.md)) |
+| **GPU Auto-Detect** | CUDA, MPS, ROCm, and CPU routing with per-engine checks ([performance](docs/performance.md)) |
 | **AI Watermark** | AudioSeal embedding and detection |
-| **MCP Server** | Synthesis and transcription tools for MCP clients |
-| **Diagnostics** | Self-checks, error journal, logs, and scrubbed support bundles |
+| **MCP Server** | Synthesis and transcription tools for MCP clients ([guide](docs/mcp.md)) |
+| **Diagnostics** | Self-checks, error journal, logs, and scrubbed support bundles ([troubleshooting](docs/install/troubleshooting.md)) |
 | **Local-first** | Core creation stays local; network-backed features are explicit opt-ins |
-| **Extensible** | Registry-based TTS, ASR, and plugin interfaces |
+| **Extensible** | Registry-based TTS, ASR, and plugin interfaces ([acceptance](docs/engine-acceptance.md)) |
 
 <table>
 <tr>
@@ -157,10 +185,20 @@ Requirements vary by engine. These values cover the default local workflow.
 | **Disk** | 10 GB free | 20 GB+ SSD |
 | **GPU** | Optional; CPU mode is supported | NVIDIA CUDA or Apple Silicon |
 | **VRAM** | 4 GB when using a GPU | 8 GB+; large optional engines need more |
-| **Python from source** | 3.11+ | 3.11–3.12 |
+| **Python from source** | 3.11+ | 3.11 or 3.12 |
 
 ROCm is Linux-only and opt-in. Windows AMD/Ryzen AI uses CPU. Systems with limited VRAM offload work to CPU when required. See [performance](docs/performance.md), [benchmarks](docs/benchmarks.md), and [engine disk usage](docs/engines/disk-usage.md).
 
+<a id="hardware-recommendations"></a>
+
+### Recommended stack by hardware
+
+| Hardware | Recommended TTS | Recommended ASR | Why |
+|---|---|---|---|
+| **Apple Silicon (M1–M4)** | [MLX-Audio](docs/engines/mlx-audio.md) · [OmniVoice](docs/engines/omnivoice.md) (MPS) | [MLX Whisper](docs/engines/mlx-whisper.md) · [Parakeet MLX](docs/engines/parakeet-mlx.md) | Native unified memory, lowest latency on macOS |
+| **NVIDIA GPU (8 GB+ VRAM)** | [OmniVoice](docs/engines/omnivoice.md) · [CosyVoice 3](docs/engines/cosyvoice.md) | [WhisperX](docs/engines/whisperx.md) | High-fidelity zero-shot cloning, word timestamps, diarization |
+| **Low VRAM / CPU-only** | [PocketTTS](docs/engines/pockettts.md) · [Sherpa-ONNX](docs/engines/sherpa-onnx.md) · [KittenTTS](docs/engines/kittentts.md) | [Moonshine](docs/engines/moonshine.md) · [Faster-Whisper](docs/engines/faster-whisper.md) (`int8`) | Low memory footprint, optimized CPU inference |
+
 <a id="engines"></a>
 
 ## Engines
@@ -173,22 +211,22 @@ Engine support is capability-specific. Check cloning, language, platform, memory
 
 | Engine | Languages | Clone | Instruct | Linux | macOS ARM | Windows | License |
 |---|:---:|:---:|:---:|:---:|:---:|:---:|---|
-| **VoiceStudio** (default, powered by k2-fsa/OmniVoice) | 600+ | Yes | Yes | CUDA/CPU | MPS | CUDA/CPU | [AGPL-3.0](LICENSE) app · [Apache-2.0](LICENSE-NOTICE.md) model |
-| **CosyVoice 3** | 9 + 18 dialects | Yes | Yes | CUDA/CPU | CPU | CUDA/CPU | Apache-2.0 |
-| **GPT-SoVITS** | 5 | Yes | — | CUDA/CPU | — | CUDA/CPU | MIT |
-| **VoxCPM2** | 30 | Yes | Yes | CUDA/CPU | MPS | CUDA/CPU | Apache-2.0 |
-| **MOSS-TTS-Nano** | 20 | Yes | — | CUDA/CPU | CPU | CUDA/CPU | Apache-2.0 |
-| **KittenTTS** | English | — | — | CPU | CPU | CPU | MIT |
-| **MLX-Audio** | Model-dependent | Varies | Varies | — | MLX | — | Varies |
-| **Sherpa-ONNX** | 20+ | — | — | CUDA/CPU | CPU | CUDA/CPU | Apache-2.0 |
-| **IndexTTS 2.5** ⚡ | ZH · EN · JA · ES · AR | Yes | — | CUDA/CPU | CPU | CUDA/CPU | Bilibili model license¹ |
-| **OmniVoice GGUF** ⚡ | 600+ | Yes | Yes | CUDA/CPU | MPS/CPU | CUDA/CPU | [AGPL-3.0](LICENSE) app · [Apache-2.0](LICENSE-NOTICE.md) model |
-| **OmniVoice (subprocess)** ⚡ | 600+ | Yes | Yes | CUDA/CPU | MPS | CUDA/CPU | [AGPL-3.0](LICENSE) app · [Apache-2.0](LICENSE-NOTICE.md) model |
-| **PocketTTS** ⚡ | EN · FR · DE · PT · IT · ES | Yes | — | CPU | CPU | CPU | CC-BY-4.0, gated² |
-| **Supertonic 3** ⚡ | 31 | — | — | CPU | CPU | CPU | OpenRAIL-M |
-| **MOSS-TTS-v1.5** ⚡ | 31 | Yes | — | CUDA/CPU | CPU | CUDA/CPU | Apache-2.0 |
-| **dots.tts** ⚡ | 24 | Yes | — | CUDA/CPU | CPU | — | Apache-2.0 |
-| **Confucius4-TTS** ⚡ | 14 | Yes | — | CUDA/CPU | CPU | CUDA/CPU | Apache-2.0 |
+| [**VoiceStudio** (default, powered by k2-fsa/OmniVoice)](docs/engines/omnivoice.md) | 600+ | Yes | Yes | CUDA/CPU | MPS | CUDA/CPU | [AGPL-3.0](LICENSE) app · [Apache-2.0 code, CC-BY-NC weights](https://huggingface.co/k2-fsa/OmniVoice#license)³ |
+| [**CosyVoice 3**](docs/engines/cosyvoice.md) | 9 + 18 dialects | Yes | Yes | CUDA/CPU | CPU | CUDA/CPU | Apache-2.0 |
+| [**GPT-SoVITS**](docs/engines/gpt-sovits.md) | 5 | Yes | No | CUDA/CPU | No | CUDA/CPU | MIT |
+| [**VoxCPM2**](docs/engines/voxcpm2.md) | 30 | Yes | Yes | CUDA/CPU | MPS | CUDA/CPU | Apache-2.0 |
+| [**MOSS-TTS-Nano**](docs/engines/moss-tts-nano.md) | 20 | Yes | No | CUDA/CPU | CPU | CUDA/CPU | Apache-2.0 |
+| [**KittenTTS**](docs/engines/kittentts.md) | English | No | No | CPU | CPU | CPU | MIT |
+| [**MLX-Audio**](docs/engines/mlx-audio.md) | Model-dependent | Varies | Varies | No | MLX | No | Varies |
+| [**Sherpa-ONNX**](docs/engines/sherpa-onnx.md) | 20+ | No | No | CUDA/CPU | CPU | CUDA/CPU | Apache-2.0 |
+| [**IndexTTS 2.5** ⚡](docs/engines/indextts.md) | ZH · EN · JA · ES · AR | Yes | No | CUDA/CPU | CPU | CUDA/CPU | Bilibili model license¹ |
+| [**OmniVoice GGUF** ⚡](docs/engines/omnivoice-gguf.md) | 600+ | Yes | Yes | CUDA/CPU | MPS/CPU | CUDA/CPU | [AGPL-3.0](LICENSE) app · [review the derivative model terms](https://huggingface.co/Serveurperso/OmniVoice-GGUF#license)³ |
+| [**OmniVoice (subprocess)** ⚡](docs/engines/omnivoice-subprocess.md) | 600+ | Yes | Yes | CUDA/CPU | MPS | CUDA/CPU | [AGPL-3.0](LICENSE) app · [Apache-2.0 code, CC-BY-NC weights](https://huggingface.co/k2-fsa/OmniVoice#license)³ |
+| [**PocketTTS** ⚡](docs/engines/pockettts.md) | EN · FR · DE · PT · IT · ES | Yes | No | CPU | CPU | CPU | CC-BY-4.0, gated² |
+| [**Supertonic 3** ⚡](docs/engines/supertonic3.md) | 31 | No | No | CPU | CPU | CPU | OpenRAIL-M |
+| [**MOSS-TTS-v1.5** ⚡](docs/engines/moss-tts-v15.md) | 31 | Yes | No | CUDA/CPU | CPU | CUDA/CPU | Apache-2.0 |
+| [**dots.tts** ⚡](docs/engines/dots-tts.md) | 24 | Yes | No | CUDA/CPU | CPU | No | Apache-2.0 |
+| [**Confucius4-TTS** ⚡](docs/engines/confucius4-tts.md) | 14 | Yes | No | CUDA/CPU | CPU | CUDA/CPU | Apache-2.0 |
 
 ⚡ Installed or registered on demand.
 
@@ -196,6 +234,8 @@ Engine support is capability-specific. Check cloning, language, platform, memory
 
 ² PocketTTS shows its gated-access and CC-BY-4.0 terms before first use.
 
+³ The OmniVoice snapshot also includes an audio tokenizer under separate [Boson Higgs Audio 2 and Meta Llama community terms](https://huggingface.co/k2-fsa/OmniVoice/blob/main/audio_tokenizer/LICENSE). VoiceStudio's application license does not replace model or tokenizer terms.
+
 Clone-less engines cannot preserve a reference speaker in dubbing or pinned-voice batch jobs. VoiceStudio rejects those jobs instead of silently changing engines. Heavy engines have separate memory and platform limits; check their engine guide first.
 
 <a id="asr-engines"></a>
@@ -204,17 +244,17 @@ Clone-less engines cannot preserve a reference speaker in dubbing or pinned-voic
 
 | Engine | ID | Languages | Best fit |
 |---|---|:---:|---|
-| **WhisperX** (default) | `whisperx` | ~100 | Dubbing, subtitles, word-level timing |
-| **Faster-Whisper** | `faster-whisper` | ~100 | General cross-platform transcription |
-| **Faster-Whisper (isolated)** | `faster-whisper-isolated` | ~100 | Crash-isolated batch transcription |
-| **MLX Whisper** | `mlx-whisper` | ~100 | Apple Silicon |
-| **PyTorch Whisper** | `pytorch-whisper` | ~100 | CUDA, MPS, and CPU fallback |
-| **Parakeet TDT** | `nemo-parakeet` | English + 25 EU | Fast CPU/CUDA transcription |
-| **Parakeet TDT v3 (MLX)** | `parakeet-mlx` | 25 EU | Apple Silicon dictation and word timestamps |
-| **Moonshine** | `moonshine` | English | Low-power, low-latency ONNX |
-| **FunASR** | `funasr` | 50+ | VAD and inline diarization |
-| **sherpa-onnx** (live dictation) | `sherpa-onnx-asr` | Model-dependent | Streaming CPU dictation |
-| **OpenAI-compatible** ⚠️ remote | `openai-compat-asr` | Server-dependent | Qwen3-ASR or another compatible endpoint; audio leaves the machine |
+| [**WhisperX** (default)](docs/engines/whisperx.md) | `whisperx` | ~100 | Dubbing, subtitles, word-level timing |
+| [**Faster-Whisper**](docs/engines/faster-whisper.md) | `faster-whisper` | ~100 | General cross-platform transcription |
+| [**Faster-Whisper (isolated)**](docs/engines/faster-whisper-isolated.md) | `faster-whisper-isolated` | ~100 | Crash-isolated batch transcription |
+| [**MLX Whisper**](docs/engines/mlx-whisper.md) | `mlx-whisper` | ~100 | Apple Silicon |
+| [**PyTorch Whisper**](docs/engines/pytorch-whisper.md) | `pytorch-whisper` | ~100 | CUDA, MPS, and CPU fallback |
+| [**Parakeet TDT**](docs/engines/nemo-parakeet.md) | `nemo-parakeet` | English + 25 EU | Fast CPU/CUDA transcription |
+| [**Parakeet TDT v3 (MLX)**](docs/engines/parakeet-mlx.md) | `parakeet-mlx` | 25 EU | Apple Silicon dictation and word timestamps |
+| [**Moonshine**](docs/engines/moonshine.md) | `moonshine` | English | Low-power, low-latency ONNX |
+| [**FunASR**](docs/engines/funasr.md) | `funasr` | 50+ | VAD and inline diarization |
+| [**sherpa-onnx** (live dictation)](docs/engines/sherpa-onnx-asr.md) | `sherpa-onnx-asr` | Model-dependent | Streaming CPU dictation |
+| [**OpenAI-compatible** ⚠️ configured server](docs/engines/openai-compatible-asr.md) | `openai-compat-asr` | Server-dependent | Local gigastt/Qwen3-ASR or a remote endpoint; audio goes only to that server |
 
 WhisperX and Faster-Whisper retry with `int8` when efficient `float16` is unavailable. Pin `ASR_COMPUTE_TYPE=int8` or `float32` only if automatic selection still fails.
 
@@ -249,12 +289,12 @@ FastAPI backend
 
 - The desktop talks to a loopback-only backend on `localhost:3900`.
 - Loopback API calls need no server key. Remote access requires a share PIN or API key.
-- Remote workers and OpenAI-compatible ASR are opt-in. The UI identifies when audio leaves the machine.
-- Analytics is off until consent. If enabled, it sends allowlisted, content-free usage metadata—not text, audio, file names, or projects.
+- Remote workers and OpenAI-compatible ASR are opt-in. Loopback ASR may use HTTP and keeps audio on the machine; non-loopback endpoints require HTTPS, and redirects are not followed.
+- Analytics is off until consent. If enabled, it sends allowlisted, content-free usage metadata. It never sends text, audio, file names, or projects.
 
 <a id="api"></a>
 
-## OpenAI-compatible API
+## Local speech platform and OpenAI-compatible API
 
 Point an OpenAI-compatible audio client at the local backend:
 
@@ -267,6 +307,8 @@ Point an OpenAI-compatible audio client at the local backend:
 |---|---|
 | `POST /v1/audio/speech` | TTS to `mp3`, `opus`, `aac`, `flac`, `wav`, or `pcm`; select a profile with `voice` and an engine with `model` |
 | `POST /v1/audio/transcriptions` | STT to `json`, `text`, `verbose_json`, `srt`, or `vtt` |
+| `WS /v1/audio/transcriptions/stream` | Live PCM/WebM transcription with partial, utterance, and session-final events |
+| `GET /.well-known/voicestudio-speech` | Discover HTTP, WebSocket, MCP, and native dictation-control transports |
 | `GET /v1/audio/voices` | List local voice profiles and engines |
 
 ```python
@@ -283,19 +325,61 @@ with client.audio.speech.with_streaming_response.create(
     response.stream_to_file("speech.wav")
 ```
 
-The full API reference is in **Settings → OpenAPI Reference**. For LAN, Tailscale, or proxy access, read [API authentication](docs/api-auth.md) before exposing the backend.
+```bash
+# Quick test via cURL
+curl http://localhost:3900/v1/audio/speech \
+  -H "Content-Type: application/json" \
+  -d '{"model": "tts-1", "input": "Made on my own hardware.", "voice": "default", "response_format": "wav"}' \
+  --output speech.wav
+```
+
+The bundled Rust control sidecar lets Herdr, coding agents, VS Code, desktop apps,
+and TUIs trigger the system-wide dictation flow or reuse its native text
+insertion. See the [speech platform guide](docs/speech-platform.md). The full API
+reference is in **Settings → OpenAPI Reference**. For LAN, Tailscale, or proxy
+access, read [API authentication](docs/api-auth.md) before exposing the backend.
 
 ### Agent skills
 
 Install the VoiceStudio skills for Claude Code, Codex, Cursor, and other [skills.sh](https://skills.sh)-compatible agents:
 
 ```bash
-npx skills add debpalash/omnivoice-studio
+npx skills add debpalash/VoiceStudio
 ```
 
 - `omnivoice`: synthesize speech and transcribe audio through local VoiceStudio.
 - `oss-maintainer`: the repository's open-source maintenance workflow.
 
+### Model Context Protocol (MCP)
+
+VoiceStudio mounts an MCP server at `http://localhost:3900/mcp` for Claude Desktop, Cursor, and AI agents:
+
+```json
+{
+  "mcpServers": {
+    "voicestudio": {
+      "url": "http://localhost:3900/mcp"
+    }
+  }
+}
+```
+
+For clients requiring stdio transport, use the bundled local shim (`docs/mcp.json`):
+
+```json
+{
+  "mcpServers": {
+    "voicestudio": {
+      "command": "python",
+      "args": ["-m", "backend.mcp_shim"],
+      "cwd": "/path/to/VoiceStudio"
+    }
+  }
+}
+```
+
+See the [MCP guide](docs/mcp.md) for tools (`generate_speech`, `clone_voice`, `transcribe`), file streaming modes, and client bindings.
+
 ### Google Colab
 
 [![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/debpalash/VoiceStudio/blob/main/notebooks/OmniVoice_Studio_Colab.ipynb)
@@ -312,11 +396,13 @@ The [notebook](notebooks/OmniVoice_Studio_Colab.ipynb) runs the app and web UI o
 | Fix setup | [Troubleshooting](docs/install/troubleshooting.md) · [model downloads](docs/downloading-models.md) · [Hugging Face token](docs/setup/huggingface-token.md) |
 | Choose an engine | [Engine guides](docs/engines/README.md) · [benchmarks](docs/benchmarks.md) · [expressive speech](docs/expressive-speech.md) |
 | Tune hardware | [Performance](docs/performance.md) · [remote workers](docs/remote-workers.md) |
-| Build integrations | [API auth](docs/api-auth.md) · [MCP](docs/mcp.md) · [examples](examples/README.md) |
+| Build integrations | [Speech platform](docs/speech-platform.md) · [Private production API](docs/production-private-api.md) · [API auth](docs/api-auth.md) · [MCP](docs/mcp.md) · [examples](examples/README.md) |
 | Build VoiceStudio | [Contributing](.github/CONTRIBUTING.md) · [engine acceptance](docs/engine-acceptance.md) |
 | Track changes | [Changelog](CHANGELOG.md) · [roadmap](docs/ROADMAP.md) · [latest release](https://github.com/debpalash/VoiceStudio/releases/latest) |
 | Remove everything | [Uninstall guide](docs/install/uninstall.md) |
 
+<a id="faq"></a>
+
 ## FAQ
 
 <details>
@@ -328,19 +414,19 @@ Apple Silicon is supported with MPS and MLX options. Intel Macs cannot run the l
 <details>
 <summary><strong>How much VRAM do I need?</strong></summary>
 
-A GPU is optional. Use 4 GB VRAM as the minimum for accelerated work and 8 GB+ for the default multi-stage workflow. Large optional engines can require 12–16 GB or more. Check the [benchmarks](docs/benchmarks.md) and engine guide.
+A GPU is optional. Use 4 GB VRAM as the minimum for accelerated work and 8 GB+ for the default multi-stage workflow. Large optional engines can require 12 to 16 GB or more. Check the [benchmarks](docs/benchmarks.md) and engine guide.
 </details>
 
 <details>
 <summary><strong>Why does a longer reference clip not always improve the clone?</strong></summary>
 
-Cloning is zero-shot: the clip is a prompt, not training data. Use 5–15 seconds of one speaker, close to the microphone, without music, noise, or reverb. Match the tone and pace you want in the output. For training, see [data preparation](docs/data_preparation.md) and [training](docs/training.md).
+Cloning is zero-shot: the clip is a prompt, not training data. Use 5 to 15 seconds of one speaker, close to the microphone, without music, noise, or reverb. Match the tone and pace you want in the output. For training, see [data preparation](docs/data_preparation.md) and [training](docs/training.md).
 </details>
 
 <details>
 <summary><strong>Can I use generated audio commercially?</strong></summary>
 
-Yes under VoiceStudio's AGPL-3.0 terms. Optional engines and model weights may use different licenses; review the selected engine's license before commercial use.
+VoiceStudio's application license does not restrict generated audio, but it does not grant rights under a model's separate terms. The default OmniVoice repository labels its pretrained weights CC-BY-NC and includes a tokenizer under separate community terms. Review the selected model terms before commercial use.
 </details>
 
 <details>
@@ -362,17 +448,30 @@ Use `scripts/uninstall.sh` on macOS/Linux or `scripts\uninstall.ps1` on Windows.
 - [Good first issues](https://github.com/debpalash/VoiceStudio/labels/good%20first%20issue) for a scoped starting point.
 - [Contributing guide](.github/CONTRIBUTING.md) for setup, tests, and pull requests.
 
+<p align="center">
+  <a href="https://star-history.com/#debpalash/VoiceStudio&Date">
+    <img src="https://api.star-history.com/svg?repos=debpalash/VoiceStudio&type=Date" alt="Star History Chart" width="100%" />
+  </a>
+</p>
+
 ## Support development
 
 VoiceStudio is free and has no paid tier. Donations fund development and infrastructure.
 
 [Ko-fi](https://ko-fi.com/debpalash) · [PayPal](https://paypal.me/palashCoder) · [Sponsorship details](SPONSORS.md)
 
+## Responsible use and safety
+
+VoiceStudio enables zero-shot voice cloning and speech generation on personal hardware. Please use it responsibly:
+- **Consent:** Only clone or synthesize voices with explicit permission from the speaker.
+- **Audio provenance:** VoiceStudio integrates [AudioSeal](https://github.com/facebookresearch/audioseal) imperceptible watermarking by default to detect and identify synthetic speech without altering sound quality.
+- **Local privacy:** For the default local workflow, audio recordings, transcripts, voices, and projects remain strictly on your local disk; data leaves your device only when you explicitly configure remote workers or external ASR endpoints.
+
 ## License
 
-VoiceStudio is licensed under [AGPL-3.0](LICENSE). You may run it, modify it, use it internally, and sell generated audio. If you modify VoiceStudio and provide that modified version as a network service, AGPL requires you to offer the corresponding source under the same license. A commercial license is available for proprietary embedding; contact **VoiceStudio@palash.dev**. See [LICENSE-NOTICE.md](LICENSE-NOTICE.md) for the plain-language scope.
+VoiceStudio is licensed under [AGPL-3.0](LICENSE). You may run it, modify it, and use it internally. The application license itself does not restrict selling generated audio, but downloaded model and tokenizer terms may. If you modify VoiceStudio and provide that modified version as a network service, AGPL requires you to offer the corresponding source under the same license. A commercial license for VoiceStudio-owned code is available for proprietary embedding; it does not relicense third-party models. Contact **VoiceStudio@palash.dev**. See [LICENSE-NOTICE.md](LICENSE-NOTICE.md) for the plain-language scope.
 
-Optional engines and downloaded models retain their own licenses. The bundled `omnivoice/` model remains Apache-2.0 upstream.
+Optional engines and downloaded models retain their own licenses. The bundled `omnivoice/` Python code is Apache-2.0 upstream; the default downloaded weights and audio tokenizer use separate terms.
 
 ## Acknowledgments
 

README_CN.md

@@ -20,6 +20,7 @@
   </p>
 
   <p>
+    <a href="https://github.com/debpalash/VoiceStudio/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/debpalash/VoiceStudio/ci.yml?branch=main&style=flat-square&label=CI" alt="CI 状态" /></a>
     <a href="https://github.com/debpalash/VoiceStudio/stargazers"><img src="https://img.shields.io/github/stars/debpalash/VoiceStudio?style=flat-square&color=f59e0b" alt="Star 数" /></a>
     <a href="https://github.com/debpalash/VoiceStudio/releases/latest"><img src="https://img.shields.io/github/v/release/debpalash/VoiceStudio?style=flat-square&color=10b981" alt="版本" /></a>
     <a href="LICENSE"><img src="https://img.shields.io/badge/license-AGPL--3.0-blue?style=flat-square" alt="许可证" /></a>
@@ -65,11 +66,27 @@
 - 🐧 **Linux** — [docs/install/linux.md](docs/install/linux.md)
 - 🐳 **Docker** — [docs/install/docker.md](docs/install/docker.md) · [Docker Hub: `palashdeb/omnivoice-studio`](https://hub.docker.com/r/palashdeb/omnivoice-studio)
 
+```bash
+# Docker 快速运行 (CPU / 本地环回模式)
+docker run -d -p 127.0.0.1:3900:3900 -v omnivoice-data:/app/omnivoice_data --name voicestudio palashdeb/omnivoice-studio:stable
+```
+
 **三步克隆出你的第一个声音：**
 
 1. **安装并启动。** 首次启动会自动搭建 Python 运行环境并下载模型权重——启动画面会逐步显示进度（仅首次，需要几分钟；之后即开即用）。
 2. 从启动台打开**语音克隆**，拖入任意声音的 **3 秒音频**。
-3. **输入一句话，点击生成。** 音频完全属于你——在你的设备上生成和保存，支持 646 种语言。
+3. **输入一句话，点击生成。** 音频在你的设备上生成并保存，支持 646 种语言（商业使用前请审阅所选模型与分词器的许可条款）。
+
+### 🎧 音频示例
+
+在线试听 VoiceStudio 本地生成的实际音频样例：
+
+| 工作流 | 提示词 / 参考音频 | 生成音频 |
+|---|---|---|
+| **声音克隆** | [demo_voice.wav](backend/assets/samples/demo_voice.wav) | [demo_clone_output.wav](backend/assets/samples/demo_clone_output.wav) |
+| **声音设计** (美语新闻主播) | *"清晰、权威的美国广播级音色"* | [demo_voice_design_us_news_anchor.wav](backend/assets/samples/voice_design/demo_voice_design_us_news_anchor.wav) |
+| **声音设计** (英式有声书) | *"温暖生动的英式故事讲述音色"* | [demo_voice_design_audiobook_uk_narrator.wav](backend/assets/samples/voice_design/demo_voice_design_audiobook_uk_narrator.wav) |
+| **视频配音** (多语种) | [source.src.wav](backend/assets/samples/demo/dubbing/source.src.wav) | [西班牙语](backend/assets/samples/demo/dubbing/dubbed_es.src.wav) · [法语](backend/assets/samples/demo/dubbing/dubbed_fr.src.wav) · [日语](backend/assets/samples/demo/dubbing/dubbed_ja.src.wav) · [中文](backend/assets/samples/demo/dubbing/dubbed_zh.src.wav) |
 
 觉得慢？[docs/performance.md](docs/performance.md) 讲清了生成时间到底花在哪里、有哪些调优开关，以及“它变慢了”的三个经典原因。各引擎/设备的实测数据见 [docs/benchmarks.md](docs/benchmarks.md)。
 
@@ -217,6 +234,16 @@ Hugging Face Token 的配置见
 > [!IMPORTANT]
 > **macOS Intel（x86_64）不支持本地后端：** 应用 UI 可以安装，但 Python 后端无法运行，因为 PyTorch 已不再发布 Intel Mac 轮子（[#889](https://github.com/debpalash/VoiceStudio/issues/889)）。Intel Mac 用户仍可让 UI 指向另一台机器上的远程后端——参见 [docs/install/macos.md](docs/install/macos.md)。
 
+<a id="hardware-recommendations"></a>
+
+### 💡 按硬件推荐引擎配置
+
+| 硬件配置 | 推荐 TTS 引擎 | 推荐 ASR 语音识别 | 优势 |
+|---|---|---|---|
+| **Apple Silicon (M1–M4)** | [MLX-Audio](docs/engines/mlx-audio.md) · [OmniVoice](docs/engines/omnivoice.md) (MPS) | [MLX Whisper](docs/engines/mlx-whisper.md) · [Parakeet MLX](docs/engines/parakeet-mlx.md) | 原生统一内存，macOS 上延迟最低、性能最强 |
+| **NVIDIA 显卡 (8 GB+ 显存)** | [OmniVoice](docs/engines/omnivoice.md) · [CosyVoice 3](docs/engines/cosyvoice.md) | [WhisperX](docs/engines/whisperx.md) | 极致零样本克隆品质、字级时间戳对齐与说话人分离 |
+| **低显存 / 仅 CPU 设备** | [PocketTTS](docs/engines/pockettts.md) · [Sherpa-ONNX](docs/engines/sherpa-onnx.md) · [KittenTTS](docs/engines/kittentts.md) | [Moonshine](docs/engines/moonshine.md) · [Faster-Whisper](docs/engines/faster-whisper.md) (`int8`) | 超低内存占用，针对 CPU 指令集深度优化 |
+
 <a id="tts-engines"></a>
 
 ### 🗣️ TTS 引擎
@@ -338,9 +365,9 @@ print(result.text)
 
 ### 📓 在 Google Colab 上运行
 
-[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/debpalash/VoiceStudio/blob/main/notebooks/VoiceStudio_Studio_Colab.ipynb)
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/debpalash/VoiceStudio/blob/main/notebooks/OmniVoice_Studio_Colab.ipynb)
 
-没有本地 GPU？官方笔记本（[notebooks/VoiceStudio_Studio_Colab.ipynb](notebooks/VoiceStudio_Studio_Colab.ipynb)）可在免费的 Colab T4 上启动完整应用（包含 Web 界面）：在笔记本内直接构建前端，用 uv 安装后端（复用 Colab 预装的 CUDA PyTorch），并通过 Colab 内置端口代理打开界面。无需第三方隧道，也无需任何 API 密钥。随后还有一套覆盖全部主要功能的 API 导览，全部可在笔记本内直接播放：多语言 TTS、声音克隆与声音设计、已保存的声音档案、语音转写、AI 水印检测、OpenAI 兼容 API、多角色故事、带章节的 m4b 有声书，以及一个附带人声分离音轨的迷你视频配音。
+没有本地 GPU？官方笔记本（[notebooks/OmniVoice_Studio_Colab.ipynb](notebooks/OmniVoice_Studio_Colab.ipynb)）可在免费的 Colab T4 上启动完整应用（包含 Web 界面）：在笔记本内直接构建前端，用 uv 安装后端（复用 Colab 预装的 CUDA PyTorch），并通过 Colab 内置端口代理打开界面。无需第三方隧道，也无需任何 API 密钥。随后还有一套覆盖全部主要功能的 API 导览，全部可在笔记本内直接播放：多语言 TTS、声音克隆与声音设计、已保存的声音档案、语音转写、AI 水印检测、OpenAI 兼容 API、多角色故事、带章节的 m4b 有声书，以及一个附带人声分离音轨的迷你视频配音。
 
 ### 🤝 智能体技能（Agent Skills）
 
@@ -352,6 +379,36 @@ npx skills add debpalash/omnivoice-studio
 
 内含两个 [skills](https://skills.sh)：**`omnivoice`**——让任何智能体通过你的本地安装进行语音合成与转录（包括你克隆的声音），免费且离线；以及 **`oss-maintainer`**——本项目所遵循的维护者方法论，适合任何用智能体运营自己开源项目的人。
 
+### 🔌 模型上下文协议（MCP 服务器）
+
+VoiceStudio 在 `http://localhost:3900/mcp` 挂载了 MCP 服务，可供 Claude Desktop、Cursor 与自主智能体调用：
+
+```json
+{
+  "mcpServers": {
+    "voicestudio": {
+      "url": "http://localhost:3900/mcp"
+    }
+  }
+}
+```
+
+对于需要 stdio 管道传输的客户端，请使用内置的本地桥接脚本（`docs/mcp.json`）：
+
+```json
+{
+  "mcpServers": {
+    "voicestudio": {
+      "command": "python",
+      "args": ["-m", "backend.mcp_shim"],
+      "cwd": "/path/to/VoiceStudio"
+    }
+  }
+}
+```
+
+支持 `generate_speech`、`clone_voice`、`transcribe` 等工具与流式文件输出模式，详见 [docs/mcp.md](docs/mcp.md)。
+
 ---
 
 ## 🗺️ 路线图
@@ -542,6 +599,13 @@ VoiceStudio **免费**且采用 **AGPL-3.0** 许可——没有付费版，没
 VoiceStudio 完全本地运行——卸载就是删除应用及其写入的文件夹（模型缓存、Python 环境、你的声音/项目、配置）。运行 <code>scripts/uninstall.sh</code>（macOS/Linux）或 <code>scripts\uninstall.ps1</code>（Windows）——它会先以干跑方式列出每个文件夹及其大小，加 <code>--yes</code> 才会真正删除。完整的各平台路径列表和应用移除步骤见 <a href="docs/install/uninstall.md"><b>docs/install/uninstall.md</b></a>。
 </details>
 
+## 🛡️ 负责任使用与安全
+
+VoiceStudio 在个人硬件上提供零样本语音克隆与语音创作能力。我们提倡负责任的技术使用：
+- **明确授权：** 严禁在未经说话人本人知情并明确授权的情况下克隆其声音。
+- **AI 溯源：** VoiceStudio 默认集成 [AudioSeal](https://github.com/facebookresearch/audioseal) 不可见神经音频水印，在完全不影响听感音质的前提下精准标记合成语音。
+- **本地隐私：** 默认本地工作流下，所有音频、声音档案、项目与转录文本始终保存在你的本地设备上；仅当你主动配置远程工作节点或第三方 ASR 端点时，相应数据才会传输到对应服务。
+
 ---
 
 <a id="license"></a>

backend/api/dependencies.py

@@ -54,6 +54,15 @@ def _server_mode() -> bool:
     return os.environ.get("OMNIVOICE_SERVER_MODE", "").strip().lower() in _TRUTHY
 
 
+def validate_server_admin_key() -> None:
+    """Reject an explicitly blank key before a server-mode app starts."""
+    raw_key = os.environ.get("OMNIVOICE_API_KEY")
+    if _server_mode() and raw_key is not None and not raw_key.strip():
+        raise RuntimeError(
+            "OMNIVOICE_API_KEY is blank; configure a non-whitespace administrator key"
+        )
+
+
 def _configured_pin(request) -> str | None:
     """The active share PIN (``app.state.network_share.pin``) or None. Read via
     getattr so a bare Request stub (or a request that hit before lifespan set

backend/api/public_engine_metadata.py

@@ -49,6 +49,13 @@ def public_backends(entries: list[dict]) -> list[dict]:
             item["routing_reason"] = _public_routing_reason(
                 item.get("routing_status"), item["routing_reason"]
             )
+        evidence = item.get("execution_evidence")
+        if isinstance(evidence, dict) and evidence.get("cpu_fallback_reason") is not None:
+            evidence = dict(evidence)
+            evidence["cpu_fallback_reason"] = _public_routing_reason(
+                "cpu_fallback", evidence["cpu_fallback_reason"]
+            )
+            item["execution_evidence"] = evidence
         safe.append(item)
     return safe
 

backend/api/routers/audiobook.py

@@ -361,7 +361,7 @@ LONGFORM_NUM_STEP = 32
 LONGFORM_GUIDANCE_SCALE = 2.0
 
 
-def _seed_segment_rng(base_seed, text: str, nonce: int = 0) -> None:
+def _seed_segment_rng(base_seed, text: str, nonce: int = 0) -> int | None:
     """Apply a profile's pinned seed to this synth call (#1139).
 
     ``_resolve_voice`` has always fetched the profile ``seed`` — but only the
@@ -380,11 +380,13 @@ def _seed_segment_rng(base_seed, text: str, nonce: int = 0) -> None:
     must cover /generate and here together, not one path.
     """
     if base_seed is None:
-        return
+        return None
     import torch
 
     from services.audiobook import segment_seed
-    torch.manual_seed(segment_seed(base_seed, text, nonce))
+    seed = segment_seed(base_seed, text, nonce)
+    torch.manual_seed(seed)
+    return seed
 
 
 def _base_seed(opts: ExpressiveOptions, voice: dict):
@@ -508,16 +510,21 @@ def _build_synth(
                 "get_model": get_model, "language": language, "opts": opts}
 
     backend = cls()
-    extra = _generic_extra_kwargs(opts)
+    native_proxy = bool(getattr(cls, "supports_native_omnivoice_controls", False))
+    extra = (_omnivoice_sampling_kwargs(opts) if native_proxy
+             else _generic_extra_kwargs(opts))
     next_nonce = _make_occ_counter(opts)
 
     def synth(text, voice_id, speed=None):
         v = resolve(voice_id)
-        _seed_segment_rng(_base_seed(opts, v), text, next_nonce())
+        seed = _seed_segment_rng(_base_seed(opts, v), text, next_nonce())
+        call_extra = dict(extra)
+        if native_proxy and seed is not None:
+            call_extra["seed"] = seed
         return backend.generate(
             text, language=language, ref_audio=v["ref_audio"],
             ref_text=v["ref_text"], instruct=v["instruct"], duration=None,
-            speed=float(speed) if speed else 1.0, **extra,
+            speed=float(speed) if speed else 1.0, **call_extra,
         )
     return {"mode": "generic", "resolve": resolve, "engine_id": engine_id,
             "synth": synth, "sample_rate": backend.sample_rate}

backend/api/routers/batch.py

@@ -111,6 +111,24 @@ BATCH_WIDTH_ENV = "OMNIVOICE_DUB_BATCH_WIDTH"
 #: that costs more than the saving.
 _MAX_BATCH_WIDTH = 16
 
+# Bound each allocation while persisting multipart uploads. Video inputs can
+# be many gigabytes; `await UploadFile.read()` with no size used to mirror the
+# entire file in process memory before writing it back out.
+_UPLOAD_CHUNK_BYTES = 1024 * 1024
+
+
+async def _save_upload(upload: UploadFile, destination: str) -> None:
+    try:
+        with open(destination, "wb") as output:
+            while chunk := await upload.read(_UPLOAD_CHUNK_BYTES):
+                output.write(chunk)
+    except BaseException:
+        try:
+            unlink_if_present(destination)
+        except FileCleanupError:
+            logger.warning("Could not remove incomplete batch upload", exc_info=True)
+        raise
+
 
 def _native_batch_width(backend) -> int:
     """How many segments to render in one native batch on THIS host.
@@ -690,9 +708,7 @@ async def enqueue_batch_job(
     ext = os.path.splitext(video.filename or "video.mp4")[1] or ".mp4"
     video_path = os.path.join(batch_dir, f"{job_id}{ext}")
 
-    with open(video_path, "wb") as f:
-        content = await video.read()
-        f.write(content)
+    await _save_upload(video, video_path)
 
     job = {
         "id": job_id,

backend/api/routers/capture_ws.py

@@ -38,11 +38,14 @@ Protocol:
 from __future__ import annotations
 
 import asyncio
+import json
 import logging
 import math
 import os
 import tempfile
 import time
+import uuid
+from typing import Any
 
 from fastapi import APIRouter, WebSocket, WebSocketDisconnect
 
@@ -52,6 +55,9 @@ from services.text_polish import polish_text
 router = APIRouter()
 logger = logging.getLogger("omnivoice.capture_ws")
 
+SPEECH_PROTOCOL = "voicestudio.speech.v1"
+PLATFORM_STREAM_PATH = "/v1/audio/transcriptions/stream"
+
 # How often (seconds) to run transcription on the accumulated buffer.
 # Shorter = more responsive but more GPU load.
 PARTIAL_INTERVAL_S = float(os.environ.get("OMNIVOICE_STREAM_INTERVAL", "2.0"))
@@ -83,6 +89,39 @@ _AEC_FAR = 0x01   # playback reference frame (feed the echo model only)
 SR_MIN, SR_MAX = 8000, 96000
 
 
+def _is_end_control(text: str | None) -> bool:
+    """Accept the versioned JSON control frame and the legacy ``EOF`` frame."""
+    if text == "EOF":
+        return True
+    if not text:
+        return False
+    try:
+        message = json.loads(text)
+    except (TypeError, json.JSONDecodeError):
+        return False
+    return isinstance(message, dict) and message.get("type") == "input_audio.end"
+
+
+class _PlatformWebSocket:
+    """Add v1 session metadata without changing the legacy WebSocket contract."""
+
+    def __init__(self, websocket: WebSocket):
+        self._websocket = websocket
+        self.session_id = uuid.uuid4().hex
+
+    def __getattr__(self, name: str) -> Any:
+        return getattr(self._websocket, name)
+
+    async def send_json(self, data: Any, mode: str = "text") -> None:
+        if isinstance(data, dict):
+            data = dict(data)
+            data.setdefault("protocol", SPEECH_PROTOCOL)
+            data.setdefault("session_id", self.session_id)
+            if data.get("type") == "final":
+                data.setdefault("final_kind", "summary")
+        await self._websocket.send_json(data, mode=mode)
+
+
 def _bounded_sample_rate(query_params) -> int:
     try:
         sample_rate = int(query_params.get("sr", "16000"))
@@ -194,9 +233,24 @@ def _select_sherpa_spec(websocket: WebSocket):
     return _usable_spec(mid) if mid else None
 
 
+@router.websocket(PLATFORM_STREAM_PATH)
 @router.websocket("/ws/transcribe")
 async def ws_transcribe(websocket: WebSocket):
     """Stream audio in, get partial + final transcription out."""
+    is_platform_stream = websocket.url.path == PLATFORM_STREAM_PATH
+    if is_platform_stream:
+        websocket = _PlatformWebSocket(websocket)
+    # A browser can reach localhost regardless of the page's own origin.
+    # Reject ambient cross-site WebSocket handshakes before the loopback-host
+    # shortcut or accept(), while keeping native clients (no Origin header)
+    # and configured/same-origin browser UIs working (#1646 review).
+    origin = websocket.headers.get("origin")
+    if origin:
+        from core.csrf import origin_allowed
+
+        if not origin_allowed(websocket):
+            await websocket.close(code=1008, reason="browser origin not allowed")
+            return
     # Loopback origin guard — refuse anything not from 127.0.0.1, ::1, or
     # localhost. Privileged HTTP routers use Depends(require_admin) at router
     # level; WebSocket dependency injection differs across FastAPI versions, so we
@@ -211,6 +265,16 @@ async def ws_transcribe(websocket: WebSocket):
         return
 
     await websocket.accept()
+    if is_platform_stream:
+        await websocket.send_json({
+            "type": "session.started",
+            "input_format": (
+                "audio/pcm;encoding=s16le;channels=1"
+                if _requested_pcm_sample_rate(websocket.query_params) is not None
+                else "audio/webm;codecs=opus"
+            ),
+            "sample_rate": _bounded_sample_rate(websocket.query_params),
+        })
 
     # Live-dictation engine selection. When a sherpa-onnx model is selected
     # (via ?model= or the dictation.model_id pref) AND sherpa is installed,
@@ -333,7 +397,7 @@ async def ws_transcribe(websocket: WebSocket):
                     total_bytes += len(data)
                     last_audio_time = time.monotonic()
                     continue
-                if msg.get("text") == "EOF":
+                if _is_end_control(msg.get("text")):
                     # Client signals end-of-audio but stays connected for `final`.
                     running = False
                     break
@@ -656,7 +720,7 @@ async def _recv_pcm_frame(websocket: WebSocket, aec):
                 return "skip", b""
             return "near", aec.process_near_end(payload)
         return "near", data
-    if msg.get("text") == "EOF":
+    if _is_end_control(msg.get("text")):
         return "eof", b""
     return "skip", b""
 

backend/api/routers/dub_core.py

@@ -134,6 +134,82 @@ _save_job          = dub_pipeline.save_job
 # paste (or a mis-aimed binary) burn CPU in the parser.
 _MAX_SUBTITLE_PASTE_CHARS = 2_000_000
 
+_SRT_REPLACED_FIELDS = {
+    "id",
+    "start",
+    "end",
+    "text",
+    "text_original",
+    "translations",
+    "translate_error",
+    "translate_degraded",
+}
+
+
+def _best_overlapping_segment(cue: dict, existing: list[dict]) -> dict | None:
+    """Return the prior segment with the strongest temporal overlap."""
+    cue_start = float(cue.get("start") or 0.0)
+    cue_end = float(cue.get("end") or cue_start)
+    cue_mid = (cue_start + cue_end) / 2.0
+    best = None
+    best_key = None
+    for index, segment in enumerate(existing):
+        start = float(segment.get("start") or 0.0)
+        end = float(segment.get("end") or start)
+        overlap = min(cue_end, end) - max(cue_start, start)
+        if overlap <= 0:
+            continue
+        midpoint_distance = abs(cue_mid - ((start + end) / 2.0))
+        key = (overlap, -midpoint_distance, -index)
+        if best_key is None or key > best_key:
+            best = segment
+            best_key = key
+    return best
+
+
+def _carry_srt_voice_metadata(
+    cues: list[dict],
+    existing: list[dict],
+    segment_clones: dict | None,
+    speaker_clones: dict | None = None,
+) -> tuple[list[dict], dict]:
+    """Replace subtitle content while retaining the source cast assignment."""
+    source_clones = dict(segment_clones or {})
+    source_speaker_clones = dict(speaker_clones or {})
+    # Replacement cues get new positional ids. Starting from the old map would
+    # let an unmatched cue whose new id happens to equal an old id inherit an
+    # unrelated reference. Only explicitly overlap-matched references survive.
+    clones = {}
+    merged_segments = []
+    for new_id, cue in enumerate(cues):
+        prior = _best_overlapping_segment(cue, existing)
+        metadata = {
+            key: value
+            for key, value in (prior or {}).items()
+            if key not in _SRT_REPLACED_FIELDS
+        }
+        merged = {
+            **metadata,
+            "id": new_id,
+            "start": cue.get("start", 0.0),
+            "end": cue.get("end", 0.0),
+            "text": cue.get("text", ""),
+            "text_original": cue.get("text", ""),
+        }
+        if not merged.get("speaker_id"):
+            merged["speaker_id"] = cue.get("speaker_id") or "Speaker 1"
+        if prior is not None:
+            prior_id = str(prior.get("id", ""))
+            clone = source_clones.get(prior_id)
+            if clone is None:
+                clone = source_speaker_clones.get(prior.get("speaker_id"))
+            if clone is not None:
+                clones[str(new_id)] = clone
+                if merged.get("profile_id") == f"auto-seg:{prior_id}":
+                    merged["profile_id"] = f"auto-seg:{new_id}"
+        merged_segments.append(merged)
+    return merged_segments, clones
+
 
 @router.post("/dub/parse-subtitle-text")
 def dub_parse_subtitle_text(req: ParseSubtitleTextRequest):
@@ -234,7 +310,32 @@ async def dub_import_srt(job_id: str, file: UploadFile = File(...)):
     else:
         segments = result.segments
 
+    prior_segments = [
+        segment for segment in (job.get("segments") or []) if isinstance(segment, dict)
+    ]
+    segments, segment_clones = _carry_srt_voice_metadata(
+        segments,
+        prior_segments,
+        job.get("segment_clones"),
+        job.get("speaker_clones"),
+    )
     job["segments"] = segments
+    job["segment_clones"] = segment_clones
+    # A pooled speaker clone is keyed only by a display label. Replacement
+    # cues can reuse that label without overlapping the original speaker, so
+    # retain matched pooled references as segment-specific clones above and
+    # drop the global map before rebuilding the cast.
+    job["speaker_clones"] = {}
+    if segment_clones:
+        from services.speaker_clone import build_cast_sources
+
+        job["cast_sources"] = build_cast_sources(
+            segments,
+            None,
+            segment_clones,
+        )
+    else:
+        job.pop("cast_sources", None)
     # `source_lang` stays whatever the user (or the upload step) set; we
     # don't try to language-detect off the cue text — that's noisy and the
     # user usually knows what their .srt is.
@@ -351,12 +452,13 @@ async def preview_upload(video: UploadFile = File(...)):
     safe_name = f"{uuid.uuid4().hex[:12]}"
     vid_path = os.path.join(PREVIEW_DIR, f"{safe_name}{ext}")
     wav_path = os.path.join(PREVIEW_DIR, f"{safe_name}.wav")
-    
-    with open(vid_path, "wb") as f:
-        f.write(await video.read())
-        
-    has_audio = False
-    if ext not in [".wav", ".mp3", ".m4a", ".aac"]:
+    payload = await video.read()
+
+    def _write_and_extract() -> bool:
+        with open(vid_path, "wb") as f:
+            f.write(payload)
+        if ext in {".wav", ".mp3", ".m4a", ".aac"}:
+            return False
         try:
             ffmpeg_cmd = [
                 find_ffmpeg(), "-y", "-i", vid_path,
@@ -368,10 +470,16 @@ async def preview_upload(video: UploadFile = File(...)):
                 stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL,
                 timeout=300,
             )
-            has_audio = True
+            return True
         except Exception as e:
             logger.warning("FFmpeg extraction failed: %s", log_safe(e))
-            pass
+            return False
+
+    # File writes and ffmpeg are blocking operations. Keep them on the bounded
+    # CPU pool so a large preview cannot stall unrelated API requests (#1667).
+    has_audio = await asyncio.get_running_loop().run_in_executor(
+        _cpu_pool, _write_and_extract
+    )
 
     return {
         "url": f"/preview/{safe_name}{ext}",
@@ -410,12 +518,52 @@ _ingest_gen       = dub_pipeline.ingest_pipeline
 #: container so a mislabelled video can't slip past the video-skipping branch.
 _AUDIO_EXTS = {".wav", ".mp3", ".m4a", ".aac", ".flac", ".ogg", ".opus", ".wma"}
 
+# Source-language choices exposed by the first-party dub UI, plus every
+# language code Whisper can write back after auto-detection. A restored job
+# may reuse that detected value as the next upload's override, so rejecting our
+# own persisted codes strands otherwise valid dubbing sessions (#1737).
+# Keeping this an allow-list still rejects language names and private-use
+# BCP-47 tags. Values are normalized to lowercase below.
+_DUB_SOURCE_LANG_CODES = frozenset({
+    "af", "sq", "am", "ar", "hy", "az", "eu", "be", "bn", "bs", "bg",
+    "my", "ca", "cmn-hans", "cmn-hant", "hr", "cs", "da", "nl", "en",
+    "et", "fi", "fr", "gl", "ka", "de", "el", "gu", "ht", "ha", "haw",
+    "he", "hi", "hu", "is", "id", "it", "ja", "jw", "kn", "kk", "km",
+    "ko", "ku", "ky", "lo", "la", "lv", "lt", "mk", "ms", "ml", "mt",
+    "mi", "mr", "mn", "ne", "no", "ps", "fa", "pl", "pt", "pa", "ro",
+    "ru", "sm", "gd", "sr", "sn", "sd", "si", "sk", "sl", "so", "es",
+    "su", "sw", "sv", "tg", "ta", "te", "th", "tr", "uk", "ur", "uz",
+    "vi", "cy", "xh", "yi", "yo", "zu",
+    "as", "ba", "bo", "br", "fo", "lb", "ln", "mg", "nn", "oc", "sa",
+    "tk", "tl", "tt", "yue", "zh",
+})
+
+
+def _source_lang_override(value: str | None) -> str | None:
+    """Normalize a user-selected source language; auto/und means detect."""
+    code = (value or "").strip().lower()
+    if code in {"", "auto", "und"}:
+        return None
+    if code not in _DUB_SOURCE_LANG_CODES:
+        raise HTTPException(status_code=400, detail="Invalid source language code")
+    return code
+
+
+def _detected_source_lang(value: str | None) -> str:
+    """Normalize an ASR language without truncating valid three-letter codes."""
+    code = (value or "en").split("_", 1)[0].strip().lower()
+    if code in _DUB_SOURCE_LANG_CODES:
+        return code
+    short = code[:2]
+    return short if short in _DUB_SOURCE_LANG_CODES else "en"
+
 
 @router.post("/dub/upload")
 async def dub_upload(
     video: UploadFile = File(...),
     job_id: Optional[str] = Form(None),
     input_type: str = Form("video"),
+    source_lang: Optional[str] = Form(None),
 ):
     """Accept a media upload, write to disk, queue background prep task.
 
@@ -445,6 +593,7 @@ async def dub_upload(
             detail=f"Audio-only dubbing needs an audio file ({', '.join(sorted(_AUDIO_EXTS))}); got '{ext or 'no extension'}'.",
         )
 
+    source_lang_override = _source_lang_override(source_lang)
     os.makedirs(job_dir, exist_ok=True)
 
     video_path = os.path.join(job_dir, f"original{ext}")
@@ -456,7 +605,13 @@ async def dub_upload(
     await task_manager.add_task(
         task_id, "prep",
         _ingest_gen, job_id, job_dir,
-        {"kind": "file", "path": video_path, "input_type": input_type}, filename,
+        {
+            "kind": "file",
+            "path": video_path,
+            "input_type": input_type,
+            "source_lang": source_lang_override,
+        },
+        filename,
     )
     return JSONResponse(
         status_code=202,
@@ -478,6 +633,7 @@ async def dub_ingest_url(req: DubIngestUrlRequest, request: Request):
             status_code=400,
             detail="URL must start with http:// or https://. Paste a full video link (e.g. https://youtube.com/watch?v=…) or drop a local file instead.",
         )
+    source_lang_override = _source_lang_override(req.source_lang)
 
     try:
         import yt_dlp  # noqa: F401
@@ -513,6 +669,7 @@ async def dub_ingest_url(req: DubIngestUrlRequest, request: Request):
         "fetch_subs": bool(req.fetch_subs),
         "sub_langs": req.sub_langs or None,
         "cookie_file": cookie_path,
+        "source_lang": source_lang_override,
     }
     try:
         await task_manager.add_task(
@@ -1099,21 +1256,13 @@ async def dub_transcribe_stream(
                         "error_code": failure["code"],
                     }
 
-            # Retry a failed/timed-out chunk once on a fresh pool before giving
-            # up. Otherwise a transient wedge on the FIRST chunk (whisperx often
-            # cold-loads its model there, the #730 hang) drops that whole window
-            # and the transcript is "missing the beginning, only middle+end".
-            # The retry reuses the same audio window, so a recovered chunk fills
-            # the hole instead of leaving silent gaps.
+            # Retry an ordinary completed failure once. A timed-out native call
+            # is different: its thread is still executing and must not overlap
+            # a retry against the same backend (#1669).
             part = None
+            timed_out = False
             for _attempt in range(1, _CHUNK_TRANSCRIBE_ATTEMPTS + 1):
-                # A wedged chunk gets the SAME guarded-timeout + pool-reset
-                # semantics as the whole-file paths (#730/#851):
-                # run_transcribe_guarded bounds the call, abandons the poisoned
-                # pool so the retry (and any concurrent TTS work) gets a fresh
-                # worker, and raises the actionable ASRTimeoutError. Run it as
-                # a task and poll so we can keep yielding pings — the
-                # EventSource connection drops without them.
+                # Run as a task and poll so pings keep the EventSource alive.
                 task = asyncio.ensure_future(run_transcribe_guarded(
                     _gpu_pool, _transcribe_chunk,
                     what=f"Dub chunk {i + 1}/{chunks_n}",
@@ -1128,9 +1277,12 @@ async def dub_transcribe_stream(
                 try:
                     part = task.result()
                 except ASRTimeoutError:
-                    # The guard already reset the pool; keep the actionable
-                    # message (it names the durable fixes, and — after repeated
-                    # timeouts — the crash-isolated engine escape hatch).
+                    # Python cannot kill an in-process native transcribe. Do
+                    # not swap pools and retry over the still-running call:
+                    # concurrent whisperx/CTranslate2 access caused the native
+                    # Windows access violation in #1669. Stop this transcript;
+                    # the worker remains honestly occupied until it exits.
+                    timed_out = True
                     logger.error(
                         "Transcribe chunk %d/%d timed out after %.0fs (attempt %d/%d, job=%s)",
                         i + 1, chunks_n, transcribe_timeout_s, _attempt,
@@ -1149,20 +1301,22 @@ async def dub_transcribe_stream(
                 # error-part; the timeout path already reset the pool).
                 if part is not None and not part.get("error"):
                     break
-                if _attempt < _CHUNK_TRANSCRIBE_ATTEMPTS:
+                if timed_out:
+                    break
+                if not timed_out and _attempt < _CHUNK_TRANSCRIBE_ATTEMPTS:
                     logger.warning(
                         "Retrying transcribe chunk %d/%d after failure/timeout (next attempt %d/%d, job=%s)",
                         i + 1, chunks_n, _attempt + 1, _CHUNK_TRANSCRIBE_ATTEMPTS, log_safe(job_id),
                     )
-                    # A completed exception did not wedge the worker. Resetting
-                    # the pool here leaked a healthy executor on every ordinary
-                    # decode failure; run_transcribe_guarded already resets the
-                    # pool on the only case that needs it: a real timeout.
+                    # A completed exception did not leave native work behind,
+                    # so retrying this same audio window is safe.
             if part.get("error"):
                 chunk_errors.append(part["error"])
                 if part.get("error_code"):
                     chunk_error_codes.append(part["error_code"])
                 logger.warning("Chunk %d/%d error: %s", i + 1, chunks_n, log_safe(part["error"]))
+            if timed_out:
+                break
             if detected_lang is None and part.get("language"):
                 detected_lang = part["language"]
             asr_speaker_turns.extend(part.get("speaker_turns") or [])
@@ -1669,7 +1823,9 @@ async def dub_transcribe_stream(
         except Exception as e:
             logger.warning("speaker_clone extraction skipped: %s", e)
 
-        job["source_lang"] = ((detected_lang or "en").split("_")[0][:2] or "en").lower()
+        job["source_lang"] = job.get("source_lang_override") or _detected_source_lang(
+            detected_lang
+        )
         job["full_transcript"] = " ".join(s.get("text", "") for s in final_segs)
         _save_job(job_id, job)
 
@@ -1866,7 +2022,9 @@ async def dub_transcribe(job_id: str, num_speakers: Optional[int] = None):
             except Exception as e:
                 logger.warning("Failed to unload ASR backend: %s", e)
 
-        job["source_lang"] = (detected_lang or "en").split("_")[0][:2].lower()
+        job["source_lang"] = job.get("source_lang_override") or _detected_source_lang(
+            detected_lang
+        )
 
         scene_cuts = job.get("scene_cuts") or []
         segments = segment_transcript(result, duration=job.get("duration", 0.0), scene_cuts=scene_cuts)
@@ -1922,7 +2080,8 @@ async def dub_transcribe(job_id: str, num_speakers: Optional[int] = None):
             # Bound the whole-file transcribe (#730): a wedged whisperx/CTranslate2
             # call would otherwise hold its GPU-pool worker forever and starve
             # every other request into a "can't reach backend". run_transcribe_guarded
-            # also resets the pool on timeout so capacity is restored.
+            # leaves an unkillable native worker accounted for on timeout so a
+            # retry cannot overlap it (#1669).
             segments_result = await run_transcribe_guarded(_gpu_pool, _transcribe, what="Dub")
         except asyncio.CancelledError:
             job["aborted"] = True

backend/api/routers/dub_export.py

@@ -23,6 +23,7 @@ from services.ffmpeg_utils import (
     find_ffmpeg,
     run_ffmpeg,
 )
+from services.karaoke_ass import build_ass, scale_words
 from services.video_retime import (
     DRIFT_TOLERANCE_S,
     RetimeError,
@@ -403,6 +404,27 @@ def _write_burn_srt(job: dict, exports_dir: str, stamp: str, dual: bool,
     return sub_path
 
 
+def _write_burn_ass(job: dict, exports_dir: str, stamp: str,
+                    fitted_segments: "list[dict] | None" = None,
+                    lang: "str | None" = None) -> str | None:
+    """Karaoke variant of ``_write_burn_srt``: word-timed ASS via ``build_ass``.
+
+    Same text/timing resolution (``_segments_for_lang`` + fitted-cue overlay,
+    which also scales per-word times onto the fitted timeline); the basename
+    is plain ASCII under exports_dir so it is ffmpeg-filter-safe. Returns
+    None if there are no segments to render.
+    """
+    segments = _segments_for_lang(job, lang)
+    if not segments:
+        return None
+    if fitted_segments:
+        segments = _apply_fitted_times(segments, fitted_segments)
+    sub_path = os.path.join(exports_dir, f"burn_subs_{stamp}.ass")
+    with open(sub_path, "w", encoding="utf-8") as f:
+        f.write(build_ass(segments))
+    return sub_path
+
+
 def _ffmpeg_filter_escape(path: str) -> str:
     """Escape a path for use inside an ffmpeg filter value (subtitles=...).
 
@@ -515,6 +537,20 @@ def _apply_fitted_times(segments: list[dict], fitted: list[dict]) -> list[dict]:
         patched = dict(seg)
         patched["start"] = float(cue["start"])
         patched["end"] = float(cue["end"])
+        # Karaoke burn-in: persisted word times live on the original timeline;
+        # scale them linearly onto the fitted cue span so the highlight sweep
+        # follows the retimed audio. Degenerate spans drop the words — export
+        # then falls back to an even split over the fitted span. Inert for
+        # SRT/VTT, which never read ``words``.
+        if isinstance(seg.get("words"), list) and seg.get("words"):
+            scaled = scale_words(
+                seg["words"], seg.get("start", 0.0), seg.get("end", 0.0),
+                patched["start"], patched["end"],
+            )
+            if scaled is not None:
+                patched["words"] = scaled
+            else:
+                patched.pop("words", None)
         out.append(patched)
     return out
 
@@ -577,6 +613,7 @@ async def dub_download(
     save_authorization: str = Header("", alias="X-VoiceStudio-Path-Authorization"),
     burn_subs: bool = Query(False, description="Burn subtitles into the video stream (forces re-encode). Uses dual-subtitle layout when dual=1."),
     dual: bool = Query(False, description="When burn_subs=1, render translated on top of italicised original."),
+    karaoke: bool = Query(False, description="When burn_subs=1, burn a word-timed karaoke highlight (ASS) instead of line subtitles. Ignored when dual=1 (dual karaoke is unsupported — the line burn renders instead)."),
     out_format: str = Query("m4a", description="Audio-only jobs (#119): output container — wav, m4a, mp3, or flac. Ignored for video jobs."),
 ):
     # Strict allowlist on the path param BEFORE it reaches any filesystem
@@ -729,7 +766,18 @@ async def dub_download(
     fitted_segments = _fitted_segments_for(job, default_track) if default_track and default_track != "original" else None
     # Burn the DEFAULT track's text (P1.2) — it's the audio the viewer hears.
     _burn_lang = default_track if default_track and default_track != "original" else None
-    sub_path = _write_burn_srt(job, exports_dir, stamp, dual, fitted_segments=fitted_segments, lang=_burn_lang) if burn_subs else None
+    # Karaoke (word-highlight) burn writes an ASS instead of the line SRT.
+    # Dual layout keeps the line burn — dual karaoke is out of scope, matching
+    # the disabled control in the Export drawer. The default (karaoke off)
+    # takes exactly the legacy SRT path.
+    sub_path = None
+    sub_is_ass = False
+    if burn_subs:
+        if karaoke and not dual:
+            sub_path = _write_burn_ass(job, exports_dir, stamp, fitted_segments=fitted_segments, lang=_burn_lang)
+            sub_is_ass = sub_path is not None
+        if sub_path is None:
+            sub_path = _write_burn_srt(job, exports_dir, stamp, dual, fitted_segments=fitted_segments, lang=_burn_lang)
 
     # ── Smart Fit video retime (two-tier) ─────────────────────────────────
     # Tier 1 (≤48 chunks): single filter_complex graph inlined into the mux
@@ -829,14 +877,16 @@ async def dub_download(
         esc = _ffmpeg_filter_escape(sub_path)
         # Burn AFTER any retime so cues (already on the fitted timeline for
         # Smart Fit) land on the retimed video. Without retime this reduces
-        # to the legacy `[0:v]subtitles=…[vsub]` graph.
+        # to the legacy `[0:v]subtitles=…[vsub]` graph. Karaoke burns the
+        # word-timed ASS through the ass filter at the same graph position.
         if video_map.startswith("["):
             sub_src = video_map
         elif retimed_idx is not None:
             sub_src = f"[{retimed_idx}:v]"
         else:
             sub_src = "[0:v]"
-        filter_parts.append(f"{sub_src}subtitles='{esc}'[vsub]")
+        _sub_filter = "ass" if sub_is_ass else "subtitles"
+        filter_parts.append(f"{sub_src}{_sub_filter}='{esc}'[vsub]")
         video_map = "[vsub]"
     if stretch_entry:
         orig_dur = float(stretch_entry.get("orig_duration") or job.get("duration") or 0.0)
@@ -1599,7 +1649,10 @@ async def dub_download_audio(
         return _native_save(wav_path, save_path, dl_name, media_type="audio/wav")
     return FileResponse(
         wav_path, media_type="audio/wav",
-        headers={"Content-Disposition": content_disposition(dl_name)},
+        headers={
+            "Cache-Control": "no-store",
+            "Content-Disposition": content_disposition(dl_name),
+        },
     )
 
 
@@ -1741,6 +1794,50 @@ async def dub_export_vtt(
     )
 
 
+@router.get("/dub/ass/{job_id}")
+@router.get("/dub/ass/{job_id}/{filename}")
+async def dub_export_ass(
+    job_id: str,
+    lang: str = Query(None, description="Track language code. Same text/timing resolution as /dub/srt, rendered as a karaoke (word-highlight) ASS sidecar."),
+):
+    """Karaoke ASS sidecar — the same script the karaoke burn-in renders.
+
+    Raw text body like /dub/srt and /dub/vtt (the Tauri side writes the file
+    itself; no ?save_path= variant — see the comment above /dub/srt).
+    """
+    _job_dir_or_400(job_id)
+    lang = _safe_lang_or_400(lang)
+    job = _get_job(job_id)
+    if not job:
+        raise HTTPException(status_code=404, detail="Job not found")
+
+    segments = _segments_for_lang(job, lang)
+    if not segments:
+        raise HTTPException(status_code=400, detail="No transcript segments available")
+
+    # Same strategy-aware cue timing as /dub/srt. The fitted overlay also
+    # scales word times; the stretch_video cue path has no per-word record,
+    # so words are dropped and build_ass even-splits over the new spans.
+    fitted = _fitted_segments_for(job, lang)
+    if fitted:
+        segments = _apply_fitted_times(segments, fitted)
+    else:
+        cues = _fitted_cue_times(job, lang)
+        if cues:
+            segments = [
+                {**{k: v for k, v in seg.items() if k != "words"}, "start": s, "end": e}
+                for seg, (s, e) in zip(segments, cues)
+            ]
+
+    base_name = os.path.splitext(job.get('filename', 'video'))[0]
+    dl_name = f"subtitles_{base_name}_karaoke.ass"
+    return Response(
+        content=build_ass(segments),
+        media_type="text/plain",
+        headers={"Content-Disposition": content_disposition(dl_name)},
+    )
+
+
 @router.get("/dub/export-segments/{job_id}")
 async def dub_export_segments_zip(job_id: str, lang: str = Query(None)):
     import zipfile

backend/api/routers/dub_generate.py

@@ -503,6 +503,18 @@ async def dub_generate(job_id: str, req: DubRequest):
         backend = await resolve_generation_backend(require_cloning=True)
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e))
+    except Exception as e:
+        from core.failure import is_gpu_oom
+
+        if not is_gpu_oom(e):
+            raise
+        from core.public_errors import public_exception_response
+
+        payload = public_exception_response(
+            e,
+            fallback="The TTS model could not be loaded.",
+        )
+        raise HTTPException(status_code=503, detail=payload["detail"]) from e
 
     async def _stream(task_id):
         total = len(req.segments)
@@ -1291,7 +1303,15 @@ async def dub_generate(job_id: str, req: DubRequest):
                         pass
                     _release_audio_tensors()
             except Exception as e:
-                yield f"data: {json.dumps({'type': 'error', 'segment': i, 'error': str(e)})}\n\n"
+                # A task-stream error bypasses the global exception handler.
+                # Never publish engine exception text here: allocator errors
+                # carry process tables and arbitrary failures can carry paths,
+                # tokens, or source text. The shared helper enriches recognized
+                # classes using VoiceStudio-owned constants only.
+                from core.public_errors import stream_generation_failure
+
+                error_detail = stream_generation_failure(e)["detail"]
+                yield f"data: {json.dumps({'type': 'error', 'segment': i, 'error': error_detail})}\n\n"
                 sr = backend.sample_rate
                 all_segment_wavs.append(_store_mix_wav(seg.start, seg.end, torch.zeros(1, max(0, int(seg_duration * sr))), sr, f"mix_{seg_id}"))
                 sync_scores.append(1.0)

backend/api/routers/engines.py

@@ -75,6 +75,21 @@ def list_tts_backends():
     return _family_payload("tts", tts_backend)
 
 
+@router.get(
+    "/engines/{engine_id}/disk-usage",
+    dependencies=[Depends(require_admin_action)],
+)
+def engine_disk_usage(engine_id: str):
+    """Measure owned engine bytes only when a catalogue row is opened."""
+    try:
+        tts_backend.get_backend_class(engine_id)
+    except ValueError:
+        raise HTTPException(status_code=404, detail="Unknown TTS engine")
+    from services.engine_disk_usage import disk_usage_for
+
+    return disk_usage_for(engine_id)
+
+
 @router.get("/engines/asr")
 def list_asr_backends():
     return _family_payload("asr", asr_backend)

backend/api/routers/generation.py

@@ -8,6 +8,7 @@ import asyncio
 import tempfile
 import contextlib
 import logging
+import threading
 import traceback
 from typing import Optional
 from fastapi import APIRouter, File, Form, UploadFile, HTTPException
@@ -32,6 +33,83 @@ router = APIRouter()
 logger = logging.getLogger("omnivoice.generate")
 
 
+class _TempReferenceLease:
+    """Delete a request-owned reference once every abandoned reader drains."""
+
+    def __init__(self, path: str):
+        self.path = path
+        self._lock = threading.Lock()
+        self._active = 0
+        self._request_done = False
+        self._deleted = False
+
+    def acquire(self):
+        with self._lock:
+            if self._request_done:
+                raise RuntimeError("reference lease acquired after request cleanup")
+            self._active += 1
+        once_lock = threading.Lock()
+        released = False
+
+        def release() -> None:
+            nonlocal released
+            with once_lock:
+                if released:
+                    return
+                released = True
+            self._release()
+
+        return release
+
+    def _release(self) -> None:
+        delete = False
+        with self._lock:
+            self._active -= 1
+            if self._active < 0:
+                raise RuntimeError("reference lease released too many times")
+            if self._request_done and self._active == 0 and not self._deleted:
+                self._deleted = True
+                delete = True
+        if delete:
+            with contextlib.suppress(OSError):
+                os.remove(self.path)
+
+    def finish_request(self) -> None:
+        delete = False
+        with self._lock:
+            self._request_done = True
+            if self._active == 0 and not self._deleted:
+                self._deleted = True
+                delete = True
+        if delete:
+            with contextlib.suppress(OSError):
+                os.remove(self.path)
+
+
+async def _run_with_reference_lease(lease, factory):
+    """Hold an ad-hoc reference through one local GPU-pool dispatch."""
+    if lease is None:
+        return await factory(None)
+    release = lease.acquire()
+    abandoned = False
+    try:
+        return await factory(release)
+    except GpuPoolBusyError:
+        # Busy means no job started; release now. The callback may already have
+        # done so, and the lease token is deliberately idempotent.
+        release()
+        abandoned = True
+        raise
+    except (asyncio.CancelledError, GpuJobTimeoutError):
+        # The guard owns release now: immediately for a queued cancellation,
+        # or from the worker finalizer after an in-flight job drains.
+        abandoned = True
+        raise
+    finally:
+        if not abandoned:
+            release()
+
+
 def _profile_instruct(row):
     """Validator-safe instruct for a stored profile row.
 
@@ -47,6 +125,96 @@ def _profile_instruct(row):
     return heal_design_instruct(row["instruct"], vd)
 
 
+def _resolve_profile_conditioning(row, *, ref_text=None, instruct=None,
+                                  seed=None, language=None):
+    """Resolve a ``voice_profiles`` row into generation conditioning.
+
+    Extracted verbatim from /generate's inline profile-resolution block so
+    other synthesis routes (POST /convert) share the exact same semantics —
+    lock wins, ``kind`` is authoritative (0005), legacy pre-0004 rows fall
+    back to the is_locked/instruct inference, and #533's language fill.
+
+    Request-supplied values (``ref_text``/``instruct``/``seed``/``language``)
+    always win over the stored row; only gaps are filled. Returns a dict with
+    ``ref_audio_path`` / ``ref_text`` / ``instruct`` / ``seed`` / ``language``
+    / ``kind`` plus ``persist_ref_text`` — True when the caller should cache
+    an auto-transcribed reference transcript back onto the row (#1032).
+    """
+    out = {
+        "ref_audio_path": None, "ref_text": ref_text, "instruct": instruct,
+        "seed": seed, "language": language, "kind": None,
+        "persist_ref_text": False,
+    }
+    # `kind` is authoritative (0005): 'design' profiles condition on their
+    # deterministic rendered sample + instruct; 'clone' on the user's
+    # reference. Lock always wins (it pins a specific take). Rows from
+    # pre-0004 DBs mid-upgrade may lack the column → fall back to the legacy
+    # is_locked/instruct inference.
+    try:
+        profile_kind = row["kind"] or "clone"
+    except (KeyError, IndexError):
+        profile_kind = "design" if (
+            row["instruct"] and not row["is_locked"] and not row["ref_audio_path"]
+        ) else "clone"
+    out["kind"] = profile_kind
+    if row["is_locked"] and row["locked_audio_path"]:
+        out["ref_audio_path"] = os.path.join(VOICES_DIR, row["locked_audio_path"])
+        if not out["ref_text"]:
+            out["ref_text"] = row["ref_text"]
+        if not out["instruct"]:
+            out["instruct"] = _profile_instruct(row)
+        if out["seed"] is None and row["seed"] is not None:
+            out["seed"] = row["seed"]
+    elif profile_kind == "design":
+        # Rendered sample (if present) carries the voice identity; instruct
+        # alone is the fallback for legacy archetype rows.
+        out["ref_audio_path"] = (
+            os.path.join(VOICES_DIR, row["ref_audio_path"]) if row["ref_audio_path"] else None
+        )
+        if out["ref_audio_path"] and not out["ref_text"] and row["ref_text"]:
+            out["ref_text"] = row["ref_text"]
+        if not out["instruct"]:
+            out["instruct"] = _profile_instruct(row)
+        if out["seed"] is None and row["seed"] is not None:
+            out["seed"] = row["seed"]
+    elif row["instruct"] and not row["is_locked"] and not row["ref_audio_path"]:
+        # Legacy design-shaped row (pre-0004 archetype materialization failure
+        # path): instruct-only conditioning.
+        if not out["instruct"]:
+            out["instruct"] = _profile_instruct(row)
+        if out["seed"] is None and row["seed"] is not None:
+            out["seed"] = row["seed"]
+    else:
+        out["ref_audio_path"] = (
+            os.path.join(VOICES_DIR, row["ref_audio_path"]) if row["ref_audio_path"] else None
+        )
+        if not out["ref_text"] and row["ref_text"]:
+            out["ref_text"] = row["ref_text"]
+        elif out["ref_audio_path"] and not out["ref_text"]:
+            # Empty stored transcript → the caller's auto-transcribe will run;
+            # cache its result onto the profile so it runs ONCE, not on every
+            # generate (#1032 perf regression).
+            out["persist_ref_text"] = True
+        if not out["instruct"] and row["instruct"]:
+            out["instruct"] = row["instruct"]
+        if out["seed"] is None and row["seed"] is not None:
+            out["seed"] = row["seed"]
+    if out["language"] == "Auto":
+        out["language"] = None
+    # #533: a profile's stored language must drive generation when the request
+    # didn't pin one. An EXPLICIT non-Auto request language still wins; we
+    # only fill the gap. `row` is a sqlite3.Row, so guard the column lookup
+    # for pre-language DBs mid-upgrade.
+    if out["language"] is None:
+        try:
+            prof_lang = row["language"]
+        except (KeyError, IndexError):
+            prof_lang = None
+        if prof_lang and prof_lang != "Auto":
+            out["language"] = prof_lang
+    return out
+
+
 def _note_generate_progress() -> None:
     """Tell the pool guard this render just finished a unit of work (#1391).
 
@@ -381,6 +549,31 @@ def _is_timeout_failure(e) -> bool:
     return False
 
 
+def _is_media_process_launch_failure(exc: BaseException) -> bool:
+    """Identify an ffmpeg/ffprobe launch ENOENT without guessing from a file name."""
+    if not isinstance(exc, FileNotFoundError):
+        return False
+
+    # A regular missing reference/model file may itself be named "ffmpeg".
+    # Require the innermost raise site to be Python's process launcher so that
+    # basename collisions keep the normal missing-file diagnosis (#1677).
+    traceback_cursor = exc.__traceback__
+    if traceback_cursor is None:
+        return False
+    while traceback_cursor.tb_next is not None:
+        traceback_cursor = traceback_cursor.tb_next
+    origin_module = traceback_cursor.tb_frame.f_globals.get("__name__", "")
+    if origin_module != "subprocess" and not origin_module.startswith("asyncio."):
+        return False
+
+    filename = getattr(exc, "filename", None)
+    if not filename:
+        return "[winerror 2]" in str(exc).lower()
+    return os.path.basename(str(filename)).lower() in {
+        "ffmpeg", "ffmpeg.exe", "ffprobe", "ffprobe.exe",
+    }
+
+
 def _oom_friendly_reraise(e):
     """Best-effort cache flush + the user-facing OOM hint shared by both
     inference paths."""
@@ -405,6 +598,21 @@ def _oom_friendly_reraise(e):
     # that lost its +x bit) is NOT an OOM — don't send the user to the Flush
     # button; tell them what's actually wrong.
     es = str(e)
+    # #1677: Windows CreateProcess reports a missing executable as a bare
+    # ``FileNotFoundError: [WinError 2] ...`` with no filename, while POSIX
+    # includes the missing ffmpeg/ffprobe name. The bundled-media downloader
+    # now republishes PATH as soon as it finishes, but a failed/blocked
+    # download still needs an actionable recovery rather than the unknown-
+    # error dead end. Keep missing reference/model files on their own path.
+    for _exc in _exception_chain(e):
+        if _is_media_process_launch_failure(_exc):
+            raise RuntimeError(
+                "A required media program couldn't be launched. Open "
+                "Settings → Audio tools and use "
+                "Download/Repair for the media engine, then retry. If Audio "
+                "tools is already ready, repair the selected TTS engine and "
+                f"restart VoiceStudio. Underlying error: {_safe_exc_text(_exc)}"
+            ) from e
     if isinstance(e, PermissionError) or "Permission denied" in es or "Errno 13" in es:
         raise RuntimeError(
             f"A required engine binary couldn't be executed (permission denied). "
@@ -695,15 +903,17 @@ def _run_backend_inference(
     backend, text, language, ref_audio_path, ref_text, instruct, duration,
     num_step, guidance_scale, speed, denoise, postprocess_output,
     used_seed, effect_preset="broadcast",
-    max_chunk_chars=None, crossfade_ms=None, *, dropped_sink=None,
+    max_chunk_chars=None, crossfade_ms=None, *, t_shift=None,
+    layer_penalty_factor=None, position_temperature=None,
+    class_temperature=None, dropped_sink=None,
 ):
     """Engine-aware twin of :func:`_run_inference` (issue #312).
 
     Runs the request through a pluggable ``TTSBackend`` adapter instead of the
-    VoiceStudio model directly. The adapter protocol is narrower than the
-    VoiceStudio-native surface — engine-specific extras (``t_shift``,
-    ``layer_penalty_factor``, …) only exist on the native path, which is why
-    VoiceStudio itself still goes through ``_run_inference``.
+    VoiceStudio model directly. A crash-isolated OmniVoice proxy advertises
+    ``supports_native_omnivoice_controls`` and receives the same advanced
+    controls and per-call seed as the native path; other adapters keep the
+    narrower protocol unchanged.
     """
     import torch
     try:
@@ -718,6 +928,18 @@ def _run_backend_inference(
             instruct=instruct, num_step=num_step, guidance_scale=guidance_scale,
             speed=speed, denoise=denoise, postprocess_output=postprocess_output,
         )
+        native_proxy = bool(
+            getattr(backend, "supports_native_omnivoice_controls", False)
+        )
+        if native_proxy:
+            gen_kwargs.update({
+                key: value for key, value in {
+                    "t_shift": t_shift,
+                    "layer_penalty_factor": layer_penalty_factor,
+                    "position_temperature": position_temperature,
+                    "class_temperature": class_temperature,
+                }.items() if value is not None
+            })
         sr = backend.sample_rate
 
         # Inline [pause Nms] markers (issue #276) work for every engine — the
@@ -727,10 +949,17 @@ def _run_backend_inference(
         has_pause = len(segments) > 1 or (segments and segments[0][1] > 0)
 
         if has_pause:
+            first_span = True
+
             def _gen_span(span_text):
+                nonlocal first_span
                 # Per-span duration is left to the engine; an explicit overall
                 # `duration` can't be meaningfully split across spans.
-                return backend.generate(span_text, duration=None, **gen_kwargs)
+                span_kwargs = dict(gen_kwargs)
+                if native_proxy and first_span and used_seed is not None:
+                    span_kwargs["seed"] = used_seed
+                first_span = False
+                return backend.generate(span_text, duration=None, **span_kwargs)
             audio_out = _render_with_pauses(_gen_span, segments, sr)
         else:
             # Wave 1.2: sentence-boundary chunking for long text (see
@@ -747,12 +976,19 @@ def _run_backend_inference(
                 for i, chunk_text in enumerate(text_chunks):
                     if used_seed is not None:
                         torch.manual_seed(used_seed + i)
-                    parts.append(backend.generate(chunk_text, duration=None, **gen_kwargs))
+                    chunk_kwargs = dict(gen_kwargs)
+                    if native_proxy and used_seed is not None:
+                        chunk_kwargs["seed"] = used_seed + i
+                    parts.append(backend.generate(
+                        chunk_text, duration=None, **chunk_kwargs
+                    ))
                     _note_generate_progress()
                 audio_out = concatenate_audio_chunks(parts, sr, _xfade_ms,
                                                      texts=text_chunks,
                                                      sink=dropped_sink)
             else:
+                if native_proxy and used_seed is not None:
+                    gen_kwargs["seed"] = used_seed
                 audio_out = backend.generate(text, duration=duration, **gen_kwargs)
 
         return _apply_effect_chain(
@@ -1297,6 +1533,7 @@ async def generate_speech(
 
     ref_audio_path = None
     cleanup_ref = False
+    ref_lease = None
     used_seed = seed
     resolved_profile_id = None
     history_mode = None  # profile.kind when a profile drives; else inferred at insert
@@ -1314,76 +1551,28 @@ async def generate_speech(
             row = conn.execute("SELECT * FROM voice_profiles WHERE id=?", (profile_id,)).fetchone()
         if row:
             resolved_profile_id = profile_id
-            # `kind` is authoritative (0005): 'design' profiles condition on
-            # their deterministic rendered sample + instruct; 'clone' on the
-            # user's reference. Lock always wins (it pins a specific take).
-            # Rows from pre-0004 DBs mid-upgrade may lack the column → fall
-            # back to the legacy is_locked/instruct inference.
-            try:
-                profile_kind = row["kind"] or "clone"
-            except (KeyError, IndexError):
-                profile_kind = "design" if (row["instruct"] and not row["is_locked"] and not row["ref_audio_path"]) else "clone"
-            history_mode = profile_kind
-            if row["is_locked"] and row["locked_audio_path"]:
-                ref_audio_path = os.path.join(VOICES_DIR, row["locked_audio_path"])
-                if not ref_text:
-                    ref_text = row["ref_text"]
-                if not instruct:
-                    instruct = _profile_instruct(row)
-                if used_seed is None and row["seed"] is not None:
-                    used_seed = row["seed"]
-            elif profile_kind == "design":
-                # Rendered sample (if present) carries the voice identity;
-                # instruct alone is the fallback for legacy archetype rows.
-                ref_audio_path = os.path.join(VOICES_DIR, row["ref_audio_path"]) if row["ref_audio_path"] else None
-                if ref_audio_path and not ref_text and row["ref_text"]:
-                    ref_text = row["ref_text"]
-                if not instruct:
-                    instruct = _profile_instruct(row)
-                if used_seed is None and row["seed"] is not None:
-                    used_seed = row["seed"]
-            elif row["instruct"] and not row["is_locked"] and not row["ref_audio_path"]:
-                # Legacy design-shaped row (pre-0004 archetype materialization
-                # failure path): instruct-only conditioning.
-                if not instruct:
-                    instruct = _profile_instruct(row)
-                if used_seed is None and row["seed"] is not None:
-                    used_seed = row["seed"]
-            else:
-                ref_audio_path = os.path.join(VOICES_DIR, row["ref_audio_path"]) if row["ref_audio_path"] else None
-                if not ref_text and row["ref_text"]:
-                    ref_text = row["ref_text"]
-                elif ref_audio_path and not ref_text:
-                    # Empty stored transcript → the auto-transcribe below will
-                    # run; cache its result onto the profile so it runs ONCE,
-                    # not on every generate (#1032 perf regression).
-                    persist_ref_text_profile_id = profile_id
-                if not instruct and row["instruct"]:
-                    instruct = row["instruct"]
-                if used_seed is None and row["seed"] is not None:
-                    used_seed = row["seed"]
-            if language == "Auto":
-                language = None
-            # #533: a profile's stored language must drive generation when the
-            # request didn't pin one. Without this the German (etc.) archetype
-            # generates with language=None and the model drifts to English —
-            # even though the archetype PREVIEW renders correctly (archetypes.py
-            # passes the language). An EXPLICIT non-Auto request language still
-            # wins; we only fill the gap. `row` is a sqlite3.Row, so guard the
-            # column lookup for pre-language DBs mid-upgrade.
-            if language is None:
-                try:
-                    prof_lang = row["language"]
-                except (KeyError, IndexError):
-                    prof_lang = None
-                if prof_lang and prof_lang != "Auto":
-                    language = prof_lang
+            # Shared with POST /convert — see _resolve_profile_conditioning
+            # for the resolution rules (kind-authoritative, lock wins, #533
+            # language fill, #1032 transcript-cache signal).
+            _cond = _resolve_profile_conditioning(
+                row, ref_text=ref_text, instruct=instruct, seed=used_seed,
+                language=language,
+            )
+            history_mode = _cond["kind"]
+            ref_audio_path = _cond["ref_audio_path"]
+            ref_text = _cond["ref_text"]
+            instruct = _cond["instruct"]
+            used_seed = _cond["seed"]
+            language = _cond["language"]
+            if _cond["persist_ref_text"]:
+                persist_ref_text_profile_id = profile_id
     elif ref_audio is not None:
         try:
             with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
                 f.write(await ref_audio.read())
                 ref_audio_path = f.name
                 cleanup_ref = True
+                ref_lease = _TempReferenceLease(ref_audio_path)
         except Exception as e:
             raise HTTPException(status_code=500, detail=str(e))
 
@@ -1400,13 +1589,19 @@ async def generate_speech(
         # built-in ASR fallback), so a timeout degrades to None rather than
         # failing the whole generate.
         try:
-            ref_text = await run_on_gpu_pool_guarded(
-                functools.partial(transcribe_reference, ref_audio_path),
-                what="Reference transcribe",
-                # Floor budget (#1190): a reference clip is seconds of audio,
-                # so the length-scaled bonus never applies — but the timeout is
-                # explicit here too, so no dispatch relies on a hidden default.
-                timeout=_generate_timeout_s("", execution_device=_routing["effective_device"]),
+            ref_text = await _run_with_reference_lease(
+                ref_lease,
+                lambda release: run_on_gpu_pool_guarded(
+                    functools.partial(transcribe_reference, ref_audio_path),
+                    what="Reference transcribe",
+                    # Floor budget (#1190): a reference clip is seconds of audio,
+                    # so the length-scaled bonus never applies — but the timeout is
+                    # explicit here too, so no dispatch relies on a hidden default.
+                    timeout=_generate_timeout_s(
+                        "", execution_device=_routing["effective_device"]
+                    ),
+                    on_abandon=release,
+                )
             )
         # TimeoutError covers both the execution bound and pool saturation:
         # this path is best-effort either way.
@@ -1662,19 +1857,30 @@ async def generate_speech(
                     "target_label": e.worker_label or _target_label,
                     "hint": e.hint,
                 })
-            except Exception:
+            except Exception as exc:
                 # Mid-job remote failure is NOT quietly redone here: the client
                 # treats a retryable error as "surface it", so the user decides
-                # whether to spend the same minutes again on this machine.
-                logger.error("Remote generation failed", exc_info=True)
-                from core.public_errors import stream_failure
-                yield _line({"type": "error", **stream_failure("generation_failed")})
+                # whether to spend the same minutes again on this machine. Like
+                # the local streaming path, this in-band frame stands in for the
+                # global 500 handler, so it journals the scrubbed failure and
+                # names a recognized cause instead of the bare generic string
+                # (#1607).
+                logger.error(
+                    "Remote generation failed (class=%s)",
+                    type(exc).__name__,
+                )
+                from core.public_errors import stream_generation_failure
+                from core import error_journal
+
+                error_journal.record(
+                    exc, route="/generate", trace=traceback.format_exc()
+                )
+                yield _line({"type": "error", **stream_generation_failure(exc)})
             finally:
                 if not render.done():
                     render.cancel()
-                if cleanup_ref and ref_audio_path:
-                    with contextlib.suppress(OSError):
-                        os.remove(ref_audio_path)
+                if cleanup_ref and ref_lease is not None:
+                    ref_lease.finish_request()
 
         return StreamingResponse(
             _remote_stream_events(),
@@ -1716,6 +1922,17 @@ async def generate_speech(
                         instruct=instruct, num_step=num_step,
                         guidance_scale=guidance_scale, speed=speed,
                         denoise=denoise, postprocess_output=postprocess_output,
+                        **({
+                            key: value for key, value in {
+                                "t_shift": t_shift,
+                                "layer_penalty_factor": layer_penalty_factor,
+                                "position_temperature": position_temperature,
+                                "class_temperature": class_temperature,
+                                "seed": used_seed + i if used_seed is not None else None,
+                            }.items() if value is not None
+                        } if getattr(
+                            _backend, "supports_native_omnivoice_controls", False
+                        ) else {}),
                     )
                     sr = _backend.sample_rate
                     skip = getattr(_backend, "applies_own_mastering", False)
@@ -1779,33 +1996,45 @@ async def generate_speech(
                 if _has_pause or len(_text_chunks) <= 1:
                     # Single-shot pipeline, unchanged — streamed as one chunk.
                     if _backend is not None:
-                        audio_tensor = await run_on_gpu_pool_guarded(
-                            functools.partial(
-                                _run_backend_inference,
-                                _backend, text, language, ref_audio_path, ref_text,
-                                instruct, duration, num_step, guidance_scale, speed,
-                                denoise, postprocess_output, used_seed, effect_preset,
-                                max_chunk_chars, crossfade_ms, dropped_sink=_dropped_sink,
-                            ),
-                            what="TTS generate",
-                            min_vram_gb=_engine_min_vram_gb,
-                            timeout=_generate_timeout_s(text, execution_device=_routing["effective_device"]),
+                        audio_tensor = await _run_with_reference_lease(
+                            ref_lease,
+                            lambda release: run_on_gpu_pool_guarded(
+                                functools.partial(
+                                    _run_backend_inference,
+                                    _backend, text, language, ref_audio_path, ref_text,
+                                    instruct, duration, num_step, guidance_scale, speed,
+                                    denoise, postprocess_output, used_seed, effect_preset,
+                                    max_chunk_chars, crossfade_ms, t_shift=t_shift,
+                                    layer_penalty_factor=layer_penalty_factor,
+                                    position_temperature=position_temperature,
+                                    class_temperature=class_temperature,
+                                    dropped_sink=_dropped_sink,
+                                ),
+                                what="TTS generate",
+                                min_vram_gb=_engine_min_vram_gb,
+                                timeout=_generate_timeout_s(text, execution_device=_routing["effective_device"]),
+                                on_abandon=release,
+                            )
                         )
                         sample_rate = _backend.sample_rate
                     else:
-                        audio_tensor = await run_on_gpu_pool_guarded(
-                            functools.partial(
-                                _run_inference,
-                                _model, text, language, ref_audio_path, ref_text,
-                                instruct, duration, num_step, guidance_scale, speed,
-                                t_shift, denoise, postprocess_output,
-                                layer_penalty_factor, position_temperature,
-                                class_temperature, used_seed, effect_preset,
-                                max_chunk_chars, crossfade_ms, dropped_sink=_dropped_sink,
-                            ),
-                            what="TTS generate",
-                            min_vram_gb=_engine_min_vram_gb,
-                            timeout=_generate_timeout_s(text, execution_device=_routing["effective_device"]),
+                        audio_tensor = await _run_with_reference_lease(
+                            ref_lease,
+                            lambda release: run_on_gpu_pool_guarded(
+                                functools.partial(
+                                    _run_inference,
+                                    _model, text, language, ref_audio_path, ref_text,
+                                    instruct, duration, num_step, guidance_scale, speed,
+                                    t_shift, denoise, postprocess_output,
+                                    layer_penalty_factor, position_temperature,
+                                    class_temperature, used_seed, effect_preset,
+                                    max_chunk_chars, crossfade_ms, dropped_sink=_dropped_sink,
+                                ),
+                                what="TTS generate",
+                                min_vram_gb=_engine_min_vram_gb,
+                                timeout=_generate_timeout_s(text, execution_device=_routing["effective_device"]),
+                                on_abandon=release,
+                            )
                         )
                         sample_rate = _model.sampling_rate
                     yield _line({
@@ -1834,14 +2063,18 @@ async def generate_speech(
                     for i, chunk_text in enumerate(_text_chunks):
                         # Bounded per chunk + pool-reset on hang (#730 class);
                         # a timeout surfaces as an "error" event below.
-                        raw, preview, sample_rate = await run_on_gpu_pool_guarded(
-                            functools.partial(_render_stream_chunk, i, chunk_text),
-                            what="TTS generate",
-                            min_vram_gb=_engine_min_vram_gb,
-                            # Budget scaled to THIS chunk (#1190) — the flat
-                            # 300s here is what made long streamed renders fail
-                            # even after the v0.3.22 scaled budget shipped.
-                            timeout=_generate_timeout_s(chunk_text, execution_device=_routing["effective_device"]),
+                        raw, preview, sample_rate = await _run_with_reference_lease(
+                            ref_lease,
+                            lambda release: run_on_gpu_pool_guarded(
+                                functools.partial(_render_stream_chunk, i, chunk_text),
+                                what="TTS generate",
+                                min_vram_gb=_engine_min_vram_gb,
+                                # Budget scaled to THIS chunk (#1190) — the flat
+                                # 300s here is what made long streamed renders fail
+                                # even after the v0.3.22 scaled budget shipped.
+                                timeout=_generate_timeout_s(chunk_text, execution_device=_routing["effective_device"]),
+                                on_abandon=release,
+                            )
                         )
                         parts.append(raw)
                         # Provenance-mark the streamed copy off the GPU pool
@@ -1915,16 +2148,32 @@ async def generate_speech(
                 logger.error("Streaming generation request rejected")
                 from core.public_errors import stream_failure
                 yield _line({"type": "error", **stream_failure("invalid_request")})
-            except Exception:
-                logger.error("Streaming generation failed unexpectedly")
-                from core.public_errors import stream_failure
-                yield _line({"type": "error", **stream_failure("generation_failed")})
+            except Exception as exc:
+                # A streaming request answers 200 and carries its failure as an
+                # in-band error frame, so it never reaches the global 500
+                # handler — which is where a classic /generate failure gets its
+                # scrubbed journal entry (Diagnostics / recent errors) AND its
+                # classified, actionable message. Both have to be reproduced
+                # here or a streaming generation failure is invisible in the
+                # diagnostic bundle and opaque to the user (#1607). The raw
+                # exception is NOT logged: it can carry a reference-clip path or
+                # a provider secret, and only the journal scrubs before storing.
+                logger.error(
+                    "Streaming generation failed unexpectedly (class=%s)",
+                    type(exc).__name__,
+                )
+                from core.public_errors import stream_generation_failure
+                from core import error_journal
+
+                error_journal.record(
+                    exc, route="/generate", trace=traceback.format_exc()
+                )
+                yield _line({"type": "error", **stream_generation_failure(exc)})
             finally:
                 # Ownership of the temp reference clip moves to this generator
                 # in stream mode (the route returns before rendering starts).
-                if cleanup_ref and ref_audio_path:
-                    with contextlib.suppress(OSError):
-                        os.remove(ref_audio_path)
+                if cleanup_ref and ref_lease is not None:
+                    ref_lease.finish_request()
 
         # Routing notice (#21): known before the stream starts, so it rides the
         # same headers the classic path uses — and now also carries "your
@@ -1962,7 +2211,11 @@ async def generate_speech(
                     _backend, text, language, ref_audio_path, ref_text, instruct,
                     duration, num_step, guidance_scale, speed, denoise,
                     postprocess_output, used_seed, effect_preset,
-                    max_chunk_chars, crossfade_ms, dropped_sink=_dropped_text,
+                    max_chunk_chars, crossfade_ms, t_shift=t_shift,
+                    layer_penalty_factor=layer_penalty_factor,
+                    position_temperature=position_temperature,
+                    class_temperature=class_temperature,
+                    dropped_sink=_dropped_text,
                 )
             else:
                 _local_render = functools.partial(
@@ -1973,14 +2226,18 @@ async def generate_speech(
                     class_temperature, used_seed, effect_preset,
                     max_chunk_chars, crossfade_ms, dropped_sink=_dropped_text,
                 )
-            audio_tensor = await gpu_gateway.run(
-                _REMOTE_OP,
-                local=gpu_gateway.LocalCall(
-                    _local_render, what="TTS generate",
-                    timeout=_generate_timeout_s(text, execution_device=_routing["effective_device"]),
-                    min_vram_gb=_engine_min_vram_gb,
-                ),
-                decision=_decision,
+            audio_tensor = await _run_with_reference_lease(
+                ref_lease,
+                lambda release: gpu_gateway.run(
+                    _REMOTE_OP,
+                    local=gpu_gateway.LocalCall(
+                        _local_render, what="TTS generate",
+                        timeout=_generate_timeout_s(text, execution_device=_routing["effective_device"]),
+                        min_vram_gb=_engine_min_vram_gb,
+                        on_abandon=release,
+                    ),
+                    decision=_decision,
+                )
             )
             # Read after generation: engines with lazy model loading report
             # their real rate only once weights are up.
@@ -2112,9 +2369,8 @@ async def generate_speech(
             ),
         )
     finally:
-        if cleanup_ref and ref_audio_path:
-            with contextlib.suppress(OSError):
-                os.remove(ref_audio_path)
+        if cleanup_ref and ref_lease is not None:
+            ref_lease.finish_request()
 
 def _safe_output_path(name):
     if not name:

backend/api/routers/settings.py

@@ -1035,9 +1035,10 @@ def set_asr_openai_compat(body: _ASROpenAICompatBody):
     from services import asr_backend, settings_store
 
     if body.base_url is not None:
-        url = body.base_url.strip().rstrip("/")
-        if url and not url.startswith(("http://", "https://")):
-            raise HTTPException(status_code=400, detail="Base URL must start with http(s)://")
+        try:
+            url = asr_backend.normalize_openai_compat_asr_base_url(body.base_url)
+        except ValueError as exc:
+            raise HTTPException(status_code=400, detail=str(exc)) from exc
         settings_store.set_text(asr_backend._ASR_OPENAI_COMPAT_BASE_URL_KEY, url)
     if body.model is not None:
         settings_store.set_text(

backend/api/routers/setup/download.py

@@ -76,6 +76,7 @@ _cancelled: set[str] = set()
 _active_installs: set[str] = set()
 _active_installs_lock = threading.Lock()
 _install_tasks: set[asyncio.Task] = set()
+_install_tasks_by_repo: dict[str, asyncio.Task] = {}
 
 
 def _download_max_workers() -> int:
@@ -420,16 +421,11 @@ async def install_model(req: InstallModelRequest):
                 f"Retry in {remaining}s or check your network."
             ),
         )
-    with _active_installs_lock:
-        if req.repo_id in _active_installs:
-            return {"status": "already_running", "repo_id": req.repo_id}
-        _active_installs.add(req.repo_id)
     loop = asyncio.get_running_loop()
 
     def _do():
         token = hf_progress.current_repo_id.set(req.repo_id)
         target_token = hf_progress.current_target.set("local")
-        _cancelled.discard(req.repo_id)  # clear any stale cancel from a prior run
         hf_progress.emit({
             "repo_id": req.repo_id,
             "filename": req.repo_id,
@@ -688,17 +684,53 @@ async def install_model(req: InstallModelRequest):
             with _active_installs_lock:
                 _active_installs.discard(req.repo_id)
 
-    try:
-        task = loop.create_task(asyncio.to_thread(_do))
-        _install_tasks.add(task)
-        task.add_done_callback(_install_tasks.discard)
-    except Exception:
-        with _active_installs_lock:
+    with _active_installs_lock:
+        if req.repo_id in _active_installs:
+            return {"status": "already_running", "repo_id": req.repo_id}
+        _active_installs.add(req.repo_id)
+        # Admission and task publication are one atomic generation boundary:
+        # cancellation can never observe an admitted install without its task.
+        _cancelled.discard(req.repo_id)
+        try:
+            task = loop.create_task(asyncio.to_thread(_do))
+            _install_tasks.add(task)
+            _install_tasks_by_repo[req.repo_id] = task
+        except Exception:
             _active_installs.discard(req.repo_id)
-        raise
+            raise
+
+    def install_finished(completed: asyncio.Task) -> None:
+        with _active_installs_lock:
+            _install_tasks.discard(completed)
+            if _install_tasks_by_repo.get(req.repo_id) is completed:
+                _install_tasks_by_repo.pop(req.repo_id, None)
+
+    task.add_done_callback(install_finished)
     return {"status": "install_started", "repo_id": req.repo_id}
 
 
+async def cancel_install_and_wait(repo_id: str) -> None:
+    """Request cancellation and retain authority until its thread exits."""
+    from worker.async_utils import drain_task  # noqa: PLC0415
+
+    with _active_installs_lock:
+        _cancelled.add(repo_id)
+        _install_cooldowns.pop(repo_id, None)
+        task = _install_tasks_by_repo.get(repo_id)
+    if task is None:
+        return
+    try:
+        # asyncio.to_thread cannot stop snapshot_download mid-file. Cancelling
+        # its wrapper would only detach the thread, so wait until the blocking
+        # call observes the flag or naturally returns.
+        await drain_task(task)
+    finally:
+        with _active_installs_lock:
+            current = _install_tasks_by_repo.get(repo_id)
+            if current is None or current is task:
+                _cancelled.discard(repo_id)
+
+
 @router.post("/models/install/cancel")
 async def cancel_install(req: InstallModelRequest):
     """Request cancellation of an in-flight install (FDL-11).

backend/api/routers/speech_platform.py

@@ -0,0 +1,160 @@
+"""Discovery contract for VoiceStudio's local speech platform.
+
+Interfaces should discover this document instead of hard-coding whichever
+dictation route the desktop happens to use.  Endpoint URLs are relative so the
+same response works on loopback, a tailnet GPU host, and a reverse proxy.
+"""
+from __future__ import annotations
+
+import os
+from typing import Literal
+
+from fastapi import APIRouter
+from pydantic import BaseModel, Field
+
+from core.version import APP_VERSION
+
+router = APIRouter(tags=["Speech Platform"])
+
+SPEECH_PROTOCOL = "voicestudio.speech.v1"
+STREAM_PATH = "/v1/audio/transcriptions/stream"
+
+
+class EndpointCapability(BaseModel):
+    path: str
+    transport: Literal["http", "websocket", "mcp-streamable-http", "mcp-stdio"]
+    method: str | None = None
+    protocol: str | None = None
+
+
+class StreamInputCapability(BaseModel):
+    framing: Literal["binary"] = "binary"
+    formats: list[str]
+    default_format: str
+    sample_rate_query: str = "sr"
+    end_control: dict[str, str]
+
+
+class StreamOutputCapability(BaseModel):
+    framing: Literal["json"] = "json"
+    events: list[str]
+    final_kinds: list[str]
+
+
+class SpeechFeatureCapabilities(BaseModel):
+    batch_transcription: bool = True
+    streaming_transcription: bool = True
+    partial_transcripts: bool = True
+    utterance_finals: bool = True
+    session_summary: bool = True
+    word_timestamps: bool = True
+    local_refinement: bool = True
+    acoustic_echo_cancellation: bool = True
+    native_dictation_control: bool = False
+
+
+class SpeechAuthCapabilities(BaseModel):
+    loopback: Literal["none"] = "none"
+    remote: Literal["bearer"] = "bearer"
+    header: str = "Authorization: Bearer <OMNIVOICE_API_KEY>"
+    browser_session_endpoint: str = "/api/auth/session"
+    websocket_ticket_endpoint: str = "/api/auth/ws-ticket"
+    websocket_ticket_query_parameter: Literal["ws_ticket"] = "ws_ticket"
+
+
+class SpeechCapabilities(BaseModel):
+    schema_: Literal["voicestudio.speech-capabilities"] = Field(
+        default="voicestudio.speech-capabilities",
+        serialization_alias="schema",
+    )
+    protocol: Literal["voicestudio.speech.v1"] = SPEECH_PROTOCOL
+    protocol_version: Literal["1.0"] = "1.0"
+    service: str = "VoiceStudio"
+    service_version: str = APP_VERSION
+    local_first: bool = True
+    endpoints: dict[str, EndpointCapability]
+    stream_input: StreamInputCapability
+    stream_output: StreamOutputCapability
+    features: SpeechFeatureCapabilities
+    authentication: SpeechAuthCapabilities
+
+
+def speech_capabilities() -> SpeechCapabilities:
+    """Return the stable, side-effect-free integration contract."""
+    endpoints = {
+        "capabilities": EndpointCapability(
+            path="/.well-known/voicestudio-speech",
+            transport="http",
+            method="GET",
+        ),
+        "batch_transcription": EndpointCapability(
+            path="/v1/audio/transcriptions",
+            transport="http",
+            method="POST",
+            protocol="openai.audio.transcriptions",
+        ),
+        "streaming_transcription": EndpointCapability(
+            path=STREAM_PATH,
+            transport="websocket",
+            protocol=SPEECH_PROTOCOL,
+        ),
+        "mcp": EndpointCapability(
+            path="/mcp",
+            transport="mcp-streamable-http",
+            method="POST",
+            protocol="mcp",
+        ),
+        "mcp_stdio": EndpointCapability(
+            path="python -m backend.mcp_shim",
+            transport="mcp-stdio",
+            protocol="mcp",
+        ),
+    }
+    native_control = False
+    try:
+        control_port = int(os.environ.get("VOICESTUDIO_SPEECH_CONTROL_PORT", ""))
+    except (TypeError, ValueError):
+        control_port = 0
+    if 0 < control_port <= 65535:
+        native_control = True
+        endpoints["native_dictation_control"] = EndpointCapability(
+            path=f"http://127.0.0.1:{control_port}/v1/capabilities",
+            transport="http",
+            method="GET",
+            protocol=SPEECH_PROTOCOL,
+        )
+
+    return SpeechCapabilities(
+        endpoints=endpoints,
+        stream_input=StreamInputCapability(
+            formats=[
+                "audio/pcm;encoding=s16le;channels=1",
+                "audio/webm;codecs=opus",
+            ],
+            default_format="audio/webm;codecs=opus",
+            end_control={"type": "input_audio.end"},
+        ),
+        stream_output=StreamOutputCapability(
+            events=["session.started", "status", "partial", "final", "error"],
+            final_kinds=["utterance", "summary"],
+        ),
+        features=SpeechFeatureCapabilities(
+            native_dictation_control=native_control,
+        ),
+        authentication=SpeechAuthCapabilities(),
+    )
+
+
+@router.get(
+    "/.well-known/voicestudio-speech",
+    response_model=SpeechCapabilities,
+    response_model_by_alias=True,
+)
+@router.get(
+    "/v1/audio/capabilities",
+    response_model=SpeechCapabilities,
+    response_model_by_alias=True,
+)
+async def get_speech_capabilities() -> SpeechCapabilities:
+    """Advertise batch, streaming, and agent-facing speech transports."""
+    return speech_capabilities()

backend/api/routers/system.py

@@ -203,8 +203,22 @@ def system_info():
     """
     try:
         _ffmpeg = find_ffmpeg()
+        from services import model_manager as _mm
+        from core import prefs as _prefs_mod
         return {
             "app_version": APP_VERSION,
+            "generate_timeout_s": _mm.GPU_JOB_TIMEOUT_S,
+            "cpu_generate_timeout_s": _mm.CPU_JOB_TIMEOUT_S,
+            # #1787 review fix: a saved prefs.json value for either key can be
+            # silently shadowed by an external env var (os.environ.setdefault
+            # in core.prefs.restore_env is a no-op when one is already
+            # present) — the Settings panel must say so rather than promise a
+            # restart will apply a value that never will.
+            "generate_timeout_shadowed": _prefs_mod.is_env_shadowed(
+                "OMNIVOICE_GENERATE_TIMEOUT_S"),
+            "cpu_generate_timeout_shadowed": _prefs_mod.is_env_shadowed(
+                "OMNIVOICE_CPU_GENERATE_TIMEOUT_S"),
+            "code_fingerprint": os.environ.get("OMNIVOICE_BUILD_FINGERPRINT", ""),
             "data_dir": DATA_DIR,
             "outputs_dir": OUTPUTS_DIR,
             "crash_log_path": CRASH_LOG_PATH,
@@ -240,6 +254,11 @@ def system_info():
         logger.exception("system_info failed — returning safe defaults")
         return {
             "app_version": APP_VERSION,
+            "generate_timeout_s": 300.0,
+            "cpu_generate_timeout_s": 600.0,
+            "generate_timeout_shadowed": False,
+            "cpu_generate_timeout_shadowed": False,
+            "code_fingerprint": os.environ.get("OMNIVOICE_BUILD_FINGERPRINT", ""),
             "data_dir": DATA_DIR,
             "outputs_dir": OUTPUTS_DIR,
             "crash_log_path": str(CRASH_LOG_PATH),
@@ -848,6 +867,14 @@ PERSISTENT_KEYS = {
     # the Rust sidecar reads OMNIVOICE_PORT at startup and the backend derives
     # the LAN-share/UI ports from the others.
     "OMNIVOICE_PORT", "OMNIVOICE_SHARE_PORT", "OMNIVOICE_UI_PORT",
+    # Per-job compute-time budgets (#1787). Both are captured at import time
+    # by services/model_manager.py (GPU_JOB_TIMEOUT_S / CPU_JOB_TIMEOUT_S), so
+    # a value saved here takes effect on the NEXT backend restart — same
+    # contract as OMNIVOICE_PORT above. Restored into os.environ during the
+    # "env_prefs" startup step (main.py), which runs before model_manager is
+    # first imported ("ml_imports"), so the restored value is what the module
+    # captures. The Settings UI must say so (RestartBadge).
+    "OMNIVOICE_GENERATE_TIMEOUT_S", "OMNIVOICE_CPU_GENERATE_TIMEOUT_S",
 }
 
 # Sidecar-engine install dirs (OMNIVOICE_INDEXTTS_DIR, …). The one-click
@@ -865,6 +892,16 @@ except Exception:  # pragma: no cover — defensive: env panel > installer wirin
 # being set so a bad value never reaches uvicorn / the share listener.
 _PORT_KEYS = {"OMNIVOICE_PORT", "OMNIVOICE_SHARE_PORT", "OMNIVOICE_UI_PORT"}
 
+# Keys whose value is a wall-clock compute-time budget in seconds (#1787).
+# Validated the same way as _PORT_KEYS: reject anything that isn't a
+# positive number before it reaches services/model_manager.py. Upper bound is
+# generous — long enough that a legitimate multi-hour, audiobook-length CPU
+# render is never blocked — but still bounded, so a fat-fingered extra digit
+# (300 -> 3000000) can't turn a wedged job into one that silently occupies a
+# worker for days before the guard ever fires.
+_TIMEOUT_KEYS = {"OMNIVOICE_GENERATE_TIMEOUT_S", "OMNIVOICE_CPU_GENERATE_TIMEOUT_S"}
+_MAX_GENERATE_TIMEOUT_S = 21600.0  # 6 hours
+
 
 @router.post("/system/set-env")
 async def set_env_var(body: dict):
@@ -908,6 +945,22 @@ async def set_env_var(body: dict):
                     status_code=400,
                     detail=f"Invalid port for {key}: must be between 1024 and 65535.",
                 )
+        if key in _TIMEOUT_KEYS:
+            try:
+                timeout_n = float(value)
+            except (TypeError, ValueError):
+                raise HTTPException(
+                    status_code=400,
+                    detail=f"Invalid timeout for {key}: '{value}' is not a number.",
+                )
+            if not (0 < timeout_n <= _MAX_GENERATE_TIMEOUT_S):
+                raise HTTPException(
+                    status_code=400,
+                    detail=(
+                        f"Invalid timeout for {key}: must be greater than 0 "
+                        f"and at most {_MAX_GENERATE_TIMEOUT_S:.0f} seconds."
+                    ),
+                )
         os.environ[key] = value
         logger.info("Environment variable set (length=%d)", len(value))
 
@@ -950,7 +1003,13 @@ async def set_env_var(body: dict):
         else:
             prefs_delete(prefs_key)
 
-    return {"key": key, "set": bool(value)}
+    # #1787 review fix: tell the caller up front when the value just saved is
+    # being shadowed by an external env var — set at THIS process's startup,
+    # before our own prefs restore ran, so it predicts the next restart too.
+    # A response that just said {"set": True} let the Settings panel promise
+    # a restart would apply a value that never will.
+    from core.prefs import is_env_shadowed
+    return {"key": key, "set": bool(value), "shadowed": is_env_shadowed(key)}
 
 
 @router.post("/clean-audio")

backend/api/routers/tts_stream.py

@@ -39,6 +39,26 @@ CHUNK_SAMPLES = int(os.environ.get("OMNIVOICE_STREAM_CHUNK", "4800"))
 _perf_counter = time.perf_counter
 
 
+async def _resolve_stream_backend(engine_id: str | None):
+    """Resolve the live-stream engine without bypassing host isolation."""
+    from services.tts_backend import (
+        OmniVoiceBackend,
+        active_backend_id,
+        get_active_tts_backend,
+        get_backend_class,
+    )
+
+    if engine_id:
+        return get_backend_class(engine_id)()
+
+    cls = get_backend_class(active_backend_id())
+    if cls is OmniVoiceBackend:
+        from services.model_manager import get_model
+
+        return get_active_tts_backend(model=await get_model())
+    return get_active_tts_backend()
+
+
 class StreamTTSRequest(BaseModel):
     """Client request for streaming TTS."""
     text: str
@@ -132,10 +152,6 @@ async def ws_tts(websocket: WebSocket):
 
             try:
                 # Resolve engine
-                from services.tts_backend import (
-                    get_active_tts_backend,
-                    get_backend_class,
-                )
                 engine_id = data.get("engine")
                 # #1224: leave a breadcrumb when memory is already tight before
                 # a heavy load. /generate has done this since the 16 GB-Mac
@@ -151,13 +167,7 @@ async def ws_tts(websocket: WebSocket):
                     log_if_low(f"TTS stream load ({engine_id or 'active engine'})")
                 except Exception:
                     pass
-                if engine_id:
-                    cls = get_backend_class(engine_id)
-                    backend = cls()
-                else:
-                    from services.model_manager import get_model
-                    model = await get_model()
-                    backend = get_active_tts_backend(model=model)
+                backend = await _resolve_stream_backend(engine_id)
 
                 # ── Routing gate (#21 — no silent CPU fallback). WebSockets have
                 # no response headers, so this uses frames: an error frame +

backend/api/routers/voice_convert.py

@@ -0,0 +1,378 @@
+"""Speech-to-speech voice changer — Studio's Convert method (POST /convert).
+
+The user drops (or records) a source clip, picks an existing voice profile,
+and gets the same words back in that profile's voice: the active ASR backend
+transcribes the clip (no word timestamps — the text is all we need), the
+active TTS engine re-synthesizes it conditioned on the profile's reference
+audio, and — by default — the take is pitch-preservingly time-stretched
+(ffmpeg atempo, clamped to one well-behaved 0.5–2.0 stage) so it lands near
+the source clip's duration.
+
+Deliberately reuses the /generate choke points instead of re-deriving them:
+
+* profile row → conditioning via ``generation._resolve_profile_conditioning``
+  (lock wins, ``kind`` authoritative, #533 language fill),
+* engine resolution via ``services.tts_backend.resolve_generation_backend``
+  (never a silent OmniVoice fallback; ``require_cloning=True`` refuses
+  clone-less engines with the actionable switch-engine message),
+* synthesis via ``generation._run_backend_inference`` on the guarded GPU
+  pool (#730 bound + reset; busy/timeout → retryable 503),
+* provenance + persistence via ``services.watermark.mark_synthetic_async``
+  and ``generation._finalize_generation`` (watermark → WAV in OUTPUTS_DIR →
+  history row → retention prune), marked AFTER the stretch so the take users
+  keep carries exactly one whole-take mark.
+
+Local-first: no network calls; ASR-model-less installs get the same typed
+409 download CTA as /transcribe; a backend mid-shutdown surfaces the global
+503 ``[shutting_down]`` (ModelLoadInterruptedByShutdown → main.py handler).
+Reachability matches /generate: loopback bind by default, with the shared
+network-share PIN / API-key middleware gating any non-loopback exposure.
+"""
+from __future__ import annotations
+
+import asyncio
+import functools
+import logging
+import os
+import tempfile
+import time
+
+from fastapi import APIRouter, File, Form, HTTPException, UploadFile
+
+router = APIRouter()
+logger = logging.getLogger("omnivoice.convert")
+
+#: ffmpeg's atempo filter is well-behaved in [0.5, 2.0] per stage. Convert
+#: clamps to ONE stage by design: needing more than 2× either way means the
+#: synthesized speech differs so much from the source that "matching" it
+#: would produce chipmunk/slow-motion artifacts worse than the mismatch.
+ATEMPO_MIN = 0.5
+ATEMPO_MAX = 2.0
+
+#: Within this relative tolerance the durations already match — stretching
+#: would resample the whole take for an inaudible gain.
+_MATCH_TOLERANCE = 0.02
+
+#: Convert clips are short conversational inputs, not long-form media. Stream
+#: them to disk in bounded chunks so a network-share client cannot make the
+#: backend materialize an arbitrarily large multipart upload in memory.
+_MAX_SOURCE_AUDIO_BYTES = 64 * 1024 * 1024
+_UPLOAD_CHUNK_BYTES = 1024 * 1024
+
+
+async def _copy_source_upload(audio: UploadFile, destination) -> int:
+    """Stream ``audio`` into ``destination`` with the Convert upload cap."""
+    total = 0
+    while True:
+        chunk = await audio.read(_UPLOAD_CHUNK_BYTES)
+        if not chunk:
+            return total
+        total += len(chunk)
+        if total > _MAX_SOURCE_AUDIO_BYTES:
+            raise HTTPException(
+                status_code=413,
+                detail="Source audio is too large (maximum 64 MB).",
+            )
+        destination.write(chunk)
+
+
+def _clamped_tempo_ratio(tts_duration_s: float, source_duration_s: float) -> "float | None":
+    """The atempo ratio that fits the take into the source duration, or None.
+
+    ratio > 1 speeds the take up (it came out longer than the source),
+    ratio < 1 slows it down. Clamped to a single atempo stage's [0.5, 2.0];
+    None when either duration is unusable or they already match.
+    """
+    if not source_duration_s or source_duration_s <= 0:
+        return None
+    if not tts_duration_s or tts_duration_s <= 0:
+        return None
+    ratio = tts_duration_s / source_duration_s
+    if abs(ratio - 1.0) <= _MATCH_TOLERANCE:
+        return None
+    return min(ATEMPO_MAX, max(ATEMPO_MIN, ratio))
+
+
+async def _match_source_duration(audio_tensor, sample_rate: int, source_duration_s: float):
+    """Best-effort pitch-preserving stretch of the take toward the source
+    clip's duration. Returns the input unchanged when no stretch is needed
+    or ffmpeg fails — a duration mismatch is better than a failed convert."""
+    n_samples = int(audio_tensor.shape[-1])
+    ratio = _clamped_tempo_ratio(n_samples / sample_rate, source_duration_s)
+    if ratio is None:
+        return audio_tensor
+    target_samples = max(1, int(round(n_samples / ratio)))
+    from services.ffmpeg_utils import _pitch_preserving_stretch
+    try:
+        return await _pitch_preserving_stretch(audio_tensor, target_samples, sample_rate)
+    except Exception as e:  # noqa: BLE001 — stretch is opt-in polish, never fatal
+        logger.warning("duration match skipped — atempo stretch failed: %s", e)
+        return audio_tensor
+
+
+async def _transcribe_source(tmp_path: str, *, source_lease=None) -> dict:
+    """Active-ASR transcription of the uploaded clip (no word timestamps).
+
+    Mirrors POST /transcribe: typed 409 + download CTA before any backend
+    is constructed (never a silent multi-GB auto-download), the guarded GPU
+    pool dispatch (#730), 504 on timeout, and the same 409 when the loader
+    degrades onto an engine with no weights on disk (#1185).
+    """
+    from services.asr_backend import (
+        ASRModelMissingError,
+        ASRTimeoutError,
+        asr_model_missing_detail,
+        asr_model_missing_error,
+        run_transcribe_guarded,
+    )
+
+    missing = await asyncio.to_thread(asr_model_missing_error, purpose="transcribe")
+    if missing is not None:
+        raise HTTPException(
+            status_code=409,
+            detail={**missing, "message": asr_model_missing_detail(missing)},
+        )
+
+    def _run():
+        # `load_*`, not `get_*`: the loader runs ensure_loaded() and degrades
+        # past an engine whose deep import chain is broken (#1185).
+        from services.asr_backend import load_active_asr_backend
+        backend = load_active_asr_backend()
+        return backend.transcribe(tmp_path, word_timestamps=False)
+
+    from services.model_manager import _gpu_pool
+    release = source_lease.acquire() if source_lease is not None else None
+    abandoned = False
+    try:
+        return await run_transcribe_guarded(
+            _gpu_pool,
+            _run,
+            what="Voice convert",
+            on_abandon=release,
+        )
+    except asyncio.CancelledError:
+        # The guard now owns the lease token until the native worker drains.
+        abandoned = True
+        raise
+    except ASRTimeoutError as e:
+        abandoned = True
+        logger.warning("Convert transcription timed out: %s", e)
+        raise HTTPException(status_code=504, detail=str(e))
+    except ASRModelMissingError as e:
+        raise HTTPException(
+            status_code=409,
+            detail={**e.payload, "message": asr_model_missing_detail(e.payload)},
+        )
+    finally:
+        if release is not None and not abandoned:
+            release()
+
+
+@router.post("/convert")
+async def convert_speech(
+    audio: UploadFile = File(...),
+    profile_id: str = Form(...),
+    match_duration: bool = Form(True),
+):
+    """Convert a spoken clip into an existing voice profile's voice.
+
+    Multipart form: ``audio`` (the source clip), ``profile_id`` (an existing
+    voice profile), optional ``match_duration`` (default on — atempo the take
+    toward the source clip's length, clamped to 0.5–2.0×).
+
+    Returns JSON ``{audio_url, text, duration_s, id}`` — the take is saved to
+    OUTPUTS_DIR and served from the ``/audio`` mount like every other take.
+    """
+    from core.db import db_conn
+    from api.routers.generation import _resolve_profile_conditioning, _TempReferenceLease
+
+    # ── Profile first: strict 404, unlike /generate's silent skip — Convert
+    # has no meaning without a target voice.
+    with db_conn() as conn:
+        row = conn.execute(
+            "SELECT * FROM voice_profiles WHERE id=?", (profile_id,)
+        ).fetchone()
+    if not row:
+        raise HTTPException(
+            status_code=404,
+            detail="That voice profile doesn't exist. It may have been deleted from another tab.",
+        )
+    cond = _resolve_profile_conditioning(row)
+
+    # ── Save the upload before loading an engine. Every ASR backend (and
+    # ffprobe) needs a file path; the bounded streaming copy rejects oversized
+    # network-share requests without materializing them in process memory or
+    # starting heavyweight model work.
+    ext = os.path.splitext(audio.filename or "audio.wav")[1] or ".wav"
+    tmp = tempfile.NamedTemporaryFile(delete=False, suffix=ext)
+    source_lease = None
+    try:
+        try:
+            await _copy_source_upload(audio, tmp)
+        finally:
+            tmp.close()
+        source_lease = _TempReferenceLease(tmp.name)
+
+        # ── Engine gate before ASR/TTS work: the shared resolver refuses a
+        # clone-less engine with the actionable switch-engine message (→ 400),
+        # and a backend mid-shutdown raises ModelLoadInterruptedByShutdown out
+        # of the model load → the global 503 [shutting_down] handler.
+        from services.tts_backend import resolve_generation_backend
+        try:
+            backend = await resolve_generation_backend(
+                require_cloning=True, cloning_purpose="voice conversion",
+            )
+        except ValueError as e:
+            raise HTTPException(status_code=400, detail=str(e))
+
+        result = await _transcribe_source(tmp.name, source_lease=source_lease)
+
+        segments = result.get("segments", [])
+        text = result.get("text", "")
+        if not text and segments:
+            text = " ".join(s.get("text", "") for s in segments).strip()
+        # Same final-text hygiene as /transcribe: strip Whisper hallucination
+        # loops, then deterministic polish (leading capital + terminal
+        # punctuation) so the TTS input reads as typed text.
+        from services.refinement import collapse_repetitive_artifacts
+        from services.text_polish import polish_text
+        text = polish_text(collapse_repetitive_artifacts(text))
+        if not text or not text.strip():
+            raise HTTPException(
+                status_code=422,
+                detail=(
+                    "No speech was recognized in the source clip, so there is "
+                    "nothing to convert. Record or drop a clip with clear, "
+                    "audible speech and try again."
+                ),
+            )
+
+        # #308/#1032 parity with /generate: a clone profile saved without a
+        # transcript conditions better when its reference clip is transcribed,
+        # and that transcript is cached onto the row so it happens ONCE, not
+        # per convert. Best-effort exactly like /generate — a timeout/failure
+        # degrades to ref_text=None and the engine's own fallback. The ASR
+        # model is already warm here (the source transcribe above just used it).
+        if cond["ref_audio_path"] and not cond["ref_text"]:
+            from api.routers.generation import (
+                _generate_timeout_s,
+                _persist_profile_ref_text,
+            )
+            from services.asr_backend import transcribe_reference
+            from services.model_manager import run_on_gpu_pool_guarded
+            try:
+                cond["ref_text"] = await run_on_gpu_pool_guarded(
+                    functools.partial(transcribe_reference, cond["ref_audio_path"]),
+                    what="Reference transcribe",
+                    timeout=_generate_timeout_s(""),
+                )
+            except TimeoutError as e:
+                logger.warning(
+                    "reference transcribe hung (%s); using engine ASR fallback", e,
+                )
+                cond["ref_text"] = None
+            if cond["ref_text"] and cond["persist_ref_text"]:
+                _persist_profile_ref_text(profile_id, cond["ref_text"])
+
+        # Source duration for the optional match: the container's own length
+        # (ffprobe), falling back to the last ASR segment end. Best-effort —
+        # None just skips the stretch.
+        source_duration_s = None
+        if match_duration:
+            from services.ffmpeg_utils import probe_duration
+            source_duration_s = await probe_duration(
+                tmp.name, allowed_root=os.path.dirname(tmp.name),
+            )
+            if not source_duration_s and segments:
+                source_duration_s = max((s.get("end", 0) or 0) for s in segments) or None
+
+        # ── Same text choke point as /generate: engine-agnostic normalization
+        # (numbers→words, junk strip) on the fully resolved language.
+        from services.text_normalization import normalize_for_tts
+        language = cond["language"]
+        text = normalize_for_tts(text, language)
+
+        used_seed = cond["seed"]
+        if used_seed is None:
+            import random
+            used_seed = random.randint(0, 2**31 - 1)
+
+        from api.routers.generation import (
+            _finalize_generation,
+            _generate_timeout_s,
+            _run_backend_inference,
+        )
+        from services.model_manager import (
+            GpuJobTimeoutError,
+            GpuPoolBusyError,
+            run_on_gpu_pool_guarded,
+        )
+
+        start_time = time.time()
+        _render = functools.partial(
+            _run_backend_inference,
+            backend, text, language, cond["ref_audio_path"], cond["ref_text"],
+            cond["instruct"],
+            None,        # duration — the model picks; match_duration owns pacing
+            16, 2.0,     # num_step / guidance_scale (the /generate defaults)
+            1.0,         # speed
+            True, True,  # denoise / postprocess_output
+            used_seed,
+        )
+        try:
+            audio_tensor = await run_on_gpu_pool_guarded(
+                _render,
+                what="Voice convert",
+                timeout=_generate_timeout_s(text),
+                min_vram_gb=getattr(type(backend), "min_vram_gb", 0.0),
+            )
+        except GpuPoolBusyError as e:
+            raise HTTPException(
+                status_code=503, detail=str(e),
+                headers={"Retry-After": str(e.retry_after),
+                         "X-OmniVoice-Retryable": "true"},
+            ) from e
+        except GpuJobTimeoutError as e:
+            raise HTTPException(
+                status_code=503, detail=str(e),
+                headers={"Retry-After": "30", "X-OmniVoice-Retryable": "true"},
+            ) from e
+        except ValueError as e:
+            raise HTTPException(status_code=400, detail=str(e)) from e
+        sample_rate = backend.sample_rate
+
+        if match_duration and source_duration_s:
+            audio_tensor = await _match_source_duration(
+                audio_tensor, sample_rate, source_duration_s,
+            )
+
+        # Provenance mark AFTER the stretch (one whole-take mark on the audio
+        # the user actually keeps), then the shared finalize tail — WAV in
+        # OUTPUTS_DIR, self-healing history row, retention prune, event emit.
+        from services.watermark import mark_synthetic_async
+        audio_tensor = await mark_synthetic_async(
+            audio_tensor, sample_rate, context="convert.finalize",
+        )
+        _, meta = await _finalize_generation(
+            audio_tensor, sample_rate, text=text, history_mode="convert",
+            ref_audio_path=cond["ref_audio_path"], language=language,
+            instruct=cond["instruct"], resolved_profile_id=profile_id,
+            used_seed=used_seed, start_time=start_time,
+            already_marked=True,
+        )
+
+        return {
+            "id": meta["id"],
+            "audio_url": f"/audio/{meta['filename']}",
+            "text": text,
+            "duration_s": meta["duration"],
+            "gen_time_s": meta["gen_time"],
+        }
+    finally:
+        if source_lease is not None:
+            source_lease.finish_request()
+        else:
+            try:
+                os.unlink(tmp.name)
+            except OSError:
+                pass

backend/api/routers/workers.py

@@ -23,13 +23,16 @@ appears and is replaced by the GPU gateway.
 from __future__ import annotations
 
 import asyncio
+import contextlib
 import logging
 
 from fastapi import APIRouter, Depends, HTTPException, Request
+from fastapi.responses import JSONResponse
 from pydantic import BaseModel, Field
 
 from api.dependencies import require_admin
 from worker import registry, routing, service
+from worker.async_utils import drain_task, to_thread_and_defer_cancellation
 
 logger = logging.getLogger("omnivoice.worker")
 
@@ -158,6 +161,19 @@ def agent_status() -> dict:
     return worker_agent.agent.status()
 
 
+@router.get("/agent/readiness", include_in_schema=False, response_model=None)
+def agent_readiness() -> JSONResponse:
+    """Container readiness: 200 only after this process registered as a worker."""
+    from worker import agent as worker_agent  # noqa: PLC0415
+
+    readiness = worker_agent.agent.readiness()
+    return JSONResponse(
+        status_code=200 if readiness["ready"] else 503,
+        content=readiness,
+        headers={} if readiness["ready"] else {"Retry-After": "2"},
+    )
+
+
 def _refuse_when_env_pinned(worker_agent) -> None:
     """OMNIVOICE_WORKER_MODE wins over the setting everywhere else.
 
@@ -176,6 +192,63 @@ def _refuse_when_env_pinned(worker_agent) -> None:
         )
 
 
+async def _finish_cleanup(awaitable):
+    """Run rollback to completion even if its HTTP task was cancelled."""
+    task = asyncio.create_task(awaitable)
+    await drain_task(task)
+    return task.result()
+
+
+async def _set_worker_mode(worker_agent, enabled: bool) -> None:
+    _result, cancelled = await to_thread_and_defer_cancellation(
+        worker_agent.set_worker_mode_enabled, enabled
+    )
+    if cancelled:
+        raise asyncio.CancelledError
+
+
+async def _restore_agent_transaction(
+    worker_agent, previous: dict, *, was_running: bool
+) -> None:
+    """Restore durable enrollment/settings and the exact prior live state."""
+    try:
+        await _finish_cleanup(worker_agent.agent.stop())
+        await _finish_cleanup(worker_agent.restore_enrollment(previous))
+        if was_running and not worker_agent.agent.running:
+            await _finish_cleanup(worker_agent.agent.start())
+        elif not was_running and worker_agent.agent.running:
+            await _finish_cleanup(worker_agent.agent.stop())
+    except worker_agent.EnrollmentRollbackError:
+        raise
+    except BaseException as exc:
+        message = (
+            "The previous worker state could not be restored safely. "
+            "Worker mode remains stopped; fix its enrollment/settings storage, then retry."
+        )
+        with contextlib.suppress(BaseException):
+            await _finish_cleanup(worker_agent.agent.stop())
+        worker_agent.agent.last_error = message
+        raise worker_agent.EnrollmentRollbackError(message) from exc
+
+
+def _raise_agent_transaction_failure(
+    worker_agent, operation: BaseException, rollback: BaseException | None
+) -> None:
+    if isinstance(operation, asyncio.CancelledError):
+        if rollback is not None:
+            logger.error(
+                "Worker rollback failed during request cancellation",
+                exc_info=(type(rollback), rollback, rollback.__traceback__),
+            )
+        raise operation
+    if rollback is not None:
+        raise HTTPException(status_code=409, detail=str(rollback)) from rollback
+    if isinstance(operation, Exception):
+        worker_agent.agent.last_error = str(operation)
+        raise HTTPException(status_code=409, detail=str(operation)) from operation
+    raise operation
+
+
 @router.post("/agent/join")
 async def join_control_plane(request: JoinRequest) -> dict:
     """Redeem a join code and start working for that control plane.
@@ -200,26 +273,37 @@ async def join_control_plane(request: JoinRequest) -> dict:
     # says it joined and never lends anything (CodeRabbit).
     _refuse_when_env_pinned(worker_agent)
     async with worker_agent.agent.lifecycle:
-        # A rejoin replaces a working enrollment. Keep enough to put it back:
-        # pinning the new certificate overwrites the old one on disk, so a
-        # failed rejoin would otherwise leave the machine unable to reconnect
-        # to the control plane it was already serving.
-        previous = worker_agent.snapshot_enrollment()
-        await worker_agent.agent.stop()
         try:
+            previous, cancelled = await to_thread_and_defer_cancellation(
+                worker_agent.snapshot_enrollment
+            )
+        except worker_agent.EnrollmentStateError as exc:
+            worker_agent.agent.last_error = str(exc)
+            raise HTTPException(status_code=409, detail=str(exc)) from exc
+        if cancelled:
+            raise asyncio.CancelledError
+        was_running = worker_agent.agent.running
+
+        # A rejoin stops a working agent before the replacement is accepted.
+        # Stop, acceptance and the durable setting are one transaction: every
+        # failure, including cancellation, restores both trust and live state.
+        try:
+            await worker_agent.agent.stop()
             await worker_agent.agent.start(token_text=token)
             # Success is the control plane ACCEPTING this worker, not the
             # connection being scheduled — see wait_until_registered.
             await worker_agent.agent.wait_until_registered()
-        except Exception as exc:
-            worker_agent.agent.last_error = str(exc)
-            await worker_agent.agent.stop()
-            await worker_agent.restore_enrollment(previous)
-            raise HTTPException(status_code=409, detail=str(exc)) from exc
+            await _set_worker_mode(worker_agent, True)
+        except BaseException as exc:
+            rollback_exc = None
+            try:
+                await _restore_agent_transaction(
+                    worker_agent, previous, was_running=was_running
+                )
+            except BaseException as rollback_error:
+                rollback_exc = rollback_error
+            _raise_agent_transaction_failure(worker_agent, exc, rollback_exc)
         worker_agent.agent.last_error = ""
-        # Persisted only after the join actually worked: a machine that failed
-        # to enrol must not come back up trying again forever.
-        worker_agent.set_worker_mode_enabled(True)
         return worker_agent.agent.status()
 
 
@@ -235,19 +319,35 @@ async def set_agent_enabled(request: EnableRequest) -> dict:
 
     _refuse_when_env_pinned(worker_agent)
     async with worker_agent.agent.lifecycle:
-        if request.enabled:
-            try:
+        try:
+            previous, cancelled = await to_thread_and_defer_cancellation(
+                worker_agent.snapshot_enrollment
+            )
+        except worker_agent.EnrollmentStateError as exc:
+            worker_agent.agent.last_error = str(exc)
+            raise HTTPException(status_code=409, detail=str(exc)) from exc
+        if cancelled:
+            raise asyncio.CancelledError
+        was_running = worker_agent.agent.running
+
+        try:
+            if request.enabled:
                 await worker_agent.agent.start()
                 await worker_agent.agent.wait_until_registered()
-            except Exception as exc:
-                worker_agent.agent.last_error = str(exc)
+                await _set_worker_mode(worker_agent, True)
+            else:
                 await worker_agent.agent.stop()
-                raise HTTPException(status_code=409, detail=str(exc)) from exc
-            worker_agent.agent.last_error = ""
-            worker_agent.set_worker_mode_enabled(True)
-        else:
-            await worker_agent.agent.stop()
-            worker_agent.set_worker_mode_enabled(False)
+                await _set_worker_mode(worker_agent, False)
+        except BaseException as exc:
+            rollback_exc = None
+            try:
+                await _restore_agent_transaction(
+                    worker_agent, previous, was_running=was_running
+                )
+            except BaseException as rollback_error:
+                rollback_exc = rollback_error
+            _raise_agent_transaction_failure(worker_agent, exc, rollback_exc)
+        worker_agent.agent.last_error = ""
         return worker_agent.agent.status()
 
 
@@ -263,9 +363,14 @@ def create_enrollment(request: EnrollRequest) -> dict:
             status_code=409,
             detail="Remote workers are turned off. Enable them in Settings → System → Remote workers first.",
         )
-    token = service.control_plane.create_enrollment(
-        endpoint=request.endpoint, label=request.label, ttl_seconds=request.ttl_seconds
-    )
+    try:
+        token = service.control_plane.create_enrollment(
+            endpoint=request.endpoint,
+            label=request.label,
+            ttl_seconds=request.ttl_seconds,
+        )
+    except service.EndpointCertificateError as exc:
+        raise HTTPException(status_code=409, detail=str(exc)) from exc
     return {
         "token": token.encode(),
         "endpoint": token.endpoint,
@@ -275,23 +380,55 @@ def create_enrollment(request: EnrollRequest) -> dict:
     }
 
 
+def _persist_worker_update(
+    worker_id: str, request: WorkerUpdate
+):
+    """Write policy on a worker thread; live publication stays loop-owned."""
+    return registry.update_policy(
+        worker_id,
+        name=request.name,
+        enabled=request.enabled,
+        priority=request.priority,
+    )
+
+
 @router.patch("/{worker_id}")
-def update_worker(worker_id: str, request: WorkerUpdate) -> dict:
-    worker = registry.get(worker_id)
-    if worker is None:
+async def update_worker(worker_id: str, request: WorkerUpdate) -> dict:
+    pool = service.control_plane.pool if service.control_plane.running else None
+    live = None
+    was_pending = False
+    if pool is not None:
+        # Quiesce dispatch before releasing authority for the SQLite write.
+        # The publication after the await restores the exact prior state, so a
+        # concurrent registration handoff remains quiesced for its own reason.
+        with registry.authority_guard():
+            live = pool.get(worker_id)
+            if live is not None:
+                was_pending = live.registration_pending
+                live.registration_pending = True
+    updated = None
+    cancelled = False
+    try:
+        updated, cancelled = await to_thread_and_defer_cancellation(
+            _persist_worker_update, worker_id, request
+        )
+    finally:
+        if pool is not None:
+            with registry.authority_guard():
+                if updated is not None:
+                    # Pool state, including the cached record the scheduler
+                    # reads, belongs to the app's event loop.
+                    pool.refresh_record(updated)
+                current = pool.get(worker_id)
+                if current is live:
+                    current.registration_pending = was_pending
+    if updated is None:
+        if cancelled:
+            raise asyncio.CancelledError
         raise HTTPException(status_code=404, detail="No such worker.")
-    if request.name is not None:
-        registry.rename(worker_id, request.name)
-    if request.enabled is not None:
-        registry.set_enabled(worker_id, request.enabled)
-    if request.priority is not None:
-        registry.set_priority(worker_id, request.priority)
-    updated = registry.get(worker_id)
-    # Keep the live copy in step, so the scheduler and its logs do not go on
-    # using the name or priority this worker had when it connected.
-    if updated is not None and service.control_plane.running:
-        service.control_plane.pool.refresh_record(updated)
-    return updated.to_dict() if updated else {}
+    if cancelled:
+        raise asyncio.CancelledError
+    return updated.to_dict()
 
 
 @router.post("/{worker_id}/consent")
@@ -305,7 +442,7 @@ def grant_consent(worker_id: str) -> dict:
 
 
 @router.post("/{worker_id}/resume")
-def clear_breaker(worker_id: str) -> dict:
+async def clear_breaker(worker_id: str) -> dict:
     """Clear a paused worker's circuit breakers.
 
     The user fixed the machine and knows it — a breaker with no manual clear is
@@ -320,18 +457,53 @@ def clear_breaker(worker_id: str) -> dict:
 
 
 @router.delete("/{worker_id}")
-def revoke_worker(worker_id: str) -> dict:
+async def revoke_worker(worker_id: str) -> dict:
     """Remove a worker — which means revoke its key, not hide the row.
 
     Its in-flight work is released so it can be retried elsewhere rather than
     waiting out a lease on a machine that will never answer again.
     """
-    if registry.get(worker_id) is None:
+    pool = service.control_plane.pool if service.control_plane.running else None
+    live = None
+    was_pending = False
+    if pool is not None:
+        with registry.authority_guard():
+            live = pool.get(worker_id)
+            if live is not None:
+                was_pending = live.registration_pending
+                live.registration_pending = True
+    try:
+        revoked, cancelled = await to_thread_and_defer_cancellation(
+            registry.revoke, worker_id
+        )
+    except BaseException:
+        if pool is not None:
+            with registry.authority_guard():
+                current = pool.get(worker_id)
+                if current is live:
+                    current.registration_pending = was_pending
+        raise
+    if not revoked:
+        if pool is not None:
+            with registry.authority_guard():
+                current = pool.get(worker_id)
+                if current is live:
+                    current.registration_pending = was_pending
+        if cancelled:
+            raise asyncio.CancelledError
         raise HTTPException(status_code=404, detail="No such worker.")
-    registry.revoke(worker_id)
-    if service.control_plane.running:
-        service.control_plane.scheduler.on_disconnected(worker_id)
-        service.control_plane.pool.breakers.forget_worker(worker_id)
+
+    # The tombstone committed before any egress/session mutation. Everything
+    # below is loop-owned and published under the same scheduler authority read
+    # used by next_assignment(), so no task can bind in the handoff window.
+    with registry.authority_guard():
+        if service.control_plane.running:
+            if service.control_plane.servicer is not None:
+                service.control_plane.servicer.revoke_worker_sessions(worker_id)
+            service.control_plane.scheduler.on_disconnected(worker_id)
+            service.control_plane.pool.breakers.forget_worker(worker_id)
+    if cancelled:
+        raise asyncio.CancelledError
     return {"ok": True, "revoked": worker_id}
 
 
@@ -375,7 +547,9 @@ async def submit_task(request: Request, body: SubmitTaskRequest) -> dict:
 
     scheduler = service.control_plane.scheduler
     try:
-        task = scheduler.submit(
+        submit = getattr(scheduler, "submit_async", None)
+        submit = submit if callable(submit) else scheduler.submit
+        submitted = submit(
             operation=body.operation,
             engine=body.engine,
             model_id=body.model_id,
@@ -384,6 +558,7 @@ async def submit_task(request: Request, body: SubmitTaskRequest) -> dict:
             deadline_seconds=body.deadline_seconds,
             pinned_worker_id=routing.decide().worker_id or None,
         )
+        task = await submitted if asyncio.iscoroutine(submitted) else submitted
     except QueueFull as exc:
         raise HTTPException(status_code=429, detail=str(exc)) from exc
 
@@ -505,8 +680,33 @@ async def set_inbound_enabled(request: InboundEnableRequest) -> dict:
                 "machine. Change that environment setting and restart VoiceStudio."
             ),
         )
+
+    requested_bind = (
+        inbound_service.normalise_bind_host(request.bind)
+        if request.bind
+        else inbound_service.bind_host()
+    )
+    requested_port = request.port or inbound_service.bind_port()
+    if (
+        request.enabled
+        and inbound_service.node.running
+        and (
+            requested_bind != inbound_service.bind_host()
+            or requested_port != inbound_service.node.port
+        )
+    ):
+        # start() is intentionally idempotent while a listener owns its
+        # socket. Persisting a new endpoint here would make the UI report a
+        # narrower/different bind while the original socket stayed live.
+        raise HTTPException(
+            status_code=409,
+            detail=(
+                "Turn off Accept connections before changing its bind address "
+                "or port."
+            ),
+        )
     if request.bind:
-        inbound_service.set_bind_host(request.bind)
+        inbound_service.set_bind_host(requested_bind)
     if request.port:
         inbound_service.set_bind_port(request.port)
     inbound_service.set_enabled(request.enabled)
@@ -535,6 +735,7 @@ def issue_inbound_key(request: IssueKeyRequest) -> dict:
     is stored, so it cannot be shown again, only replaced.
     """
     from worker.inbound import service as inbound_service  # noqa: PLC0415
+    from worker.inbound.keys import KeyLimitExceeded  # noqa: PLC0415
 
     if not inbound_service.node.running:
         raise HTTPException(
@@ -544,7 +745,10 @@ def issue_inbound_key(request: IssueKeyRequest) -> dict:
                 "Settings → System → Remote workers → Accept connections first."
             ),
         )
-    issued = inbound_service.node.keys.issue(request.label)
+    try:
+        issued = inbound_service.node.keys.issue(request.label)
+    except KeyLimitExceeded as exc:
+        raise HTTPException(status_code=409, detail=str(exc)) from exc
     return {
         "key_id": issued.key.key_id,
         "label": issued.key.label,
@@ -555,12 +759,12 @@ def issue_inbound_key(request: IssueKeyRequest) -> dict:
 
 
 @router.delete("/inbound/keys/{key_id}")
-def revoke_inbound_key(key_id: str) -> dict:
+async def revoke_inbound_key(key_id: str) -> dict:
     """Revoke one panel. Everyone else stays connected — the whole reason keys
     are per panel rather than one shared node key."""
     from worker.inbound import service as inbound_service  # noqa: PLC0415
 
-    if not inbound_service.node.keys.revoke(key_id):
+    if not await inbound_service.node.revoke_key(key_id):
         raise HTTPException(status_code=404, detail="No such key.")
     return inbound_service.node.snapshot()
 
@@ -579,6 +783,7 @@ async def add_inbound_connection(request: ConnectRequest) -> dict:
     """Paste a connection string from a GPU machine and dial it."""
     from worker.inbound import service as inbound_service  # noqa: PLC0415
     from worker.inbound.connection_string import InvalidConnectionString  # noqa: PLC0415
+    from worker.inbound.connector import InboundConnectionError  # noqa: PLC0415
 
     if not service.control_plane.running:
         raise HTTPException(
@@ -597,12 +802,18 @@ async def add_inbound_connection(request: ConnectRequest) -> dict:
         # surfaces as "cannot connect", which is what a firewall, a wrong port
         # and a dead node all say too.
         raise HTTPException(status_code=400, detail=str(exc)) from exc
+    except InboundConnectionError as exc:
+        raise HTTPException(status_code=409, detail=str(exc)) from exc
     return {"endpoint": connection.endpoint, "connections": inbound_service.outbound.snapshot()}
 
 
 @router.delete("/inbound/connections/{endpoint}")
 async def remove_inbound_connection(endpoint: str) -> dict:
     from worker.inbound import service as inbound_service  # noqa: PLC0415
+    from worker.inbound.connector import InboundConnectionError  # noqa: PLC0415
 
-    await inbound_service.outbound.remove(endpoint)
+    try:
+        await inbound_service.outbound.remove(endpoint)
+    except InboundConnectionError as exc:
+        raise HTTPException(status_code=409, detail=str(exc)) from exc
     return {"connections": inbound_service.outbound.snapshot()}

backend/api/schemas.py

@@ -26,6 +26,21 @@ class SystemInfoResponse(BaseModel):
     model_config = ConfigDict(extra="allow")
 
     app_version: str = ""
+    # Effective compute-time budgets (seconds) for one synthesis job — the
+    # values services/model_manager.py's GPU_JOB_TIMEOUT_S / CPU_JOB_TIMEOUT_S
+    # captured at backend import time (#1787). A value just saved via
+    # /system/set-env is NOT reflected here until the next restart.
+    generate_timeout_s: float = 300.0
+    cpu_generate_timeout_s: float = 600.0
+    # True when an external env var (shell, `.env`, Docker, …) is currently
+    # shadowing a prefs.json save for this key — see core.prefs.is_env_shadowed.
+    generate_timeout_shadowed: bool = False
+    cpu_generate_timeout_shadowed: bool = False
+    # #1770: the desktop attach handshake's code fingerprint — whatever
+    # Tauri set OMNIVOICE_BUILD_FINGERPRINT to when it spawned this process,
+    # echoed back verbatim. Blank when unset (dev mode, a manually started
+    # backend). See frontend/src-tauri/src/backend.rs::code_fingerprint_is_current.
+    code_fingerprint: str = ""
     data_dir: str
     outputs_dir: str
     crash_log_path: str

backend/core/contained_subprocess.py

@@ -0,0 +1,716 @@
+"""Nested subprocess ownership for desktop-managed backend operations.
+
+The desktop owns the backend with an OS process group/Job.  Engine and
+installer operations also need an independently terminable subtree: killing
+only their direct child on a timeout leaves uv/git/model workers holding pipes
+and mutating files.
+
+On POSIX a small supervisor is the unreaped leader of a nested process group.
+A control-pipe EOF (including kernel EOF when the backend dies) kills that
+group; the parent also drains the group before reaping its stable leader. On
+Windows the backend retains a nested kill-on-close Job directly and assigns
+the suspended operation before resuming it. The outer desktop Job remains the
+terminal fallback.
+
+Standalone/server launches use the same nested owner, preserving their
+independently terminable subtree without relying on ``taskkill`` or discovery.
+"""
+from __future__ import annotations
+
+import os
+import signal
+import struct
+import subprocess
+import sys
+import threading
+import time
+from pathlib import Path
+from typing import Any, Optional
+
+
+_RESULT = struct.Struct("!i")
+_DESKTOP_MARKER = "OMNIVOICE_DESKTOP_CONTAINED"
+_DRAIN_FD_ENV = "OMNIVOICE_DESKTOP_DRAIN_FD"
+
+
+def backend_drain_fd(*, required: bool = False) -> Optional[int]:
+    """Validated Rust-owned drain writer inherited by the desktop backend."""
+    if os.name != "posix" or os.environ.get(_DESKTOP_MARKER) != "1":
+        return None
+    try:
+        fd = int(os.environ[_DRAIN_FD_ENV])
+        os.fstat(fd)
+    except (KeyError, ValueError, OSError) as exc:
+        if required:
+            raise RuntimeError(
+                "desktop backend is missing its live nested-operation drain descriptor"
+            ) from exc
+        return None
+    return fd
+
+
+def secure_backend_drain_fd() -> None:
+    """Restore CLOEXEC after Rust's one intentional backend inheritance."""
+    fd = backend_drain_fd(required=True)
+    if fd is not None:
+        os.set_inheritable(fd, False)
+
+
+class OwnedPopen:
+    """Popen-compatible handle for a desktop-owned nested operation."""
+
+    def __init__(
+        self,
+        proc: subprocess.Popen,
+        control_fd: int,
+        result_fd: int,
+    ) -> None:
+        self._proc = proc
+        self._control_fd: Optional[int] = control_fd
+        self._result_fd: Optional[int] = result_fd
+        self._returncode: Optional[int] = None
+        self._lock = threading.RLock()
+
+        # Popen callers use these directly (protocol pipes and log drains).
+        self.stdin = proc.stdin
+        self.stdout = proc.stdout
+        self.stderr = proc.stderr
+
+    @property
+    def pid(self) -> int:
+        return self._proc.pid
+
+    @property
+    def args(self) -> Any:
+        return self._proc.args
+
+    @property
+    def returncode(self) -> Optional[int]:
+        return self._returncode
+
+    def _close_control(self) -> None:
+        fd, self._control_fd = self._control_fd, None
+        if fd is not None:
+            try:
+                os.close(fd)
+            except OSError:
+                # Cleanup is idempotent; another teardown path already closed it.
+                pass
+
+    def _read_result(self, fallback: int) -> int:
+        fd, self._result_fd = self._result_fd, None
+        if fd is None:
+            return fallback
+        try:
+            payload = b""
+            while len(payload) < _RESULT.size:
+                chunk = os.read(fd, _RESULT.size - len(payload))
+                if not chunk:
+                    break
+                payload += chunk
+            return _RESULT.unpack(payload)[0] if len(payload) == _RESULT.size else fallback
+        except OSError:
+            return fallback
+        finally:
+            try:
+                os.close(fd)
+            except OSError:
+                # The descriptor may have been closed by cancellation cleanup.
+                pass
+
+    def _posix_exited_unreaped(self) -> bool:
+        flags = os.WEXITED | os.WNOHANG | os.WNOWAIT
+        info = os.waitid(os.P_PID, self.pid, flags)
+        return info is not None and info.si_pid != 0
+
+    def _posix_exited_reaping(self) -> Optional[int]:
+        """macOS fallback for :meth:`_posix_exited_unreaped` (#1656).
+
+        CPython on macOS does not expose ``os.waitid`` (HAVE_WAITID is not set
+        in its build), so the WNOWAIT probe is unavailable there. This
+        fallback *reaps* the wrapper with ``waitpid(WNOHANG)``: it returns
+        the wrapper's exit code once it has exited, None while it is still
+        running, and raises ``ChildProcessError`` when another owner already
+        reaped it (the same refusal the waitid probe gives).
+
+        Reaping earlier than the WNOWAIT dance loses the pre-reap group kill
+        in :meth:`poll`; that is safe because the supervisor's control-pipe
+        EOF already terminates the whole nested group (#1635 design).
+        """
+        pid, status = os.waitpid(self.pid, os.WNOHANG)
+        if pid != self.pid:
+            return None
+        rc = os.waitstatus_to_exitcode(status)
+        # Publish on the underlying Popen so its own wait()/poll() no-op.
+        self._proc.returncode = rc
+        return rc
+
+    def _posix_exit_state_reaping(self) -> Optional[int]:
+        """:meth:`_posix_exited_reaping` plus one concession: if the leader
+        was already reaped through *this* Popen (``_proc.returncode`` known),
+        report that code rather than refusing — reaping by our own handle is
+        not the foreign reaper the ECHILD refusal exists for."""
+        try:
+            return self._posix_exited_reaping()
+        except ChildProcessError:
+            return self._proc.returncode
+
+    def _signal_owned_group(self, sig: int) -> None:
+        # The numeric group is safe only while its direct-child leader remains
+        # ours and unreaped.  ECHILD therefore refuses rather than guessing.
+        try:
+            os.waitid(os.P_PID, self.pid, os.WEXITED | os.WNOHANG | os.WNOWAIT)
+        except ChildProcessError:
+            return
+        except AttributeError:
+            # macOS CPython has no os.waitid (#1656). waitpid still proves
+            # that this exact numeric pid is our live child: ECHILD refuses a
+            # foreign-reaped/reused pid, while pid == self.pid records an exit
+            # without ever signalling the now-unowned process-group number.
+            try:
+                pid, status = os.waitpid(self.pid, os.WNOHANG)
+            except ChildProcessError:
+                return
+            if pid == self.pid:
+                self._proc.returncode = os.waitstatus_to_exitcode(status)
+                return
+        try:
+            os.killpg(self.pid, sig)
+        except ProcessLookupError:
+            # The owned group exited between the waitid probe and the signal.
+            pass
+
+    def poll(self) -> Optional[int]:
+        with self._lock:
+            if self._returncode is not None:
+                return self._returncode
+            if os.name == "posix":
+                try:
+                    if hasattr(os, "waitid"):
+                        if not self._posix_exited_unreaped():
+                            return None
+                        self._signal_owned_group(signal.SIGKILL)
+                        wrapper_rc = self._proc.wait()
+                    else:
+                        # macOS CPython: no os.waitid (#1656) — the reaping
+                        # probe already terminated/killed nothing; the group
+                        # is torn down by the control-pipe EOF in _close_control.
+                        wrapper_rc = self._posix_exit_state_reaping()
+                        if wrapper_rc is None:
+                            return None
+                except ChildProcessError:
+                    # Never signal a potentially reused group after another
+                    # owner reaped the stable leader.
+                    return None
+            else:
+                wrapper_rc = self._proc.poll()
+                if wrapper_rc is None:
+                    return None
+            self._close_control()
+            self._returncode = self._read_result(wrapper_rc)
+            return self._returncode
+
+    def wait(self, timeout: Optional[float] = None) -> int:
+        deadline = None if timeout is None else time.monotonic() + timeout
+        while True:
+            rc = self.poll()
+            if rc is not None:
+                return rc
+            if deadline is not None and time.monotonic() >= deadline:
+                raise subprocess.TimeoutExpired(self.args, timeout)
+            time.sleep(0.01)
+
+    def terminate(self) -> None:
+        with self._lock:
+            if self._returncode is not None:
+                return
+            self._close_control()
+            if os.name == "posix":
+                self._signal_owned_group(signal.SIGTERM)
+            else:
+                # Closing the control pipe asks the supervisor to terminate
+                # its nested Job.  The stable wrapper handle is a fallback.
+                try:
+                    self._proc.terminate()
+                except OSError:
+                    # The wrapper exited after the return-code check.
+                    pass
+
+    def kill(self) -> None:
+        with self._lock:
+            if self._returncode is not None:
+                return
+            self._close_control()
+            if os.name == "posix":
+                self._signal_owned_group(signal.SIGKILL)
+            else:
+                try:
+                    self._proc.kill()
+                except OSError:
+                    # The wrapper exited after the return-code check.
+                    pass
+
+    def __getattr__(self, name: str) -> Any:
+        return getattr(self._proc, name)
+
+    def __del__(self) -> None:
+        self._close_control()
+        fd, self._result_fd = self._result_fd, None
+        if fd is not None:
+            try:
+                os.close(fd)
+            except OSError:
+                # Finalization may race explicit wait or cancellation cleanup.
+                pass
+
+
+class WindowsJobPopen:
+    """Popen-compatible handle whose child tree lives in a retained Job.
+
+    Windows Job handles already provide the stable ownership that POSIX needs
+    a supervisor process group for. Keeping the handle in the backend means an
+    abrupt backend exit closes it in the kernel and kills the whole operation
+    tree, without inserting a second Python process in the sidecar loader path
+    (#1734).
+    """
+
+    def __init__(self, proc: subprocess.Popen, job: Any, kernel32: Any) -> None:
+        self._proc = proc
+        self._job = job
+        self._kernel32 = kernel32
+        self._lock = threading.RLock()
+        self.stdin = proc.stdin
+        self.stdout = proc.stdout
+        self.stderr = proc.stderr
+
+    @property
+    def pid(self) -> int:
+        return self._proc.pid
+
+    @property
+    def args(self) -> Any:
+        return self._proc.args
+
+    @property
+    def returncode(self) -> Optional[int]:
+        return self._proc.returncode
+
+    def _close_job(self, *, terminate: bool) -> None:
+        job, self._job = self._job, None
+        if job is None:
+            return
+        try:
+            if terminate:
+                self._kernel32.TerminateJobObject(job, 1)
+        finally:
+            self._kernel32.CloseHandle(job)
+
+    def poll(self) -> Optional[int]:
+        with self._lock:
+            rc = self._proc.poll()
+            if rc is None:
+                return None
+            # A successful direct child may leave helpers behind. Match the
+            # supervisor contract by draining the retained Job before return.
+            self._close_job(terminate=True)
+            return rc
+
+    def wait(self, timeout: Optional[float] = None) -> int:
+        try:
+            rc = self._proc.wait(timeout=timeout)
+        except subprocess.TimeoutExpired:
+            raise
+        with self._lock:
+            self._close_job(terminate=True)
+        return rc
+
+    def terminate(self) -> None:
+        with self._lock:
+            self._close_job(terminate=True)
+
+    def kill(self) -> None:
+        self.terminate()
+
+    def __getattr__(self, name: str) -> Any:
+        return getattr(self._proc, name)
+
+    def __del__(self) -> None:
+        try:
+            self._close_job(terminate=True)
+        except Exception:
+            pass  # interpreter shutdown; closing the OS handle is best-effort
+
+
+def _spawn_windows_owned(argv: list[str], kwargs: dict[str, Any]) -> WindowsJobPopen:
+    """Start *argv* suspended, assign its tree to a Job, then resume it."""
+    import ctypes
+
+    job, kernel32, wintypes = _windows_job()
+    child: Optional[subprocess.Popen] = None
+    popen_kwargs = dict(kwargs)
+    supplied_env = popen_kwargs.get("env")
+    operation_env = dict(os.environ if supplied_env is None else supplied_env)
+    operation_env.pop(_DRAIN_FD_ENV, None)
+    operation_env.pop(_DESKTOP_MARKER, None)
+    popen_kwargs["env"] = operation_env
+    supplied_flags = int(popen_kwargs.pop("creationflags", 0))
+    popen_kwargs["creationflags"] = supplied_flags | 0x08000000 | 0x00000004
+    try:
+        child = subprocess.Popen(argv, **popen_kwargs)
+        assign = kernel32.AssignProcessToJobObject
+        assign.argtypes = (wintypes.HANDLE, wintypes.HANDLE)
+        assign.restype = wintypes.BOOL
+        if not assign(job, wintypes.HANDLE(child._handle)):
+            raise OSError(ctypes.get_last_error(), "AssignProcessToJobObject")
+        _resume_windows_process(kernel32, wintypes, child.pid)
+        return WindowsJobPopen(child, job, kernel32)
+    except BaseException:
+        kernel32.TerminateJobObject(job, 1)
+        if child is not None:
+            try:
+                child.kill()
+            except OSError:
+                pass  # the suspended child may already have exited
+            try:
+                child.wait(timeout=5)
+            except (OSError, subprocess.TimeoutExpired):
+                pass  # Job termination remains the authoritative cleanup
+        kernel32.CloseHandle(job)
+        raise
+
+
+def spawn_owned(
+    argv: list[str], **kwargs: Any
+) -> "subprocess.Popen | OwnedPopen | WindowsJobPopen":
+    """Spawn an operation with a stable, independently terminable owner."""
+
+    if os.name == "nt":
+        return _spawn_windows_owned(argv, kwargs)
+
+    drain_fd = backend_drain_fd(required=True)
+    control_read, control_write = os.pipe()
+    result_read, result_write = os.pipe()
+    wrapper_argv = _supervisor_argv(
+        control_read,
+        result_write,
+        argv,
+    )
+    wrapper_kwargs = dict(kwargs)
+    wrapper_kwargs["start_new_session"] = True
+    pass_fds = [control_read, result_write]
+    if drain_fd is not None:
+        pass_fds.append(drain_fd)
+        if wrapper_kwargs.get("env") is not None:
+            wrapper_env = dict(wrapper_kwargs["env"])
+            wrapper_env[_DESKTOP_MARKER] = "1"
+            wrapper_env[_DRAIN_FD_ENV] = str(drain_fd)
+            wrapper_kwargs["env"] = wrapper_env
+    wrapper_kwargs["pass_fds"] = tuple(pass_fds)
+    try:
+        proc = subprocess.Popen(wrapper_argv, **wrapper_kwargs)
+    except BaseException:
+        # The finally block exclusively owns the child-side endpoints. Closing
+        # them here as well risks closing a reused descriptor in another thread.
+        for fd in (control_write, result_read):
+            try:
+                os.close(fd)
+            except OSError:
+                # A partial spawn may already have closed a parent-side endpoint.
+                pass
+        raise
+    finally:
+        for fd in (control_read, result_write):
+            try:
+                os.close(fd)
+            except OSError:
+                # Popen may have consumed an inherited child-side endpoint.
+                pass
+    return OwnedPopen(proc, control_write, result_read)
+
+
+def _supervisor_argv(
+    control_token: int,
+    result_token: int,
+    argv: list[str],
+) -> list[str]:
+    prefix = [sys.executable]
+    if not getattr(sys, "frozen", False):
+        prefix.append(str(Path(__file__).resolve().parents[1] / "main.py"))
+    return [
+        *prefix,
+        "--supervise",
+        str(control_token),
+        str(result_token),
+        "--",
+        *map(str, argv),
+    ]
+
+
+def _write_result(fd: int, returncode: int) -> None:
+    try:
+        os.write(fd, _RESULT.pack(int(returncode)))
+    except OSError:
+        # The caller may have cancelled and closed its result reader.
+        pass
+    finally:
+        try:
+            os.close(fd)
+        except OSError:
+            # Writing or cancellation may already have closed the descriptor.
+            pass
+
+
+def _operation_env() -> dict[str, str]:
+    env = os.environ.copy()
+    # The operation intentionally does not own the Rust drain writer. Avoid
+    # exposing a stale numeric token which nested code could mistake as valid.
+    env.pop(_DRAIN_FD_ENV, None)
+    env.pop(_DESKTOP_MARKER, None)
+    return env
+
+
+def _supervise_posix(control_fd: int, result_fd: int, argv: list[str]) -> int:
+    def cancel_on_eof() -> None:
+        try:
+            while os.read(control_fd, 1):
+                pass
+        except OSError:
+            # Closing the control descriptor is itself a cancellation signal.
+            pass
+        os.killpg(os.getpgrp(), signal.SIGKILL)
+
+    threading.Thread(target=cancel_on_eof, daemon=True).start()
+    try:
+        child = subprocess.Popen(argv, close_fds=True, env=_operation_env())
+        rc = child.wait()
+    except OSError:
+        rc = 127
+    _write_result(result_fd, rc)
+    # Drain children which outlived the operation before the stable group
+    # leader exits.  SIGKILL intentionally includes this supervisor.
+    os.killpg(os.getpgrp(), signal.SIGKILL)
+    return rc  # unreachable
+
+
+def _windows_job() -> tuple[Any, Any, Any]:
+    import ctypes
+    import ctypes.wintypes as wintypes
+
+    kernel32 = ctypes.WinDLL("kernel32", use_last_error=True)
+    kernel32.CloseHandle.argtypes = (wintypes.HANDLE,)
+    kernel32.CloseHandle.restype = wintypes.BOOL
+    kernel32.TerminateJobObject.argtypes = (wintypes.HANDLE, wintypes.UINT)
+    kernel32.TerminateJobObject.restype = wintypes.BOOL
+    kernel32.ReadFile.argtypes = (
+        wintypes.HANDLE,
+        ctypes.c_void_p,
+        wintypes.DWORD,
+        ctypes.POINTER(wintypes.DWORD),
+        ctypes.c_void_p,
+    )
+    kernel32.ReadFile.restype = wintypes.BOOL
+    kernel32.WriteFile.argtypes = (
+        wintypes.HANDLE,
+        ctypes.c_void_p,
+        wintypes.DWORD,
+        ctypes.POINTER(wintypes.DWORD),
+        ctypes.c_void_p,
+    )
+    kernel32.WriteFile.restype = wintypes.BOOL
+    create = kernel32.CreateJobObjectW
+    create.argtypes = (ctypes.c_void_p, wintypes.LPCWSTR)
+    create.restype = wintypes.HANDLE
+    job = create(None, None)
+    if not job:
+        raise OSError(ctypes.get_last_error(), "CreateJobObjectW")
+
+    class BasicLimits(ctypes.Structure):
+        _fields_ = [
+            ("PerProcessUserTimeLimit", ctypes.c_longlong),
+            ("PerJobUserTimeLimit", ctypes.c_longlong),
+            ("LimitFlags", wintypes.DWORD),
+            ("MinimumWorkingSetSize", ctypes.c_size_t),
+            ("MaximumWorkingSetSize", ctypes.c_size_t),
+            ("ActiveProcessLimit", wintypes.DWORD),
+            ("Affinity", ctypes.c_size_t),
+            ("PriorityClass", wintypes.DWORD),
+            ("SchedulingClass", wintypes.DWORD),
+        ]
+
+    class IoCounters(ctypes.Structure):
+        _fields_ = [(name, ctypes.c_ulonglong) for name in (
+            "ReadOperationCount", "WriteOperationCount", "OtherOperationCount",
+            "ReadTransferCount", "WriteTransferCount", "OtherTransferCount",
+        )]
+
+    class ExtendedLimits(ctypes.Structure):
+        _fields_ = [
+            ("BasicLimitInformation", BasicLimits),
+            ("IoInfo", IoCounters),
+            ("ProcessMemoryLimit", ctypes.c_size_t),
+            ("JobMemoryLimit", ctypes.c_size_t),
+            ("PeakProcessMemoryUsed", ctypes.c_size_t),
+            ("PeakJobMemoryUsed", ctypes.c_size_t),
+        ]
+
+    info = ExtendedLimits()
+    info.BasicLimitInformation.LimitFlags = 0x00002000  # KILL_ON_JOB_CLOSE
+    set_info = kernel32.SetInformationJobObject
+    set_info.argtypes = (wintypes.HANDLE, ctypes.c_int, ctypes.c_void_p, wintypes.DWORD)
+    set_info.restype = wintypes.BOOL
+    if not set_info(job, 9, ctypes.byref(info), ctypes.sizeof(info)):
+        error = ctypes.get_last_error()
+        kernel32.CloseHandle(job)
+        raise OSError(error, "SetInformationJobObject")
+    return job, kernel32, wintypes
+
+
+def _resume_windows_process(kernel32: Any, wintypes: Any, pid: int) -> None:
+    import ctypes
+
+    class ThreadEntry(ctypes.Structure):
+        _fields_ = [
+            ("dwSize", wintypes.DWORD),
+            ("cntUsage", wintypes.DWORD),
+            ("th32ThreadID", wintypes.DWORD),
+            ("th32OwnerProcessID", wintypes.DWORD),
+            ("tpBasePri", wintypes.LONG),
+            ("tpDeltaPri", wintypes.LONG),
+            ("dwFlags", wintypes.DWORD),
+        ]
+
+    kernel32.CreateToolhelp32Snapshot.argtypes = (wintypes.DWORD, wintypes.DWORD)
+    kernel32.CreateToolhelp32Snapshot.restype = wintypes.HANDLE
+    kernel32.Thread32First.argtypes = (wintypes.HANDLE, ctypes.POINTER(ThreadEntry))
+    kernel32.Thread32First.restype = wintypes.BOOL
+    kernel32.Thread32Next.argtypes = (wintypes.HANDLE, ctypes.POINTER(ThreadEntry))
+    kernel32.Thread32Next.restype = wintypes.BOOL
+    kernel32.OpenThread.argtypes = (wintypes.DWORD, wintypes.BOOL, wintypes.DWORD)
+    kernel32.OpenThread.restype = wintypes.HANDLE
+    kernel32.ResumeThread.argtypes = (wintypes.HANDLE,)
+    kernel32.ResumeThread.restype = wintypes.DWORD
+
+    snapshot = kernel32.CreateToolhelp32Snapshot(0x00000004, 0)
+    invalid = ctypes.c_void_p(-1).value
+    if snapshot == invalid:
+        raise OSError(ctypes.get_last_error(), "CreateToolhelp32Snapshot")
+    try:
+        entry = ThreadEntry(dwSize=ctypes.sizeof(ThreadEntry))
+        found = kernel32.Thread32First(snapshot, ctypes.byref(entry))
+        while found:
+            if entry.th32OwnerProcessID == pid:
+                thread = kernel32.OpenThread(0x0002, False, entry.th32ThreadID)
+                if not thread:
+                    raise OSError(ctypes.get_last_error(), "OpenThread")
+                try:
+                    if kernel32.ResumeThread(thread) == 0xFFFFFFFF:
+                        raise OSError(ctypes.get_last_error(), "ResumeThread")
+                    return
+                finally:
+                    kernel32.CloseHandle(thread)
+            found = kernel32.Thread32Next(snapshot, ctypes.byref(entry))
+    finally:
+        kernel32.CloseHandle(snapshot)
+    raise OSError("suspended operation thread was not found")
+
+
+def _supervise_windows(control_fd: int, result_fd: int, argv: list[str]) -> int:
+    import ctypes
+
+    job, kernel32, wintypes = _windows_job()
+    cancelled = threading.Event()
+    job_lock = threading.Lock()
+    job_open = True
+
+    def terminate_job() -> None:
+        with job_lock:
+            if job_open:
+                kernel32.TerminateJobObject(job, 1)
+
+    def cancel_on_eof() -> None:
+        byte = ctypes.create_string_buffer(1)
+        count = wintypes.DWORD()
+        while kernel32.ReadFile(
+            wintypes.HANDLE(control_fd), byte, 1, ctypes.byref(count), None
+        ) and count.value:
+            pass
+        kernel32.CloseHandle(wintypes.HANDLE(control_fd))
+        cancelled.set()
+        terminate_job()
+
+    threading.Thread(target=cancel_on_eof, daemon=True).start()
+    child: Optional[subprocess.Popen] = None
+    rc = 127
+    try:
+        child = subprocess.Popen(
+            argv,
+            close_fds=True,
+            env=_operation_env(),
+            creationflags=0x08000000 | 0x00000004,  # NO_WINDOW | SUSPENDED
+        )
+        assign = kernel32.AssignProcessToJobObject
+        assign.argtypes = (wintypes.HANDLE, wintypes.HANDLE)
+        assign.restype = wintypes.BOOL
+        if not assign(job, wintypes.HANDLE(child._handle)):
+            raise OSError(ctypes.get_last_error(), "AssignProcessToJobObject")
+        if cancelled.is_set():
+            terminate_job()
+        else:
+            _resume_windows_process(kernel32, wintypes, child.pid)
+        rc = child.wait()
+        # A successful direct child may leave helpers behind; terminate the
+        # nested stable Job before reporting completion.
+        terminate_job()
+    except OSError:
+        terminate_job()
+        if child is not None:
+            try:
+                # Assignment itself may have failed, leaving this suspended
+                # process outside the nested Job.  Terminate it through its
+                # stable process handle before waiting; never strand an
+                # unassigned operation or rely on the outer desktop Job.
+                child.kill()
+            except OSError:
+                # The suspended child may have exited during Job teardown.
+                pass
+            try:
+                child.wait(timeout=5)
+            except (OSError, subprocess.TimeoutExpired):
+                # The outer desktop Job remains the terminal containment fallback.
+                pass
+    finally:
+        payload = _RESULT.pack(int(rc))
+        payload_buffer = ctypes.create_string_buffer(payload)
+        written = wintypes.DWORD()
+        kernel32.WriteFile(
+            wintypes.HANDLE(result_fd),
+            payload_buffer,
+            len(payload),
+            ctypes.byref(written),
+            None,
+        )
+        kernel32.CloseHandle(wintypes.HANDLE(result_fd))
+        with job_lock:
+            job_open = False
+            kernel32.CloseHandle(job)
+    return rc
+
+
+def supervisor_main(args: list[str]) -> int:
+    if len(args) < 5 or args[0] != "--supervise" or args[3] != "--":
+        return 2
+    control_fd = int(args[1])
+    result_fd = int(args[2])
+    argv = args[4:]
+    secure_backend_drain_fd()
+    if os.name == "posix":
+        return _supervise_posix(control_fd, result_fd, argv)
+    return _supervise_windows(control_fd, result_fd, argv)
+
+
+def _main() -> int:
+    return supervisor_main(sys.argv[1:])
+
+
+if __name__ == "__main__":
+    raise SystemExit(_main())

backend/core/device_caps.py

@@ -177,6 +177,23 @@ def gfx_for_hsa_override(value: str) -> str | None:
 #: The ROCm kernel driver interface. Its absence, or its presence without
 #: permission, are the two commonest reasons a ROCm host silently runs on CPU.
 _KFD_DEVICE = "/dev/kfd"
+_DXG_DEVICE = "/dev/dxg"
+_DXG_RUNTIME_PATHS = (
+    "/usr/lib/libdxcore.so",
+    "/usr/lib/librocdxg.so",
+    "/usr/share/rocdxg/dids.conf",
+)
+
+
+def _rocm_requires_dxg_detection(version: object) -> bool:
+    """Whether WSL's ROCDXG bridge still needs its explicit opt-in."""
+    try:
+        parts = str(version).split(".")
+        return (int(parts[0]), int(parts[1])) < (7, 13)
+    except (IndexError, TypeError, ValueError):
+        # Unknown versions get the conservative advice. The variable is
+        # harmless on newer runtimes and necessary on every older one.
+        return True
 
 
 def why_no_gpu(torch) -> tuple[str, ...]:
@@ -230,6 +247,40 @@ def why_no_gpu(torch) -> tuple[str, ...]:
         # /dev/kfd only exists on Linux; on any other platform its absence
         # says nothing, so don't invent a reason.
         if sys.platform.startswith("linux"):
+            if not os.path.exists(_KFD_DEVICE) and os.path.exists(_DXG_DEVICE):
+                if not os.access(_DXG_DEVICE, os.R_OK | os.W_OK):
+                    return (
+                        f"ROCm {hip} is installed and {_DXG_DEVICE} exists, "
+                        "but this process cannot open it — pass "
+                        "--device /dev/dxg to the WSL container",
+                    )
+                dxg_detection = os.environ.get("HSA_ENABLE_DXG_DETECTION", "").strip()
+                if dxg_detection == "0":
+                    return (
+                        f"ROCm {hip} is installed and {_DXG_DEVICE} is reachable, "
+                        "but HSA_ENABLE_DXG_DETECTION=0 explicitly disables the "
+                        "WSL GPU bridge; remove it or set it to 1",
+                    )
+                if _rocm_requires_dxg_detection(hip) and dxg_detection != "1":
+                    return (
+                        f"ROCm {hip} is installed and {_DXG_DEVICE} is "
+                        "reachable, but this pre-7.13 runtime requires "
+                        "HSA_ENABLE_DXG_DETECTION=1 inside WSL containers",
+                    )
+                missing = [
+                    path for path in _DXG_RUNTIME_PATHS if not os.path.exists(path)
+                ]
+                if missing:
+                    return (
+                        f"ROCm {hip} can reach {_DXG_DEVICE}, but the WSL "
+                        "ROCDXG runtime mounts are incomplete; missing: "
+                        f"{', '.join(missing)}",
+                    )
+                return (
+                    f"ROCm {hip} and the WSL ROCDXG bridge are reachable, "
+                    "but no GPU was enumerated — verify the AMD Windows "
+                    "driver, librocdxg/ROCm compatibility, and host `rocminfo`",
+                )
             if not os.path.exists(_KFD_DEVICE):
                 return (
                     f"ROCm {hip} is installed but {_KFD_DEVICE} is not "

backend/core/diagnose.py

@@ -21,6 +21,7 @@ Check shape:
 """
 from __future__ import annotations
 
+import importlib
 import os
 import platform
 import shutil
@@ -367,10 +368,44 @@ def run_diagnostics(include_network: bool = True, deep: bool = False) -> dict:
     counts = {OK: 0, WARN: 0, FAIL: 0}
     for c in checks:
         counts[c["status"]] += 1
+    engine_execution = []
+    for family in ("tts", "asr"):
+        active = "unknown"
+        try:
+            module = importlib.import_module(f"services.{family}_backend")
+            active = module.active_backend_id()
+            row = next((item for item in module.list_backends() if item.get("id") == active), None)
+            if row is not None:
+                engine_execution.append({
+                    "family": family,
+                    "engine_id": active,
+                    **row["execution_evidence"],
+                })
+        except Exception:  # noqa: BLE001 - evidence must not break diagnostics
+            # Preserve the other family's successful evidence and make this
+            # collection failure explicit without exposing exception text.
+            engine_execution.append({
+                "family": family,
+                "engine_id": active,
+                "implementation_variant": None,
+                "declared_device_families": [],
+                "evidence_state": "collection_failed",
+                "actual_execution_provider": None,
+                "actual_execution_device": None,
+                "gpu_name": None,
+                "gpu_architecture": None,
+                "precision_or_quantization": None,
+                "cpu_fallback_reason": None,
+                "cpu_fallback_stage": None,
+                "parent_memory_observable": None,
+                "runtime_versions": {},
+            })
+
     return {
         "app_version": APP_VERSION,
         "platform": scrub_text(platform.platform()),
         "checks": checks,
+        "engine_execution": engine_execution,
         "summary": {
             "ok": counts[FAIL] == 0,
             "passed": counts[OK],
@@ -395,6 +430,35 @@ def format_text(report: dict) -> str:
         lines.append(f"{tag[c['status']]} {c['label']}: {c['detail']}")
         if c.get("hint"):
             lines.append(f"       hint: {c['hint']}")
+    if report.get("engine_execution"):
+        lines.append("")
+        lines.append("Engine execution evidence:")
+        for item in report["engine_execution"]:
+            if item.get("actual_execution_provider"):
+                provider = item["actual_execution_provider"]
+            elif item.get("evidence_state") == "subprocess_loaded_provider_unreported":
+                provider = "loaded child; provider not reported"
+            else:
+                provider = "not loaded"
+            precision = item.get("precision_or_quantization") or "unknown"
+            device = item.get("actual_execution_device") or "unknown"
+            gpu = item.get("gpu_name") or "none"
+            architecture = item.get("gpu_architecture") or "unknown"
+            fallback_stage = item.get("cpu_fallback_stage") or "none"
+            fallback_reason = item.get("cpu_fallback_reason") or "none"
+            versions = ",".join(
+                f"{name}={version}"
+                for name, version in sorted(item.get("runtime_versions", {}).items())
+            ) or "none"
+            visible = "yes" if item.get("parent_memory_observable") else "no"
+            lines.append(
+                f"  {item['family']}:{item['engine_id']} provider={provider}; "
+                f"device={device}; gpu={gpu}; architecture={architecture}; "
+                f"precision={precision}; fallback-stage={fallback_stage}; "
+                f"fallback-reason={fallback_reason}; runtimes={versions}; "
+                f"evidence-state={item.get('evidence_state', 'unknown')}; "
+                f"parent-memory-visible={visible}"
+            )
     s = report["summary"]
     lines.append("")
     lines.append(

backend/core/failure.py

@@ -52,6 +52,7 @@ _REDACTED_VALUE = "***REDACTED***"
 # One-line "what to do" per docs-taxonomy key. Keys mirror error_docs_map's
 # taxonomy; the docs URL itself stays owned by error_docs_map.
 _HINTS: dict[str, str] = {
+    "GPU_OOM": "Close other GPU-heavy apps or unload models, then retry. You can also choose CPU in Settings → Performance & Device or select a smaller TTS engine.",
     "WORKER_AT_CAPACITY": "Wait for a running job on that worker to finish, or choose another available worker and retry.",
     "MODEL_NOT_INSTALLED": "Install or enable this engine on the worker machine, then refresh its capabilities and retry.",
     "MODEL_NOT_DOWNLOADED": "Open Models, install this model on the selected worker, then retry when the download completes.",
@@ -290,6 +291,9 @@ def append_hf_mirror_hint(text: str) -> str:
 # must NOT be added: its bare "timed out" trigger would stamp a "video server"
 # hint on a model-load timeout that leaks through the 500 handler.
 _CONTEXT_FREE_HINT_CLASSES = frozenset({
+    # Device allocator signatures are specific enough to attach the shared
+    # recovery without exposing CUDA's process table or filesystem paths.
+    "GPU_OOM",
     "SOCKS_PROXY_SUPPORT_MISSING",
     "SSL_HANDSHAKE_FAILURE",
     # Its trigger is an exact OpenSSL string, so it cannot be confused with
@@ -323,6 +327,38 @@ def append_hint(text: str) -> str:
     return f"{text} — {hint}" if hint else text
 
 
+_GPU_OOM_SIGNATURES = (
+    "cuda out of memory",
+    "cuda error: out of memory",
+    "cuda_error_out_of_memory",
+    "mps backend out of memory",
+    "hip out of memory",
+    "out of memory on device",
+)
+
+
+def is_gpu_oom(error: BaseException | str) -> bool:
+    """Recognize device OOMs through wrappers without importing torch."""
+    pending: list[BaseException] = [error] if isinstance(error, BaseException) else []
+    seen: set[int] = set()
+    while pending:
+        current = pending.pop()
+        if id(current) in seen:
+            continue
+        seen.add(id(current))
+        if type(current).__name__ == "OutOfMemoryError":
+            return True
+        if any(signature in str(current).lower() for signature in _GPU_OOM_SIGNATURES):
+            return True
+        if current.__cause__ is not None:
+            pending.append(current.__cause__)
+        if current.__context__ is not None:
+            pending.append(current.__context__)
+    if isinstance(error, str):
+        return any(signature in error.lower() for signature in _GPU_OOM_SIGNATURES)
+    return False
+
+
 def classify(reason: str) -> str:
     """Map a failure reason to a docs-taxonomy key, or "" when unknown.
 
@@ -330,6 +366,8 @@ def classify(reason: str) -> str:
     backend log / diagnostic names the same class the UI deeplink will use.
     """
     low = (reason or "").lower()
+    if is_gpu_oom(low):
+        return "GPU_OOM"
     if "pkg_resources" in low:
         return "PKG_RESOURCES_MISSING"
     if "quarantine" in low or "is damaged" in low or "gatekeeper" in low:
@@ -519,6 +557,7 @@ def classify(reason: str) -> str:
         or "unable to download video" in low
         or "remote end closed" in low
         or "timed out" in low
+        or "the page needs to be reloaded" in low
     ):
         return "VIDEO_DOWNLOAD_NETWORK"
     # #1227: Windows Smart App Control / WDAC / AppLocker refused to load a

backend/core/parent_liveness.py

@@ -0,0 +1,34 @@
+"""Terminate a desktop-contained backend when its owning shell disappears."""
+from __future__ import annotations
+
+import os
+import sys
+import threading
+from typing import BinaryIO, Callable
+
+
+def _watch_parent_pipe(reader: BinaryIO, exit_process: Callable[[int], None]) -> None:
+    """Block until the desktop-owned stdin pipe closes, then exit immediately."""
+    try:
+        while reader.read(1):
+            pass
+    except (OSError, ValueError):
+        # A broken or already-closed parent-owned pipe is equivalent to EOF.
+        pass
+    exit_process(0)
+
+
+def arm_desktop_parent_watchdog() -> bool:
+    """Use stdin EOF as an unforgeable parent-liveness signal for desktop runs."""
+    if os.environ.get("OMNIVOICE_DESKTOP_CONTAINED") != "1":
+        return False
+    reader = getattr(sys.stdin, "buffer", None)
+    if reader is None:
+        return False
+    threading.Thread(
+        target=_watch_parent_pipe,
+        args=(reader, os._exit),
+        name="desktop-parent-watchdog",
+        daemon=True,
+    ).start()
+    return True

backend/core/path_authorization.py

@@ -7,6 +7,7 @@ only the unguessable capability token crosses loopback HTTP.
 from __future__ import annotations
 
 import json
+import logging
 import os
 import re
 import secrets
@@ -14,6 +15,8 @@ import stat
 
 from core.config import DATA_DIR
 
+logger = logging.getLogger("omnivoice.path_authorization")
+
 _TOKEN_RE = re.compile(r"[0-9a-f]{64}\Z")
 _KINDS = {
     "models_dir",
@@ -40,25 +43,49 @@ def consume(token: str, expected_kind: str) -> str:
     if expected_kind not in _KINDS or not _TOKEN_RE.fullmatch(token or ""):
         raise PathAuthorizationError("Invalid or expired desktop authorization")
     root = _AUTH_DIR
+    # Distinguish "the store exists but this token isn't in it" (expired /
+    # already consumed / never issued — normal, no server-side signal) from
+    # "the store doesn't exist at all" (the desktop app and this backend are
+    # very likely pointed at different data directories, e.g. a dev backend
+    # started without OMNIVOICE_DATA_DIR, or a stale custom data folder — see
+    # #1781). The client-facing message is byte-identical either way (never
+    # leak local filesystem paths, or even which case occurred, over HTTP —
+    # CWE-200); the mismatch case additionally gets a server log line so it's
+    # diagnosable instead of a silent 403. That log line is deliberately
+    # path-free too (CWE-532: per-user filesystem paths, e.g. a home
+    # directory username, are sensitive and don't belong in application
+    # logs) — it names the failure mode, not the directory.
+    try:
+        entries = os.scandir(root)
+    except FileNotFoundError as exc:
+        logger.warning(
+            "path authorization store does not exist; the desktop app and "
+            "this backend likely resolved different data directories "
+            "(see #1781)"
+        )
+        raise PathAuthorizationError("Invalid or expired desktop authorization") from exc
+    except OSError as exc:
+        raise PathAuthorizationError("Invalid or expired desktop authorization") from exc
     candidate = None
     try:
-        for entry in os.scandir(root):
-            if not _TOKEN_RE.fullmatch(entry.name.removesuffix(".json")):
-                continue
-            if not entry.is_file(follow_symlinks=False):
-                continue
-            try:
-                with open(entry.path, "r", encoding="utf-8") as handle:
-                    probe = json.load(handle)
-            except (OSError, UnicodeError, json.JSONDecodeError):
-                continue  # Ignore corrupt/stale capabilities; they authorize nothing.
-            if isinstance(probe, dict) and secrets.compare_digest(
-                str(probe.get("token", "")), token
-            ):
-                candidate = entry.path
-                break
+        with entries:
+            for entry in entries:
+                if not _TOKEN_RE.fullmatch(entry.name.removesuffix(".json")):
+                    continue
+                if not entry.is_file(follow_symlinks=False):
+                    continue
+                try:
+                    with open(entry.path, "r", encoding="utf-8") as handle:
+                        probe = json.load(handle)
+                except (OSError, UnicodeError, json.JSONDecodeError):
+                    continue  # Ignore corrupt/stale capabilities; they authorize nothing.
+                if isinstance(probe, dict) and secrets.compare_digest(
+                    str(probe.get("token", "")), token
+                ):
+                    candidate = entry.path
+                    break
         if candidate is None:
-            raise OSError("capability not found")
+            raise PathAuthorizationError("Invalid or expired desktop authorization")
         claimed = os.path.join(root, f".consuming-{os.getpid()}-{secrets.token_hex(16)}")
         os.replace(candidate, claimed)
     except OSError as exc:

backend/core/prefs.py

@@ -91,3 +91,53 @@ def resolve(key: str, *, env: Optional[str] = None, default: Any = None) -> Any:
         if v:
             return v
     return get(key, default)
+
+
+# ── external-override detection (#1787 review fix) ──────────────────────────
+# restore_env() below uses os.environ.setdefault(), so a value already present
+# in the process's environment (shell profile, `.env`, Docker `-e`, systemd
+# unit, …) silently wins over anything saved in prefs.json — the setdefault
+# call is a no-op. That is the right behavior (env stays authoritative,
+# matching resolve()'s contract above), but a Settings control that persists a
+# value to prefs.json must not tell the user it "took effect after restart"
+# when an external source will keep shadowing it on every future restart too.
+#
+# _EXTERNALLY_PROVIDED records, once per process start, every bare key that
+# was ALREADY present in os.environ the moment restore_env() ran — i.e.
+# before our own setdefault() calls could have put it there, and before any
+# value our Settings UI ever wrote (Settings only ever writes prefs.json plus
+# the CURRENT process's os.environ; it never touches a shell profile or `.env`
+# file). Snapshotting unconditionally — not only for keys prefs.json already
+# has an entry for — means is_env_shadowed() also answers correctly for a key
+# a user is about to save for the FIRST time. Membership is stable for the
+# life of the process (nothing removes an inherited env var), and since a
+# plain restart re-inherits the same shell / container environment, it is
+# also a reliable predictor for the NEXT start: if the external source is
+# still exporting the key, the next restart will be shadowed again the same
+# way.
+_EXTERNALLY_PROVIDED: frozenset[str] = frozenset()
+
+
+def restore_env(data: dict) -> None:
+    """Restore ``env.*`` prefs into ``os.environ`` (startup only).
+
+    Called once from main.py's ``env_prefs`` step, before any user code reads
+    ``os.environ``. Snapshots which keys were already externally provided —
+    see :func:`is_env_shadowed` — then applies every saved ``env.*`` pref via
+    ``setdefault`` (never overriding an explicitly-set env var).
+    """
+    global _EXTERNALLY_PROVIDED
+    _EXTERNALLY_PROVIDED = frozenset(os.environ.keys())
+    for k, v in data.items():
+        if not k.startswith("env.") or not v:
+            continue
+        os.environ.setdefault(k[len("env."):], str(v))
+
+
+def is_env_shadowed(key: str) -> bool:
+    """Whether *key* was already present in the environment from a source
+    other than our own prefs restore, as of the last time :func:`restore_env`
+    ran. If prefs.json holds (or will hold) a saved value for *key*, that
+    value is being silently ignored — and will be again on the next restart —
+    unless the external source is removed."""
+    return key in _EXTERNALLY_PROVIDED

backend/core/public_errors.py

@@ -32,8 +32,8 @@ def stream_failure(code: str) -> dict[str, object]:
             "code": "generation_timeout",
             "detail": (
                 "Generation exceeded the compute-time limit. The backend is "
-                "still running; try a shorter passage or raise the generation "
-                "timeout."
+                "still running; try a shorter passage, or raise the "
+                "compute-time budget in Settings → Performance & Device."
             ),
             "retryable": True,
         },
@@ -74,6 +74,48 @@ def stream_failure(code: str) -> dict[str, object]:
     return dict(failures.get(code, failures["generation_failed"]))
 
 
+def stream_generation_failure(error: BaseException | object) -> dict[str, object]:
+    """``generation_failed`` stream metadata, enriched with the actual cause.
+
+    The bare "Generation failed. Check the selected engine and try again." is
+    the floor for an *unrecognized* failure. When the private exception DOES
+    classify to a known failure class — a corrupt model cache, an unreachable
+    Hugging Face mirror, a missing ffmpeg/ffprobe, a Windows paging-file limit,
+    a SOCKS/TLS proxy problem, … — the stable VoiceStudio-owned remediation for
+    that class is appended so the user can self-diagnose instead of guessing
+    which engine or which failure. This is the same enrichment the classic
+    (non-streaming) ``/generate`` 500 already gets via
+    :func:`public_exception_response`; the in-band streaming error frame
+    replaces the global 500 handler for a streaming request and used to bypass
+    it entirely (#1607).
+
+    Only VoiceStudio-owned constants are copied — never a substring of
+    ``error`` (Constitution I). Never raises: a diagnosis failure must not
+    replace the failure being diagnosed.
+    """
+    payload = stream_failure("generation_failed")
+    try:
+        enriched = public_exception_response(error, fallback=str(payload["detail"]))
+    except Exception:
+        return payload
+    hint = enriched.get("hint")
+    if hint:
+        payload["detail"] = enriched["detail"]
+        payload["hint"] = hint
+        topic = enriched.get("docs_topic")
+        if topic:
+            payload["docs_topic"] = topic
+            try:
+                from core import error_docs_map
+
+                url = error_docs_map.ERROR_DOCS.get(topic, "")
+            except Exception:
+                url = ""
+            if url:
+                payload["docs_url"] = url
+    return payload
+
+
 def public_failure(
     logger: logging.Logger,
     log_message: str,

backend/core/version.py

@@ -24,7 +24,7 @@ from pathlib import Path
 # tests/test_app_version.py::test_all_version_files_in_lockstep and bumped by
 # release.yml's version-bump job, so it stays equal to
 # pyproject/tauri.conf/Cargo/package.json.
-_FALLBACK_VERSION = "0.5.0"
+_FALLBACK_VERSION = "0.5.2"
 
 
 def _fallback_version() -> str:

backend/engines/omnivoice_gguf/backend.py

@@ -98,6 +98,7 @@ def _platform_slug() -> str:
         darwin-x86_64
         windows-x86_64
         linux-x86_64
+        linux-aarch64
     """
     system = platform.system().lower()
     machine = platform.machine().lower()
@@ -107,6 +108,8 @@ def _platform_slug() -> str:
         return "darwin-x86_64"
     if system == "windows":
         return "windows-x86_64"
+    if system == "linux" and machine in ("arm64", "aarch64"):
+        return "linux-aarch64"
     # Linux + everything else falls into the linux slug.
     return "linux-x86_64"
 

backend/engines/omnivoice_subprocess/__init__.py

@@ -1,7 +1,9 @@
 """omnivoice-subprocess: the resident OmniVoice TTS engine in a crash-isolated
 sidecar process (#730/#1190).
 
-The default ``omnivoice`` engine runs in-process on the GPU ``ThreadPoolExecutor``.
+The ``omnivoice`` engine runs in-process on CUDA, ROCm, and CPU. On MPS it is
+resolved to :class:`OmniVoiceMPSSubprocessBackend` so a fatal native allocator
+exit cannot take down the local API process.
 When a generate or load there exceeds its execution budget the pool is "reset"
 but the abandoned worker *thread* cannot be killed (Python cannot interrupt a
 native torch/MPS call), so it holds the MPS device until it finishes on its
@@ -13,16 +15,11 @@ timeout the parent's watchdog calls ``proc.kill()``, reclaiming the child's
 VRAM/device, and the next request transparently respawns a fresh sidecar. That
 is the one thing the in-process engine structurally cannot do.
 
-OPT-IN (Settings -> Engines, or ``OMNIVOICE_TTS_BACKEND=omnivoice-subprocess``);
-the in-process ``omnivoice`` stays the default so existing users see no change.
+The explicit ``omnivoice-subprocess`` id remains available on every host for
+operators who want the same containment elsewhere.
 
-Tradeoff vs the in-process engine: identical model and quality, a little extra
-per-call overhead (one stdio round-trip), and it does not carry the native
-advanced-parameter surface (``t_shift`` / ``layer_penalty_factor`` /
-``position_temperature`` / ``class_temperature``) or parent-side seed
-determinism, because the generic ``backend.generate`` path does not forward
-those. Acceptable for unattended / reaction-triggered use where reliability
-matters more than those controls.
+Tradeoff vs the in-process engine: identical model, controls, seed behavior,
+and quality, with a little extra per-call overhead (one stdio round-trip).
 
 Unlike IndexTTS / dots.tts / Supertonic-3, this sidecar runs under the PARENT
 interpreter (``venv_python() -> sys.executable``): the goal here is crash
@@ -51,10 +48,15 @@ class OmniVoiceSubprocessBackend(SubprocessBackend):
     id = "omnivoice-subprocess"
     display_name = "OmniVoice (subprocess-isolated, killable on timeout)"
     _DEFAULT_SAMPLE_RATE = 24000
-    gpu_compat = ("cuda", "mps", "cpu")
+    gpu_compat = ("cuda", "rocm", "mps", "cpu")
     # Match OmniVoiceBackend: the measured floor below which a render that
     # should take seconds runs for minutes (the #1226/#1222 4 GB reports).
     min_vram_gb = 6.0
+    # Packaged Windows hosts can spend more than the base 30 seconds starting
+    # the shared Python runtime before this stdlib-only sidecar emits ready.
+    # Keep the bound below the 300-second generation budget while avoiding the
+    # repeated false kill captured in #1711.
+    spawn_ready_timeout_s = 120.0
 
     @classmethod
     def is_available(cls) -> tuple[bool, str]:
@@ -102,4 +104,34 @@ class OmniVoiceSubprocessBackend(SubprocessBackend):
         return ["multi"]
 
 
-__all__ = ["OmniVoiceSubprocessBackend"]
+class OmniVoiceMPSSubprocessBackend(OmniVoiceSubprocessBackend):
+    """Effective ``omnivoice`` implementation on MPS.
+
+    Native torch/MPS allocator failures can terminate the process without a
+    catchable Python exception. Keeping the same engine id and model surface in
+    a child makes that failure recoverable while Settings, APIs, and saved
+    projects continue to refer to ``omnivoice``.
+    """
+
+    id = "omnivoice"
+    display_name = "VoiceStudio (k2-fsa/OmniVoice, 600+ languages)"
+    supports_native_omnivoice_controls = True
+
+    def generate(self, text: str, **kw):
+        from services.model_manager import make_room_before_generate
+
+        make_room_before_generate()
+        try:
+            return super().generate(text, **kw)
+        except RuntimeError as exc:
+            if "sidecar closed pipe mid-generate" not in str(exc):
+                raise
+            raise RuntimeError(
+                "The isolated OmniVoice engine stopped during generation, "
+                "usually because macOS reclaimed it under memory pressure. "
+                "The VoiceStudio backend is still running. Close memory-heavy "
+                "apps or select a smaller TTS engine, then retry."
+            ) from exc
+
+
+__all__ = ["OmniVoiceMPSSubprocessBackend", "OmniVoiceSubprocessBackend"]

backend/engines/omnivoice_subprocess/main.py

@@ -50,6 +50,8 @@ OMNIVOICE_SAMPLE_RATE = 24000
 _GEN_KW_ALLOWLIST = (
     "language", "instruct", "duration", "num_step", "guidance_scale",
     "speed", "denoise", "postprocess_output", "preprocess_prompt",
+    "t_shift", "layer_penalty_factor", "position_temperature",
+    "class_temperature", "audio_chunk_duration", "audio_chunk_threshold",
 )
 
 _model = None
@@ -183,6 +185,12 @@ def _handle_synthesize(msg: dict, stdout) -> None:
     ref_text = msg.get("ref_text") or None
     gen_kw = {k: msg[k] for k in _GEN_KW_ALLOWLIST if k in msg}
 
+    seed = msg.get("seed")
+    if seed is not None:
+        import torch
+
+        torch.manual_seed(int(seed))
+
     audios = model.generate(
         text=text, ref_audio=ref_audio, ref_text=ref_text, **gen_kw
     )

backend/main.py

@@ -1,4 +1,3 @@
-import math
 import os
 import sys
 
@@ -10,6 +9,24 @@ _backend_dir = os.path.dirname(os.path.abspath(__file__))
 if _backend_dir not in sys.path:
     sys.path.insert(0, _backend_dir)
 
+# PyInstaller re-executes this entry module when the frozen backend binary is
+# launched. Nested operation supervisors therefore dispatch here, before math,
+# logging, FastAPI, torch, or any application initialization. Source launches
+# use this same entry contract so frozen/source behavior cannot drift.
+if __name__ == "__main__" and len(sys.argv) > 1 and sys.argv[1] == "--supervise":
+    from core.contained_subprocess import supervisor_main
+
+    raise SystemExit(supervisor_main(sys.argv[1:]))
+
+# Rust clears CLOEXEC only for the backend exec. Re-arm PEP 446 immediately:
+# nested supervisors receive this descriptor solely through explicit pass_fds,
+# so a third-party close_fds=False child cannot hold the desktop drain barrier.
+from core.contained_subprocess import secure_backend_drain_fd  # noqa: E402
+
+secure_backend_drain_fd()
+
+import math  # noqa: E402
+
 # Windows: run every child process (ffmpeg, engine sidecars, yt-dlp, demucs, …)
 # WITHOUT popping a console window. The backend itself is spawned console-less by
 # the Tauri shell, so on Windows each console subprocess it launches would
@@ -60,6 +77,7 @@ os.environ.setdefault("FOR_DISABLE_CONSOLE_CTRL_HANDLER", "1")
 # (utils.hf_progress.SafeFileWrapper — same wrapper the patched hub tqdm
 # already uses for its own fp.)
 from utils.hf_progress import SafeFileWrapper as _SafeStdio  # noqa: E402
+from core.parent_liveness import arm_desktop_parent_watchdog  # noqa: E402
 
 # Force UTF-8 stdio before wrapping (#1155): on Windows the spawned backend's
 # stdout defaults to cp1252, and any library that prints user text (kittentts
@@ -72,6 +90,11 @@ for _stream in (sys.stdout, sys.stderr):
     except Exception:  # noqa: BLE001 — pythonw/frozen builds may lack reconfigure
         pass
 
+# The desktop keeps the backend's stdin pipe open for its own lifetime. EOF is
+# therefore a stable ownership signal that survives PID reuse and lets a child
+# terminate even when the shell crashes before its normal process-tree teardown.
+arm_desktop_parent_watchdog()
+
 if not getattr(sys.stdout, "_is_safe_wrapper", False):
     sys.stdout = _SafeStdio(sys.stdout)
 if not getattr(sys.stderr, "_is_safe_wrapper", False):
@@ -562,13 +585,14 @@ def _phase_a_build_inner() -> None:
         pass  # never block startup on the migration; it retries next launch
     # Restore persisted env vars from prefs.json (Settings UI writes them
     # there so they survive backend restarts) — before any user code reads
-    # os.environ, and never overriding an explicitly-set env var.
+    # os.environ, and never overriding an explicitly-set env var. Also
+    # snapshots which keys an external source (shell, `.env`, Docker, …)
+    # already provided, so a Settings control can tell the user their saved
+    # value is being shadowed instead of silently promising it will apply
+    # (core.prefs.is_env_shadowed — #1787 review fix).
     try:
-        from core.prefs import _load as _load_all_prefs
-        _prefs = _load_all_prefs()
-        for _k, _v in _prefs.items():
-            if _k.startswith("env.") and _v:
-                os.environ.setdefault(_k[len("env."):], str(_v))
+        from core.prefs import _load as _load_all_prefs, restore_env
+        restore_env(_load_all_prefs())
     except Exception:
         pass  # prefs.json missing or broken — fine on first run
     # yt-dlp user-update overlay: must run before anything imports yt_dlp so
@@ -649,6 +673,7 @@ def _phase_a_build_inner() -> None:
         events,
         capture,
         capture_ws,
+        speech_platform,
         dictation,
         openai_compat,
         tts_stream,
@@ -661,14 +686,15 @@ def _phase_a_build_inner() -> None:
         settings as settings_router,  # Phase 1 AUTH-03: HF token save/clear/state
         media_tools as media_tools_router,  # Audio tools: ffmpeg/ffprobe/yt-dlp
         auth as auth_router,
+        voice_convert,  # Studio Convert: speech-to-speech via ASR → TTS
     )
     from api.routers import mcp_bindings as _mcp_bindings_router  # noqa: E402
     from api.routers import workers as workers_router  # noqa: E402
     _router_modules.extend([
-        system, profiles, exports, generation, dub_core, dub_generate,
+        system, profiles, exports, generation, voice_convert, dub_core, dub_generate,
         dub_export, dub_translate, projects, glossary, engines, tools,
         stories, setup, gallery, archetypes, describe_voice, community,
-        batch, watermark, events, capture, capture_ws, dictation,
+        batch, watermark, events, capture, capture_ws, speech_platform, dictation,
         openai_compat, tts_stream, marketplace, personas, sonitranslate,
         audiobook, longform_jobs, pronunciation, settings_router,
         media_tools_router, auth_router, _mcp_bindings_router, workers_router,
@@ -991,6 +1017,9 @@ async def _phase_b(app: FastAPI) -> None:
 
 @asynccontextmanager
 async def lifespan(app: FastAPI):
+    from api.dependencies import validate_server_admin_key
+
+    validate_server_admin_key()
     # Startup watchdog (#632): a silent hang during startup (e.g. a model-load /
     # MCP deadlock on some platforms) means "Application startup complete" never
     # logs and the app sits forever with no error. If startup hasn't finished

backend/mcp_server.py

@@ -7,8 +7,8 @@ Run standalone:
 
 Tools exposed:
     generate_speech   — text → WAV audio (voice clone or design)
-    clone_voice       — base64 reference audio → new voice profile
-    transcribe        — base64 audio → text
+    clone_voice       — reference audio (base64, or a file path) → new voice profile
+    transcribe        — audio (base64, or a file path) → text
     list_voices       — enumerate saved voice profiles
     list_languages    — available TTS languages
     list_personalities — voice personality presets
@@ -17,6 +17,18 @@ Tools exposed:
 Resources exposed:
     voice://{profile_id}  — voice profile metadata
     history://recent      — last 20 generated audio items
+
+Output mode (OMNIVOICE_MCP_OUTPUT_MODE):
+    resources — generate_speech returns the WAV as base64 inline (the original
+                contract; default)
+    files     — it returns a URL to the render (and, with a base path, a WAV
+                written there); nothing large ever enters the agent's context
+    both      — both of the above
+
+File inputs (OMNIVOICE_MCP_BASE_PATH):
+    One directory that agents may read audio from (transcribe / clone_voice
+    `*_path` arguments) and receive files in (files mode). It is the security
+    boundary: with no base path configured, path-shaped inputs are refused.
 """
 from __future__ import annotations
 
@@ -25,6 +37,8 @@ import base64
 import json
 import logging
 import os
+import re
+import stat
 import sys
 
 logger = logging.getLogger("omnivoice.mcp")
@@ -69,6 +83,244 @@ def _sniff_audio_ext(raw: bytes) -> str:
     return ".wav"
 
 
+# ── Output mode + the base path boundary ─────────────────────────────────
+# An LLM agent that receives a WAV as base64 pays for every byte in context:
+# a 1.4 s clip already brushes per-result token caps, and a paragraph of
+# narration blows them outright. The ElevenLabs MCP settled this with an
+# OUTPUT_MODE (files / resources / both) and a BASE_PATH that doubles as the
+# security boundary for file-shaped inputs; the same two knobs here, named in
+# the OMNIVOICE_* family the rest of the server reads.
+
+_OUTPUT_MODES = ("resources", "files", "both")
+_MAX_INPUT_BYTES = 200 * 1024 * 1024
+_SAFE_AUDIO_ID = re.compile(r"^[A-Za-z0-9_-]{1,64}$")
+
+
+def _output_mode() -> str:
+    """How generate_speech hands audio back (OMNIVOICE_MCP_OUTPUT_MODE).
+
+    'resources' is the original base64-inline contract and stays the default
+    so existing integrations see no change; 'files' returns a URL to the
+    render (plus a WAV under the base path when one is configured); 'both'
+    returns everything. Anything unrecognized falls back to 'resources' with
+    a warning rather than failing the tool."""
+    mode = os.environ.get("OMNIVOICE_MCP_OUTPUT_MODE", "resources").strip().lower()
+    if mode not in _OUTPUT_MODES:
+        logger.warning(
+            "OMNIVOICE_MCP_OUTPUT_MODE=%r is not one of %s; using 'resources'",
+            mode, _OUTPUT_MODES,
+        )
+        return "resources"
+    return mode
+
+
+def _base_path() -> "str | None":
+    """The one directory agents may read audio from and receive files in
+    (OMNIVOICE_MCP_BASE_PATH), realpath'd; None when unset."""
+    raw = os.environ.get("OMNIVOICE_MCP_BASE_PATH", "").strip()
+    if not raw:
+        return None
+    return os.path.realpath(os.path.expanduser(raw))
+
+
+def _resolve_under_base(path: str) -> str:
+    """Absolute realpath of ``path`` when it lies inside the base path.
+
+    Relative paths resolve against the base; absolute paths must already be
+    inside it. Both sides are realpath'd, so a symlink pointing outward cannot
+    smuggle a read in. Raises ValueError with an agent-legible reason when no
+    base path is configured or the path escapes it."""
+    base = _base_path()
+    if base is None:
+        raise ValueError(
+            "OMNIVOICE_MCP_BASE_PATH is not set; file paths are refused until it "
+            "names a directory"
+        )
+    candidate = os.path.realpath(os.path.join(base, os.path.expanduser(path)))
+    if not _path_is_under_base(base, candidate):
+        raise ValueError(f"{path!r} resolves outside OMNIVOICE_MCP_BASE_PATH")
+    return candidate
+
+
+def _opened_file_is_confined(fd: int, resolved: str, base: str) -> bool:
+    """Verify that an opened descriptor still names a file under ``base``."""
+    proc_fd = f"/proc/self/fd/{fd}"
+    if os.path.exists(proc_fd):
+        return _path_is_under_base(base, os.path.realpath(proc_fd))
+    try:
+        current = os.path.realpath(resolved)
+        return _path_is_under_base(base, current) and os.path.samestat(
+            os.fstat(fd), os.stat(current, follow_symlinks=False)
+        )
+    except OSError:
+        return False
+
+
+def _path_is_under_base(base: str, candidate: str) -> bool:
+    try:
+        common = os.path.commonpath([base, candidate])
+    except ValueError:  # different drives on Windows
+        return False
+    return os.path.normcase(common) == os.path.normcase(base)
+
+
+def _open_under_base(path: str, flags: int, *, mode: int = 0o600) -> tuple[int, str]:
+    """Open ``path`` without following a component replaced after validation."""
+    base = _base_path()
+    if base is None:
+        raise ValueError(
+            "OMNIVOICE_MCP_BASE_PATH is not set; file paths are refused until it "
+            "names a directory"
+        )
+    resolved = _resolve_under_base(path)
+    relative = os.path.relpath(resolved, base)
+    parts = [part for part in relative.split(os.sep) if part not in ("", ".")]
+    if not parts or parts[0] == os.pardir:
+        raise ValueError(f"{path!r} resolves outside OMNIVOICE_MCP_BASE_PATH")
+
+    no_follow = getattr(os, "O_NOFOLLOW", 0)
+    close_on_exec = getattr(os, "O_CLOEXEC", 0)
+    binary = getattr(os, "O_BINARY", 0)
+    file_flags = flags | no_follow | close_on_exec | binary
+    supports_dir_fd = os.open in getattr(os, "supports_dir_fd", ())
+    directory_flag = getattr(os, "O_DIRECTORY", 0)
+
+    if supports_dir_fd and directory_flag:
+        directory_flags = os.O_RDONLY | directory_flag | no_follow | close_on_exec
+        directory_fd = os.open(base, directory_flags)
+        try:
+            for component in parts[:-1]:
+                next_fd = os.open(component, directory_flags, dir_fd=directory_fd)
+                os.close(directory_fd)
+                directory_fd = next_fd
+            fd = os.open(parts[-1], file_flags, mode, dir_fd=directory_fd)
+        finally:
+            os.close(directory_fd)
+    else:
+        fd = os.open(resolved, file_flags, mode)
+
+    if not _opened_file_is_confined(fd, resolved, base):
+        os.close(fd)
+        raise ValueError(f"{path!r} resolves outside OMNIVOICE_MCP_BASE_PATH")
+    return fd, resolved
+
+
+def _read_input_audio(
+    audio_base64: "str | None",
+    audio_path: "str | None",
+    *,
+    label: str = "audio_base64",
+    too_big: str = "audio exceeds 200 MB limit",
+) -> "tuple[bytes | None, str | None]":
+    """Audio bytes from exactly one of the two input lanes, or (None, error).
+
+    The base64 lane keeps its data-URI tolerance and 200 MB cap; the path lane
+    is honored only inside the base path (the security boundary) and applies
+    the same cap to the file's size before reading it."""
+    if bool(audio_base64) == bool(audio_path):
+        return None, f"pass exactly one of {label} or the matching *_path argument"
+    if audio_path:
+        try:
+            fd, _resolved = _open_under_base(audio_path, os.O_RDONLY)
+        except ValueError as e:
+            return None, str(e)
+        except FileNotFoundError:
+            return None, f"no such file under OMNIVOICE_MCP_BASE_PATH: {audio_path!r}"
+        except OSError as e:
+            return None, f"could not safely read {audio_path!r}: {e}"
+        with os.fdopen(fd, "rb") as handle:
+            info = os.fstat(handle.fileno())
+            if not stat.S_ISREG(info.st_mode):
+                return None, f"{audio_path!r} is not a regular file"
+            if info.st_size > _MAX_INPUT_BYTES:
+                return None, too_big
+            raw = handle.read(_MAX_INPUT_BYTES + 1)
+        if len(raw) > _MAX_INPUT_BYTES:
+            return None, too_big
+        if not raw:
+            return None, f"{label} is empty"
+        return raw, None
+    encoded = (
+        audio_base64.split(",", 1)[-1]
+        if audio_base64.startswith("data:")
+        else audio_base64
+    )
+    max_encoded_bytes = 4 * ((_MAX_INPUT_BYTES + 2) // 3)
+    if len(encoded) > max_encoded_bytes:
+        return None, too_big
+    raw = _decode_ref_audio(audio_base64)
+    if raw is None:
+        return None, f"{label} is not valid base64"
+    if not raw:
+        return None, f"{label} is empty"
+    if len(raw) > _MAX_INPUT_BYTES:
+        return None, too_big
+    return raw, None
+
+
+def _write_output(audio_id: str, raw: bytes) -> str:
+    """Land a render under the base path as ``<audio_id>.wav``; returns the path."""
+    if not _SAFE_AUDIO_ID.fullmatch(audio_id):
+        raise ValueError("backend returned an invalid X-Audio-Id header")
+    base = _base_path()
+    os.makedirs(base, exist_ok=True)
+    filename = f"{audio_id}.wav"
+    fd, path = _open_under_base(filename, os.O_WRONLY | os.O_CREAT | os.O_EXCL)
+    with os.fdopen(fd, "wb") as handle:
+        handle.write(raw)
+    return path
+
+
+def _post_timeout_s() -> float:
+    """Seconds the tools wait on a backend POST (OMNIVOICE_MCP_TIMEOUT_S,
+    default 120). A CPU host renders a paragraph in minutes and serializes
+    generations, so an agent behind another render used to hit the fixed
+    budget with an empty-message timeout; the knob follows the backend's own
+    OMNIVOICE_GENERATE_TIMEOUT_S when a deployment raises that."""
+    raw = os.environ.get("OMNIVOICE_MCP_TIMEOUT_S", "").strip()
+    try:
+        value = float(raw) if raw else 120.0
+    except ValueError:
+        logger.warning("OMNIVOICE_MCP_TIMEOUT_S=%r is not a number; using 120", raw)
+        return 120.0
+    return value if value > 0 else 120.0
+
+
+def _maybe_number(value):
+    """A response-header number as a number, or the raw text (e.g. '?')."""
+    try:
+        return float(value)
+    except (TypeError, ValueError):
+        return value
+
+
+def _speech_result(audio_id: str, gen_time, duration, raw: bytes, api_base: str) -> dict:
+    """The generate_speech reply shaped by the output mode.
+
+    The backend already keeps every render on disk and serves it at
+    ``/audio/<audio_id>.wav``, so files mode costs nothing but a URL - plus one
+    write when a base path invites the WAV into the agent's own directory."""
+    if not _SAFE_AUDIO_ID.fullmatch(audio_id):
+        raise ValueError("backend returned an invalid X-Audio-Id header")
+    mode = _output_mode()
+    out = {
+        "audio_id": audio_id,
+        "generation_time_s": gen_time,
+        "audio_duration_s": duration,
+        "format": "wav",
+        "output_mode": mode,
+    }
+    if mode in ("files", "both"):
+        out["audio_url"] = f"{api_base.rstrip('/')}/audio/{audio_id}.wav"
+        if _base_path() is not None:
+            out["output_path"] = _write_output(audio_id, raw)
+        else:
+            out["note"] = "set OMNIVOICE_MCP_BASE_PATH to also receive the WAV as a file"
+    if mode in ("resources", "both"):
+        out["wav_base64"] = base64.b64encode(raw).decode("ascii")
+    return out
+
+
 # ── Lazy imports — keeps startup fast when not using MCP ────────────────
 
 
@@ -147,7 +399,7 @@ def create_mcp_server():
 
     async def _api_post_form(path: str, data: dict, files: dict | None = None):
         import httpx
-        async with httpx.AsyncClient(base_url=_api_base(), timeout=120) as c:
+        async with httpx.AsyncClient(base_url=_api_base(), timeout=_post_timeout_s()) as c:
             r = await c.post(path, data=data, files=files or {})
             r.raise_for_status()
             return r
@@ -190,8 +442,12 @@ def create_mcp_server():
             steps: Diffusion steps (8=fast/draft, 16=balanced, 32=quality).
 
         Returns:
-            JSON with audio_id, generation_time, audio_duration, and
-            base64-encoded WAV data.
+            JSON with audio_id, generation_time_s, audio_duration_s and the
+            audio itself shaped by OMNIVOICE_MCP_OUTPUT_MODE: base64 WAV data
+            ('resources', the default), a URL to the render plus a WAV under
+            OMNIVOICE_MCP_BASE_PATH when one is set ('files'), or all of the
+            above ('both'). Prefer 'files' for LLM agents: nothing large
+            enters the context.
         """
         # Per-agent voice binding (Wave 2.2): explicit arg wins; otherwise
         # resolve this client's bound profile, then the global default.
@@ -218,18 +474,10 @@ def create_mcp_server():
         r = await _api_post_form("/generate", data=form)
 
         audio_id = r.headers.get("X-Audio-Id", "unknown")
-        gen_time = r.headers.get("X-Gen-Time", "?")
-        duration = r.headers.get("X-Audio-Duration", "?")
+        gen_time = _maybe_number(r.headers.get("X-Gen-Time", "?"))
+        duration = _maybe_number(r.headers.get("X-Audio-Duration", "?"))
 
-        wav_b64 = base64.b64encode(r.content).decode("ascii")
-
-        return (
-            f'{{"audio_id":"{audio_id}",'
-            f'"generation_time_s":{gen_time},'
-            f'"audio_duration_s":{duration},'
-            f'"format":"wav",'
-            f'"wav_base64":"{wav_b64}"}}'
-        )
+        return json.dumps(_speech_result(audio_id, gen_time, duration, r.content, _api_base()))
 
     @mcp.tool()
     async def list_voices() -> str:
@@ -266,30 +514,39 @@ def create_mcp_server():
         )
 
     @mcp.tool()
-    async def transcribe(audio_base64: str, language: str | None = None) -> str:
+    async def transcribe(
+        audio_base64: str | None = None,
+        audio_path: str | None = None,
+        language: str | None = None,
+    ) -> str:
         """Transcribe spoken audio to text.
 
+        Pass exactly one of audio_base64 or audio_path.
+
         Args:
             audio_base64: Base64-encoded audio bytes (wav/mp3/webm/m4a).
+            audio_path: Path to an audio file under OMNIVOICE_MCP_BASE_PATH
+                (relative to it, or absolute inside it). The base path is the
+                security boundary: with none configured, paths are refused.
+                Prefer this lane for LLM agents - the audio never enters the
+                agent's context.
             language: Optional language hint; omit for auto-detect.
 
         Returns:
             JSON with the recognized text, language, and duration.
         """
-        try:
-            raw = base64.b64decode(audio_base64, validate=True)
-        except Exception:
-            return '{"error":"audio_base64 is not valid base64"}'
-        # 200 MB cap — same spirit as voicebox's transcribe gate. Keeps a
-        # buggy/hostile agent from posting an unbounded blob.
-        if len(raw) > 200 * 1024 * 1024:
-            return '{"error":"audio exceeds 200 MB limit"}'
+        # 200 MB cap on both lanes — same spirit as voicebox's transcribe
+        # gate. Keeps a buggy/hostile agent from posting an unbounded blob.
+        raw, err = _read_input_audio(audio_base64, audio_path)
+        if err:
+            return json.dumps({"error": err})
         data = {}
         if language:
             data["language"] = language
         r = await _api_post_form(
             "/transcribe", data=data,
-            files={"audio": ("audio.wav", raw, "application/octet-stream")},
+            files={"audio": (f"audio{_sniff_audio_ext(raw)}", raw,
+                             "application/octet-stream")},
         )
         return str(r.json())
 
@@ -319,15 +576,17 @@ def create_mcp_server():
     @mcp.tool()
     async def clone_voice(
         name: str,
-        ref_audio_base64: str,
+        ref_audio_base64: str | None = None,
         ref_text: str = "",
         instruct: str = "",
         language: str = "Auto",
+        ref_audio_path: str | None = None,
     ) -> str:
         """Clone a new voice profile from a reference audio sample.
 
         The new voice is immediately available for use with generate_speech
-        (pass the returned profile_id as the profile_id argument).
+        (pass the returned profile_id as the profile_id argument). Pass
+        exactly one of ref_audio_base64 or ref_audio_path.
 
         Args:
             name: A human-friendly name for the cloned voice.
@@ -338,19 +597,20 @@ def create_mcp_server():
                 quality for some engines).
             instruct: Optional style instruction (e.g. 'whisper', 'excited').
             language: Language of the reference audio (ISO code or 'Auto').
+            ref_audio_path: Path to the reference audio under
+                OMNIVOICE_MCP_BASE_PATH (relative to it, or absolute inside
+                it); refused when no base path is configured. Prefer this
+                lane for LLM agents - the clip never enters the context.
 
         Returns:
             JSON with the new profile's id, name, and kind.
         """
-        # Reject oversized inputs before decoding (base64 is always larger
-        # than raw, so this is a safe lower bound on the decoded size).
-        if len(ref_audio_base64) > 200 * 1024 * 1024:
-            return '{"error":"reference audio exceeds 200 MB limit"}'
-        raw = _decode_ref_audio(ref_audio_base64)
-        if raw is None:
-            return '{"error":"ref_audio_base64 is not valid base64"}'
-        if not raw:
-            return '{"error":"ref_audio_base64 is empty"}'
+        raw, err = _read_input_audio(
+            ref_audio_base64, ref_audio_path,
+            label="ref_audio_base64", too_big="reference audio exceeds 200 MB limit",
+        )
+        if err:
+            return json.dumps({"error": err})
         import httpx
         try:
             r = await _api_post_form(

backend/schemas/requests.py

@@ -190,6 +190,7 @@ class ParseSubtitleTextRequest(BaseModel):
 class DubIngestUrlRequest(BaseModel):
     url: str
     job_id: Optional[str] = None
+    source_lang: Optional[str] = None
     # When true and the URL is a caption-bearing host (YouTube, Vimeo, TED…),
     # ask yt-dlp to also download the original-language + any additional
     # sub_langs as VTT. The UI uses this to seed a transcript without running

backend/services/admin_sessions.py

@@ -33,7 +33,12 @@ WS_TICKET_PREFIX = "ovs_ws_ticket_"
 _TOKEN_BYTES = 32
 _ENCODED_TOKEN_LENGTH = 43
 _TOKEN_BODY_RE = re.compile(rf"^[A-Za-z0-9_-]{{{_ENCODED_TOKEN_LENGTH}}}$")
-_ALLOWED_WS_PATHS = frozenset({"/ws/events", "/ws/transcribe"})
+# Every ticketed WebSocket route. The first-party mirror is ``ALLOWED_WS_PATHS``
+# in frontend/src/api/authSession.ts — a route missing here mints a 422 and the
+# UI consumer fails silently (#1769 added /ws/tts for the live dub preview).
+_ALLOWED_WS_PATHS = frozenset(
+    {"/ws/events", "/ws/transcribe", "/ws/tts", "/v1/audio/transcriptions/stream"}
+)
 _ADMIN_CAPABILITIES = frozenset({"consume", "admin"})
 _KEY_GENERATION_INFO = b"omnivoice-admin-key-generation-v1"
 

backend/services/asr_backend.py

@@ -24,12 +24,15 @@ faster-whisper because it's available on every platform we ship to).
 from __future__ import annotations
 
 import asyncio
+import ipaddress
 import logging
 import os
 import re
 import contextlib
 import threading
 import time
+import weakref
+from urllib.parse import urlsplit
 from utils.containment import contain_system_exit
 
 from abc import ABC, abstractmethod
@@ -88,10 +91,9 @@ def reset_pool_after_wedge(executor, *, what: str = "ASR") -> bool:
 
 
 # ── Consecutive-timeout streak → recommend the crash-isolated engine ────────
-# A pool reset restores *capacity*, but the wedged CTranslate2/whisperx thread
-# keeps its VRAM until the process exits. When guarded transcribes keep timing
-# out back-to-back in one session, resets clearly aren't recovering the
-# underlying hang — the durable fix is the crash-isolated sidecar engine
+# A timed-out CTranslate2/whisperx thread keeps its worker and VRAM until the
+# native call exits. When guarded transcribes keep timing out back-to-back in
+# one session, the durable fix is the crash-isolated sidecar engine
 # (services.subprocess_asr, #393), whose child process CAN be hard-killed to
 # reclaim the hung call and its VRAM. We only *recommend* it (log + error
 # message); we never switch engines automatically (owner rule: no silent
@@ -146,41 +148,89 @@ def _isolated_engine_hint(streak: int) -> str:
 
 async def run_transcribe_guarded(executor, fn, *, what: str = "ASR",
                                  timeout: float = ASR_TRANSCRIBE_TIMEOUT_S,
-                                 timeout_env: str = "OMNIVOICE_ASR_TRANSCRIBE_TIMEOUT_S"):
+                                 timeout_env: str = "OMNIVOICE_ASR_TRANSCRIBE_TIMEOUT_S",
+                                 reset_on_timeout: bool = False,
+                                 on_abandon=None):
     """Run a blocking transcribe ``fn`` in ``executor`` with a hard wall-clock
     bound. On timeout, raise :class:`ASRTimeoutError` with guidance instead of
     letting the request hang forever.
 
-    ``run_in_executor`` cannot cancel the underlying thread, so a wedged
-    transcribe (a CTranslate2 / whisperx / VAD hang seen on some Windows + CUDA
-    setups, #730) keeps occupying its GPU-pool worker. With a 1–2 worker pool
-    that starves every *other* request — including TTS generate — and the next
-    thing the user does surfaces as "Can't reach the local backend" even though
-    the process is alive. So on timeout we also ``reset()`` the pool when it
-    supports it (``_ResilientGpuPool``): the wedged thread is abandoned and the
-    next submit gets a fresh worker, restoring capacity without an app restart.
-    The orphaned thread still holds its VRAM until the process exits, which is
-    why the message still recommends a smaller ASR model / Flush as the durable
-    fix. Executors without ``reset`` (a plain ThreadPoolExecutor in tests) just
-    get the bound + actionable error.
+    A future cannot cancel the underlying thread, so a timed-out
+    in-process CTranslate2/whisperx call still owns its model and device. The
+    default deliberately leaves that worker accounted for: swapping in a fresh
+    pool and immediately retrying the same backend overlaps two native calls,
+    which produced the Windows access violation in #1669. A caller backed by a
+    genuinely killable process may opt into ``reset_on_timeout``.
+
+    ``on_abandon`` is called once after a timed-out or cancelled worker can no
+    longer access its inputs. Queued work cancelled before it starts calls it
+    immediately; running work calls it from the worker finalizer. Normal
+    completion leaves cleanup with the caller.
     """
     loop = asyncio.get_running_loop()
     # Same SystemExit containment as the TTS pool (#1133 class): an ASR
     # dependency written as a CLI must not be able to shut the backend down.
-    fut = loop.run_in_executor(executor, contain_system_exit(fn, what))
+    inner = contain_system_exit(fn, what)
+    abandon_lock = threading.Lock()
+    abandon_state = {
+        "requested": False,
+        "finished": False,
+        "callback_called": False,
+    }
+
+    def _fire_abandon_callback() -> None:
+        if on_abandon is None:
+            return
+        with abandon_lock:
+            if abandon_state["callback_called"]:
+                return
+            abandon_state["callback_called"] = True
+        try:
+            on_abandon()
+        except Exception:  # noqa: BLE001 — cleanup cannot hide the ASR result
+            logger.exception("%s abandon cleanup failed", what)
+
+    def _job():
+        try:
+            return inner()
+        finally:
+            with abandon_lock:
+                abandon_state["finished"] = True
+                abandoned = abandon_state["requested"]
+            if abandoned:
+                _fire_abandon_callback()
+
+    concurrent_fut = executor.submit(_job)
+    fut = asyncio.wrap_future(concurrent_fut, loop=loop)
+
+    def _abandon() -> None:
+        cancelled_before_start = concurrent_fut.cancel()
+        with abandon_lock:
+            abandon_state["requested"] = True
+            finished = abandon_state["finished"]
+        fut.cancel()
+        if cancelled_before_start or finished:
+            _fire_abandon_callback()
+
     try:
-        result = await asyncio.wait_for(fut, timeout=timeout)
+        # Shield the wrapper so timeout does not discard our ability to tell a
+        # queued cancellation from a native thread that is still running.
+        result = await asyncio.wait_for(asyncio.shield(fut), timeout=timeout)
+    except asyncio.CancelledError:
+        _abandon()
+        raise
     except asyncio.TimeoutError:
-        # Free the poisoned pool so a hung transcribe can't keep starving TTS /
-        # other ASR work (the "can't reach backend" symptom, #730).
-        reset_pool_after_wedge(executor, what=what)
+        _abandon()
+        if reset_on_timeout:
+            reset_pool_after_wedge(executor, what=what)
         streak = _note_transcribe_timeout()
         msg = (
             f"{what} transcription exceeded {timeout:.0f}s and was abandoned — "
             "the backend is running, but the ASR model is too heavy for the "
             "available compute. Most often the GPU is VRAM-starved: the resident "
             "TTS model and a large ASR model (large-v3) contend for memory. "
-            "Capacity was restored automatically, but for a durable fix Flush the "
+            "The native call cannot be killed safely, so its capacity remains "
+            "reserved until it exits. For a durable fix Flush the "
             "TTS model to free VRAM, pick a smaller ASR model in "
             f"Model Catalogue → Models, or set ASR to CPU. (Raise {timeout_env} "
             "for very long transcribes.)"
@@ -310,6 +360,16 @@ class ASRBackend(ABC):
     # broken GPU path, strictly worse than the honest `cpu_fallback`.)
     gpu_compat: tuple[str, ...] = ("cpu",)
 
+    def execution_evidence_loaded(self) -> bool:
+        """Whether this instance has live model state worth reporting."""
+        if getattr(self, "runs_out_of_process", False):
+            proc = getattr(self, "_proc", None)
+            return proc is not None and proc.poll() is None
+        return any(
+            getattr(self, attr, None) is not None
+            for attr in ("_model", "_asr", "_pipeline", "_pipe", "_transcriber", "_rec")
+        )
+
     @classmethod
     @abstractmethod
     def is_available(cls) -> tuple[bool, str]:
@@ -956,6 +1016,8 @@ class FasterWhisperBackend(ASRBackend):
         # (after the #551 compute_type / #255 OOM→CPU fallback chain).
         self._device: str | None = None
         self._compute_type: str | None = None
+        self._fallback_reason: str | None = None
+        self._fallback_stage: str | None = None
 
     @classmethod
     def is_available(cls) -> tuple[bool, str]:
@@ -1033,6 +1095,8 @@ class FasterWhisperBackend(ASRBackend):
                 except Exception:  # noqa: BLE001 — cache clear is best-effort
                     pass
                 device = "cpu"
+                self._fallback_reason = "CUDA memory was exhausted while loading the engine"
+                self._fallback_stage = "model_load"
                 candidates = _compute_type_candidates(device)
                 compute_type = candidates[0]
                 continue
@@ -1991,6 +2055,42 @@ _ASR_OPENAI_COMPAT_MODEL_KEY = "asr.openai_compat.model"
 _ASR_OPENAI_COMPAT_SECRET_NAME = "asr_openai_compat_key"
 
 
+def normalize_openai_compat_asr_base_url(value: str) -> str:
+    """Normalize a safe ASR endpoint, allowing plain HTTP only on loopback."""
+    base = (value or "").strip().rstrip("/")
+    if not base:
+        return ""
+    try:
+        parsed = urlsplit(base)
+        _ = parsed.port
+    except (TypeError, ValueError) as exc:
+        raise ValueError("Invalid OpenAI-compatible ASR base URL") from exc
+    scheme = parsed.scheme.lower()
+    if (
+        scheme not in {"http", "https"}
+        or not parsed.hostname
+        or parsed.username is not None
+        or parsed.password is not None
+        or parsed.query
+        or parsed.fragment
+    ):
+        raise ValueError(
+            "OpenAI-compatible ASR base URL must be a credential-free HTTP(S) URL"
+        )
+    host = parsed.hostname.lower()
+    loopback = host == "localhost"
+    if not loopback:
+        try:
+            address = ipaddress.ip_address(host)
+            address = getattr(address, "ipv4_mapped", None) or address
+            loopback = address.is_loopback
+        except ValueError:
+            loopback = False
+    if scheme == "http" and not loopback:
+        raise ValueError("Non-loopback OpenAI-compatible ASR endpoints require HTTPS")
+    return base
+
+
 def resolve_openai_compat_asr_base_url() -> str:
     from services import settings_store
     return (
@@ -2054,7 +2154,7 @@ def probe_openai_compat_server(
     maps to a translated message:
 
       not_configured   no base URL anywhere
-      invalid_url      base URL without an http(s):// scheme
+      invalid_url      malformed URL or non-loopback HTTP endpoint
       ok               2xx — ``model_found`` says whether the configured
                        model appears in the server's list (None = unknown)
       ok_no_models     404/405/501 — reachable, but no /models endpoint
@@ -2069,7 +2169,7 @@ def probe_openai_compat_server(
 
     from core.scrub import scrub_text
 
-    base = (base_url if base_url is not None else resolve_openai_compat_asr_base_url()).strip().rstrip("/")
+    configured_base = base_url if base_url is not None else resolve_openai_compat_asr_base_url()
     mdl = (model if model is not None else resolve_openai_compat_asr_model()).strip()
     if api_key is None:
         key = resolve_openai_compat_asr_api_key()
@@ -2085,9 +2185,11 @@ def probe_openai_compat_server(
         "model_found": None,
         "detail": None,
     }
-    if not base:
+    if not configured_base.strip():
         return out
-    if not base.startswith(("http://", "https://")):
+    try:
+        base = normalize_openai_compat_asr_base_url(configured_base)
+    except ValueError:
         out["status"] = "invalid_url"
         return out
 
@@ -2098,7 +2200,7 @@ def probe_openai_compat_server(
     try:
         with httpx.Client(
             timeout=httpx.Timeout(timeout_s, connect=min(5.0, timeout_s)),
-            follow_redirects=True,
+            follow_redirects=False,
         ) as client:
             resp = client.get(f"{base}/models", headers=headers)
     except httpx.TimeoutException as exc:
@@ -2161,13 +2263,20 @@ class OpenAICompatASRBackend(ASRBackend):
     gpu_compat = ("cpu",)  # network client only — no local compute
 
     def __init__(self):
-        self._base_url = resolve_openai_compat_asr_base_url()
+        self._base_url = normalize_openai_compat_asr_base_url(
+            resolve_openai_compat_asr_base_url()
+        )
         self._model = resolve_openai_compat_asr_model()
 
     @classmethod
     def is_available(cls) -> tuple[bool, str]:
-        if not resolve_openai_compat_asr_base_url():
+        base_url = resolve_openai_compat_asr_base_url()
+        if not base_url:
             return False, "Configure a server endpoint in Model Catalogue → Engines"
+        try:
+            normalize_openai_compat_asr_base_url(base_url)
+        except ValueError as exc:
+            return False, str(exc)
         try:
             import openai  # noqa: F401
         except ImportError:
@@ -2175,13 +2284,18 @@ class OpenAICompatASRBackend(ASRBackend):
         return True, "ready"
 
     def _client(self):
-        from openai import OpenAI
+        from openai import DefaultHttpxClient, OpenAI
         api_key = resolve_openai_compat_asr_api_key() or "not-needed"
         # max_retries=0: mirrors llm_skills.resolve_skill_client — a
         # rate-limited/slow server retrying inside the SDK would blow past
         # whatever bounded timeout the caller (dub transcribe, dictation)
         # expects from a single call.
-        return OpenAI(base_url=self._base_url, api_key=api_key, max_retries=0)
+        return OpenAI(
+            base_url=self._base_url,
+            api_key=api_key,
+            max_retries=0,
+            http_client=DefaultHttpxClient(follow_redirects=False),
+        )
 
     def transcribe(self, audio_path: str, *, word_timestamps: bool = True) -> dict:
         logger.info(
@@ -2360,6 +2474,8 @@ _LAST_ERRORS: dict[str, str] = {}
 # failing ASR wholesale. Per-process by design: repairing the env requires a
 # reinstall / ``uv sync --reinstall`` and an app restart anyway.
 _DEEP_IMPORT_BROKEN: dict[str, str] = {}
+_RUNTIME_EVIDENCE: dict[str, dict] = {}
+_RUNTIME_INSTANCES: weakref.WeakValueDictionary[str, "ASRBackend"] = weakref.WeakValueDictionary()
 
 
 def _deep_import_reason(cls: type["ASRBackend"], exc: ImportError) -> str:
@@ -2390,6 +2506,7 @@ def list_backends() -> list[dict]:
     """
     from core.device_caps import detect_host_caps
     from core.scrub import scrub_text
+    from services.engine_evidence import snapshot as execution_snapshot
     from services.engine_routing import routing_fields
     caps = detect_host_caps()
 
@@ -2414,6 +2531,24 @@ def list_backends() -> list[dict]:
             _LAST_ERRORS[bid] = scrub_text(msg)
         isolation = "subprocess" if getattr(cls, "_is_subprocess_isolated", False) else "in-process"
         gpu_compat = getattr(cls, "gpu_compat", ("cpu",))
+        routing = routing_fields(gpu_compat, caps)
+        # Cached load-time facts are valid only while their exact backend still
+        # owns live model state. Recompute from that instance so unload/reaping
+        # cannot leave ghost GPU/provider evidence in diagnostics.
+        instance = (
+            _ISOLATED_INSTANCES.get(bid)
+            if isolation == "subprocess"
+            else _RUNTIME_INSTANCES.get(bid)
+        )
+        execution_evidence = execution_snapshot(
+            engine_id=bid,
+            engine_cls=cls,
+            instance=instance,
+            routing=routing,
+            caps=caps,
+        )
+        if execution_evidence["evidence_state"] == "not_loaded":
+            _RUNTIME_EVIDENCE.pop(bid, None)
         out.append({
             "id": bid,
             "display_name": cls.display_name,
@@ -2425,7 +2560,14 @@ def list_backends() -> list[dict]:
             "last_error": _LAST_ERRORS.get(bid),
             "isolation_mode": isolation,
             "gpu_compat": list(gpu_compat),
-            **routing_fields(gpu_compat, caps),
+            **routing,
+            "execution_evidence": execution_evidence or execution_snapshot(
+                engine_id=bid,
+                engine_cls=cls,
+                instance=None,
+                routing=routing,
+                caps=caps,
+            ),
         })
     return out
 
@@ -2666,6 +2808,21 @@ def load_active_asr_backend(*, asr_pipe=None) -> ASRBackend:
                 raise ASRModelMissingError(missing)
         try:
             backend.ensure_loaded()
+            from core.device_caps import detect_host_caps
+            from services.engine_evidence import snapshot as execution_snapshot
+            from services.engine_routing import routing_fields
+
+            cls = type(backend)
+            caps = detect_host_caps()
+            routing = routing_fields(getattr(cls, "gpu_compat", ("cpu",)), caps)
+            _RUNTIME_EVIDENCE[bid] = execution_snapshot(
+                engine_id=bid,
+                engine_cls=cls,
+                instance=backend,
+                routing=routing,
+                caps=caps,
+            )
+            _RUNTIME_INSTANCES[bid] = backend
             return backend
         except ImportError as e:
             # ModuleNotFoundError and its ImportError parent ("cannot import

backend/services/dub_pipeline.py

@@ -742,6 +742,74 @@ def _ensure_browser_playable_mp4(video_path: str) -> str:
     return video_path
 
 
+async def _ensure_browser_playable_mp4_for_job(job_id: str, video_path: str) -> str:
+    """Normalize an upload through the job's cancellable process registry."""
+    is_mp4 = video_path.lower().endswith(".mp4")
+    vcodec, acodec = await asyncio.to_thread(_probe_codecs, video_path)
+    if is_mp4 and vcodec in _BROWSER_VIDEO_CODECS and acodec in _BROWSER_AUDIO_CODECS:
+        return video_path
+
+    target = os.path.splitext(video_path)[0] + ".mp4"
+    if target == video_path:
+        target = os.path.splitext(video_path)[0] + ".browser.mp4"
+    run_proc = run_proc_factory(job_id)
+    ffmpeg_bin = find_ffmpeg()
+
+    async def attempt(cmd: list[str]) -> int:
+        try:
+            proc, _stdout, _stderr = await run_proc(cmd, timeout=1800.0)
+            return proc.returncode
+        except asyncio.CancelledError:
+            raise
+        except Exception as exc:
+            logger.warning(
+                "Browser-media normalization process failed for %s: %s",
+                log_safe(video_path),
+                log_safe(exc),
+            )
+            return 1
+
+    rc = 1
+    if not is_mp4:
+        rc = await attempt(
+            [
+                ffmpeg_bin, "-y", "-i", video_path,
+                "-c:v", "copy", "-c:a", "copy",
+                "-movflags", "+faststart", target,
+            ]
+        )
+        if rc == 0 and os.path.exists(target):
+            target_vcodec, target_acodec = await asyncio.to_thread(_probe_codecs, target)
+            if (
+                target_vcodec not in _BROWSER_VIDEO_CODECS
+                or target_acodec not in _BROWSER_AUDIO_CODECS
+            ):
+                rc = 1
+        else:
+            rc = 1
+    if rc != 0:
+        rc = await attempt(
+            [
+                ffmpeg_bin, "-y", "-i", video_path,
+                "-c:v", "libx264", "-preset", "veryfast", "-crf", "23",
+                "-pix_fmt", "yuv420p", "-c:a", "aac", "-b:a", "192k",
+                "-movflags", "+faststart", target,
+            ]
+        )
+    if rc == 0 and os.path.exists(target) and target != video_path:
+        try:
+            os.remove(video_path)
+        except OSError:
+            pass  # Best effort: the normalized target is already complete.
+        return target
+    logger.warning(
+        "Could not transcode %s to browser-playable mp4 — the in-app "
+        "video player may render this file as a black box.",
+        log_safe(video_path),
+    )
+    return video_path
+
+
 # Bounded retry for transient download failures (#579/#598). yt-dlp's own
 # `retries`/`fragment_retries` cover per-fragment HTTP flakes, but a broken
 # pipe ([Errno 32]) raised while the write side of a pipe closes mid-stream
@@ -1257,6 +1325,13 @@ async def ingest_pipeline(
         except Exception:
             dur = 0.0
 
+        # URL downloads already pass through this guard in yt_download_sync.
+        # Uploaded videos did not, so a valid VP9/AV1/Opus upload could be
+        # processed successfully but remain undecodable by the in-app WebView.
+        # Codec probing/transcoding is blocking; keep it off the event loop.
+        if source.get("kind") != "url" and input_type != "audio":
+            video_path = await _ensure_browser_playable_mp4_for_job(job_id, video_path)
+
         # Content-hash cache: reuse artifacts from previous matching jobs.
         content_hash = await asyncio.to_thread(compute_file_hash, audio_path)
         cached = find_cached_job(content_hash, job_id)
@@ -1295,6 +1370,7 @@ async def ingest_pipeline(
                 "scene_cuts": scene_cuts,
                 "youtube_subs": youtube_subs_by_lang or None,
                 "input_type": input_type,
+                "source_lang_override": source.get("source_lang"),
             }
             if not put_and_save_job(
                 job_id, full_job, filename=filename, duration=dur, content_hash=content_hash,
@@ -1323,6 +1399,7 @@ async def ingest_pipeline(
                 "scene_cuts": [],
                 "youtube_subs": youtube_subs_by_lang or None,
                 "input_type": input_type,
+                "source_lang_override": source.get("source_lang"),
             }
             if not put_and_save_job(
                 job_id, partial, filename=filename, duration=dur, content_hash=content_hash,

backend/services/engine_disk_usage.py

@@ -0,0 +1,230 @@
+"""Structured pre-install and measured disk costs for TTS engines."""
+from __future__ import annotations
+
+import os
+import threading
+import time
+from functools import lru_cache
+from pathlib import Path
+
+_GIB = 1024**3
+_CACHE_TTL_SECONDS = 10.0
+_measurement_cache: dict[str, tuple[float, dict]] = {}
+_measurement_lock = threading.Lock()
+
+# Catalogue/build estimates. ``None`` is deliberate: unknown costs must stay
+# visible instead of being silently treated as zero.
+_ESTIMATES: dict[str, dict] = {
+    "omnivoice": {
+        "package_download_bytes": None,
+        "unique_installed_bytes": None,
+        "potentially_shared_bytes": None,
+        "temporary_free_bytes": None,
+        "confidence": "estimated",
+        "destination": "hf_model_cache",
+        "deduplication": None,
+    },
+    "kittentts": {
+        "package_download_bytes": None,
+        "unique_installed_bytes": None,
+        "potentially_shared_bytes": None,
+        "temporary_free_bytes": None,
+        "confidence": "estimated",
+        "destination": "hf_model_cache",
+        "deduplication": None,
+    },
+}
+_MODEL_REPOS = {
+    "omnivoice": "k2-fsa/OmniVoice",
+    "kittentts": "KittenML/kitten-tts-mini-0.8",
+}
+
+
+def _volume_root(path: Path) -> str:
+    """Mount point/drive containing a possibly not-yet-created destination."""
+    try:
+        current = path.expanduser().resolve()
+        while not current.exists() and current.parent != current:
+            current = current.parent
+        device = current.stat().st_dev
+        while current.parent != current and current.parent.stat().st_dev == device:
+            current = current.parent
+        return str(current)
+    except OSError:
+        return "unknown"
+
+
+def _hf_cache_path() -> Path:
+    configured = (
+        os.environ.get("HF_HUB_CACHE")
+        or os.environ.get("HUGGINGFACE_HUB_CACHE")
+        or os.environ.get("HF_HOME")
+    )
+    return Path(configured) if configured else Path.home() / ".cache" / "huggingface"
+
+
+@lru_cache(maxsize=None)
+def _catalog_model_bytes(engine_id: str) -> int | None:
+    """Resolve the weight estimate from config/models.yaml, its source of truth."""
+    repo_id = _MODEL_REPOS.get(engine_id)
+    if repo_id is None:
+        return None
+    try:
+        import yaml
+
+        catalog_path = Path(__file__).resolve().parents[1] / "config" / "models.yaml"
+        entries = yaml.safe_load(catalog_path.read_text(encoding="utf-8"))["models"]
+        model = next(item for item in entries if item["repo_id"] == repo_id)
+        return round(float(model["size_gb"]) * _GIB)
+    except (OSError, KeyError, StopIteration, TypeError, ValueError):
+        return None
+
+
+def _dir_size(path: Path) -> int:
+    total = 0
+    try:
+        for root, _dirs, files in os.walk(path):
+            for filename in files:
+                try:
+                    total += os.path.getsize(os.path.join(root, filename))
+                except OSError:
+                    continue
+    except OSError:
+        return 0
+    return total
+
+
+def _sidecar_estimate(engine_id: str) -> dict | None:
+    try:
+        from services.sidecar_install import get_spec, managed_root
+
+        spec = get_spec(engine_id)
+    except Exception:
+        return None
+    if spec is None:
+        return None
+    model_bytes = spec.weights_bytes
+    dependency_bytes = spec.dependency_bytes
+    return {
+        "model_download_bytes": model_bytes,
+        "package_download_bytes": dependency_bytes,
+        "unique_installed_bytes": spec.required_bytes,
+        "potentially_shared_bytes": spec.potentially_shared_bytes,
+        "temporary_free_bytes": spec.temporary_free_bytes,
+        "confidence": spec.disk_confidence,
+        "destination": "engine_data",
+        "destination_volume": _volume_root(managed_root(spec)),
+        "deduplication": "uv_same_volume",
+    }
+
+
+def estimate_for(engine_id: str) -> dict:
+    estimate = _sidecar_estimate(engine_id) or _ESTIMATES.get(engine_id)
+    if estimate is not None:
+        return {
+            "model_download_bytes": _catalog_model_bytes(engine_id),
+            "destination_volume": _volume_root(_hf_cache_path()),
+            **estimate,
+        }
+    return {
+        "model_download_bytes": None,
+        "package_download_bytes": None,
+        "unique_installed_bytes": None,
+        "potentially_shared_bytes": None,
+        "temporary_free_bytes": None,
+        "confidence": "unknown",
+        "destination": "unknown",
+        "destination_volume": "unknown",
+        "deduplication": None,
+    }
+
+
+def _measure_sidecar(engine_id: str) -> dict | None:
+    try:
+        from services.sidecar_install import get_spec, managed_checkout, managed_root
+
+        spec = get_spec(engine_id)
+    except Exception:
+        return None
+    if spec is None:
+        return None
+    checkout = managed_checkout(spec)
+    if not checkout.is_dir():
+        return None
+    model = _dir_size(checkout / spec.weights_subdir)
+    environment = _dir_size(checkout / ".venv")
+    total = _dir_size(managed_root(spec))
+    shared_cache = _dir_size(managed_root(spec).parent / ".uv-cache")
+    return {
+        "model_bytes": model,
+        "environment_bytes": environment,
+        "cache_bytes": shared_cache,
+        "total_owned_bytes": total,
+        "confidence": "measured",
+    }
+
+
+def _measure_model_cache(engine_id: str) -> dict | None:
+    repo_id = _MODEL_REPOS.get(engine_id)
+    if repo_id is None:
+        return None
+    try:
+        from huggingface_hub import scan_cache_dir
+
+        repo = next((item for item in scan_cache_dir().repos if item.repo_id == repo_id), None)
+    except Exception:
+        return None
+    if repo is None or repo.size_on_disk <= 0:
+        return None
+    size = int(repo.size_on_disk)
+    return {
+        "model_bytes": size,
+        # The model lives in this cache; cache overhead is not separately
+        # attributable without double-counting the same hardlinked blobs.
+        "environment_bytes": None,
+        "cache_bytes": 0,
+        "total_owned_bytes": size,
+        "confidence": "measured",
+    }
+
+
+def actual_for(engine_id: str) -> dict:
+    now = time.monotonic()
+    cached = _measurement_cache.get(engine_id)
+    if cached and now - cached[0] < _CACHE_TTL_SECONDS:
+        return dict(cached[1])
+    # A cache miss can recursively walk a sidecar and the shared uv cache.
+    # Coalesce concurrent requests so callers cannot multiply that work.
+    with _measurement_lock:
+        now = time.monotonic()
+        cached = _measurement_cache.get(engine_id)
+        if cached and now - cached[0] < _CACHE_TTL_SECONDS:
+            return dict(cached[1])
+        actual = _measure_sidecar(engine_id) or _measure_model_cache(engine_id) or {
+            "model_bytes": None,
+            "environment_bytes": None,
+            "cache_bytes": None,
+            "total_owned_bytes": None,
+            "confidence": "unknown",
+        }
+        _measurement_cache[engine_id] = (now, actual)
+        return dict(actual)
+
+
+def disk_usage_for(engine_id: str) -> dict:
+    """Stable API shape consumed by the engine catalogue."""
+    return {"estimate": estimate_for(engine_id), "actual": actual_for(engine_id)}
+
+
+def disk_summary_for(engine_id: str) -> dict:
+    """Cheap list payload; measurement is deferred until the row is opened."""
+    return {
+        "estimate": estimate_for(engine_id),
+        "actual": {
+            "model_bytes": None,
+            "environment_bytes": None,
+            "cache_bytes": None,
+            "total_owned_bytes": None,
+            "confidence": "unknown",
+        },
+    }

backend/services/engine_evidence.py

@@ -0,0 +1,119 @@
+"""Sanitized, reproducible execution evidence for TTS and ASR engines."""
+from __future__ import annotations
+
+import importlib.metadata
+import platform
+from typing import Any
+
+
+def _version(distribution: str) -> str | None:
+    try:
+        return importlib.metadata.version(distribution)
+    except importlib.metadata.PackageNotFoundError:
+        return None
+
+
+def _value(instance: object, *names: str) -> str | None:
+    for name in names:
+        try:
+            value = getattr(instance, name, None)
+            if value is not None and not callable(value):
+                text = str(value).strip()
+                if text and len(text) <= 80 and "/" not in text and "\\" not in text:
+                    return text
+        except Exception:
+            continue
+    return None
+
+
+def runtime_versions(engine_id: str) -> dict[str, str]:
+    """Relevant installed library versions, never paths or environment values."""
+    names = {"python": platform.python_version()}
+    candidates = ["torch"]
+    low = engine_id.lower()
+    if "faster" in low or "whisperx" in low:
+        candidates.extend(["ctranslate2", "faster-whisper"])
+    if "sherpa" in low or "moonshine" in low:
+        candidates.append("onnxruntime")
+    if "mlx" in low:
+        candidates.append("mlx")
+    for name in candidates:
+        if (version := _version(name)) is not None:
+            names[name] = version
+    return names
+
+
+def snapshot(
+    *,
+    engine_id: str,
+    engine_cls: type,
+    instance: object | None,
+    routing: dict[str, Any],
+    caps: object,
+) -> dict[str, Any]:
+    """Return fixed-shape evidence; actual fields stay null until an instance loads."""
+    isolated = bool(
+        getattr(engine_cls, "_is_subprocess_isolated", False)
+        or getattr(engine_cls, "runs_out_of_process", False)
+    )
+    loaded = False
+    probe_failed = False
+    if instance is not None:
+        try:
+            contract = getattr(instance, "execution_evidence_loaded", False)
+            loaded = bool(contract() if callable(contract) else contract)
+        except Exception:  # noqa: BLE001 - third-party lifecycle descriptors may raise
+            probe_failed = True
+
+    actual_device = None
+    provider = None
+    precision = None
+    if loaded:
+        actual_device = _value(instance, "_device", "device", "execution_device")
+        provider = _value(instance, "_provider", "provider", "execution_provider")
+        precision = _value(
+            instance, "_compute_type", "compute_type", "_dtype", "dtype", "quantization"
+        )
+        if provider is None and actual_device is not None:
+            provider = actual_device
+
+    runtime_fallback_reason = _value(instance, "_fallback_reason", "fallback_reason") if loaded else None
+    runtime_fallback_stage = _value(instance, "_fallback_stage", "fallback_stage") if loaded else None
+    status = routing.get("routing_status")
+    fallback = status == "cpu_fallback" or runtime_fallback_reason is not None
+    evidence_state = "not_loaded"
+    if probe_failed:
+        evidence_state = "probe_error"
+    elif loaded:
+        evidence_state = "loaded"
+        if isolated and provider is None and actual_device is None:
+            evidence_state = "subprocess_loaded_provider_unreported"
+    return {
+        "implementation_variant": f"{engine_cls.__module__}.{engine_cls.__name__}",
+        "declared_device_families": list(getattr(engine_cls, "gpu_compat", ("cpu",))),
+        "evidence_state": evidence_state,
+        "actual_execution_provider": provider,
+        "actual_execution_device": actual_device,
+        "gpu_name": getattr(caps, "device_name", "") or None,
+        "gpu_architecture": _gpu_architecture(getattr(caps, "family", "cpu")),
+        "precision_or_quantization": precision,
+        "cpu_fallback_reason": runtime_fallback_reason or (routing.get("routing_reason") if fallback else None),
+        "cpu_fallback_stage": runtime_fallback_stage or ("routing_preflight" if fallback else None),
+        "parent_memory_observable": not isolated,
+        "runtime_versions": runtime_versions(engine_id),
+    }
+
+
+def _gpu_architecture(family: str) -> str | None:
+    if family not in {"cuda", "rocm"}:
+        return "apple-silicon" if family == "mps" else None
+    try:
+        import torch
+
+        if family == "rocm":
+            props = torch.cuda.get_device_properties(0)
+            return str(getattr(props, "gcnArchName", "") or "") or None
+        major, minor = torch.cuda.get_device_capability(0)
+        return f"sm_{major}{minor}"
+    except Exception:
+        return None

backend/services/gpu_gateway.py

@@ -177,6 +177,9 @@ class LocalCall:
     queue_timeout: Optional[float] = None
     # The engine's declared VRAM floor; only shapes the timeout message.
     min_vram_gb: float = 0.0
+    # Called once a local worker abandoned by its waiter can no longer touch
+    # request-owned inputs. Normal completion does not call it (#1668).
+    on_abandon: Optional[Callable[[], None]] = None
     # Some remote-first callers cannot construct the local callable without
     # loading the very model they are trying to offload.  Prepare it only when
     # routing/fallback actually selects this machine.
@@ -465,6 +468,7 @@ async def _run_local(call: LocalCall, *, admit: bool = False, executor=None) ->
         queue_timeout=call.queue_timeout,
         min_vram_gb=call.min_vram_gb,
         executor=executor,
+        on_abandon=call.on_abandon,
     )
 
 
@@ -499,7 +503,9 @@ async def _run_remote(
         deadline = _default_deadline(call.operation, params.get("text"))
 
     try:
-        task = scheduler.submit(
+        submit = getattr(scheduler, "submit_async", None)
+        submit = submit if callable(submit) else scheduler.submit
+        submitted = submit(
             operation=call.operation,
             engine=call.engine,
             model_id=call.model_id,
@@ -508,6 +514,7 @@ async def _run_remote(
             deadline_seconds=deadline,
             pinned_worker_id=decision.worker_id,
         )
+        task = await submitted if asyncio.iscoroutine(submitted) else submitted
     except QueueFull as exc:
         raise _NotDispatched(str(exc)) from exc
 

backend/services/karaoke_ass.py

@@ -0,0 +1,220 @@
+"""Karaoke (word-highlight) ASS builder for dub hardsub export.
+
+Pure text-in/text-out: no ffmpeg, no models, no filesystem. ``build_ass``
+turns subtitle cues into an ASS script whose lines carry ``\\k``/``\\kf``
+karaoke tags, so ffmpeg's ``ass=`` filter burns a word-by-word highlight
+sweep instead of the static line the SRT path renders.
+
+Word timing sources, in order:
+
+1. ``cue["words"]`` — per-word ``{text, start, end}`` persisted at
+   transcribe time (services.segmentation). Used only when the words still
+   spell the cue's display text: after translation the persisted ASR words
+   are source-language tokens, so re-using their timing would burn the
+   wrong language. The display text is always authoritative.
+2. Even split — the cue text's whitespace tokens spread uniformly across
+   ``[start, end]``. This is the compatibility path for jobs transcribed
+   before word persistence and for translated tracks.
+
+Dual-layout karaoke is intentionally unsupported (out of scope): callers
+must fall back to the line (SRT) burn when the dual layout is requested.
+"""
+
+from __future__ import annotations
+
+import re
+from typing import Optional, Sequence
+
+_WS = re.compile(r"\s+")
+
+#: Default ASS canvas. libass scales the script to the real video size, so
+#: one reference resolution keeps font/margin proportions stable everywhere.
+DEFAULT_PLAY_RES = (1920, 1080)
+
+_HEADER_TEMPLATE = """[Script Info]
+; Generated by VoiceStudio karaoke burn-in
+ScriptType: v4.00+
+PlayResX: {res_x}
+PlayResY: {res_y}
+WrapStyle: 0
+ScaledBorderAndShadow: yes
+
+[V4+ Styles]
+Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding
+Style: Default,Arial,64,&H0000E7FF,&H00FFFFFF,&H00101010,&H7F000000,0,0,0,0,100,100,0,0,1,3,1,2,96,96,48,1
+
+[Events]
+Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
+"""
+
+
+def _norm(text: object) -> str:
+    return _WS.sub(" ", str(text or "").strip())
+
+
+def _ass_time(seconds: float) -> str:
+    """``H:MM:SS.CC`` (centiseconds) — the ASS event timestamp format."""
+    cs = max(0, int(round(float(seconds) * 100)))
+    h, rem = divmod(cs, 360000)
+    m, rem = divmod(rem, 6000)
+    s, c = divmod(rem, 100)
+    return f"{h}:{m:02d}:{s:02d}.{c:02d}"
+
+
+def _ass_escape(text: str) -> str:
+    """Escape a display token for an ASS Dialogue text field.
+
+    Braces would open an override block (user text like ``{\\b1}`` must render
+    literally, never execute); newlines become ASS hard line breaks.
+    """
+    return (
+        str(text)
+        .replace("{", "\\{")
+        .replace("}", "\\}")
+        .replace("\r\n", "\\N")
+        .replace("\n", "\\N")
+        .replace("\r", "\\N")
+    )
+
+
+def _cs(seconds: float) -> int:
+    """Karaoke tag duration in centiseconds; ≥1 so a tag never renders as 0."""
+    return max(1, int(round(float(seconds) * 100)))
+
+
+def even_split_words(text: str, start: float, end: float) -> list[dict]:
+    """Uniformly distribute the cue text's whitespace tokens over [start, end].
+
+    The export fallback for jobs transcribed before per-word persistence and
+    for translated tracks (whose persisted words are source-language tokens).
+    """
+    tokens = [tok for tok in _WS.split(str(text or "").strip()) if tok]
+    if not tokens:
+        return []
+    start = float(start)
+    dur = max(0.0, float(end) - start) / len(tokens)
+    return [
+        {"text": tok, "start": start + i * dur, "end": start + (i + 1) * dur}
+        for i, tok in enumerate(tokens)
+    ]
+
+
+def scale_words(
+    words: Sequence[dict],
+    orig_start: float,
+    orig_end: float,
+    new_start: float,
+    new_end: float,
+) -> Optional[list[dict]]:
+    """Map word times linearly from [orig_start, orig_end] → [new_start, new_end].
+
+    Used when Smart Fit moves a cue onto the fitted timeline: the persisted
+    word times live on the original timeline and must ride along. Returns
+    ``None`` when either span is degenerate (caller should drop the words so
+    export falls back to an even split over the new span).
+    """
+    orig_span = float(orig_end) - float(orig_start)
+    new_span = float(new_end) - float(new_start)
+    if orig_span <= 0 or new_span <= 0:
+        return None
+    ratio = new_span / orig_span
+    out: list[dict] = []
+    for w in words:
+        try:
+            ws = float(w["start"])
+            we = float(w["end"])
+        except (KeyError, TypeError, ValueError):
+            return None
+        out.append({
+            **w,
+            "start": round(float(new_start) + (ws - float(orig_start)) * ratio, 3),
+            "end": round(float(new_start) + (we - float(orig_start)) * ratio, 3),
+        })
+    return out
+
+
+def _usable_words(cue: dict, text: str) -> Optional[list[tuple[str, float, float]]]:
+    """Persisted words, iff well-formed AND they spell the cue's display text."""
+    words = cue.get("words")
+    if not isinstance(words, list) or not words:
+        return None
+    clean: list[tuple[str, float, float]] = []
+    for w in words:
+        if not isinstance(w, dict):
+            return None
+        wtext = _norm(w.get("text"))
+        try:
+            ws = float(w["start"])
+            we = float(w["end"])
+        except (KeyError, TypeError, ValueError):
+            return None
+        if wtext:
+            clean.append((wtext, ws, we))
+    if not clean:
+        return None
+    if _norm(" ".join(t for t, _, _ in clean)) != text:
+        return None
+    return clean
+
+
+def _karaoke_text(cue: dict, text: str, start: float, end: float) -> str:
+    """One Dialogue text field: ``{\\k…}`` lead-in + per-word ``{\\kf…}`` tags.
+
+    Each word's sweep runs until the next word starts (the classic karaoke
+    layout — inter-word gaps finish the previous word's fill), and the last
+    word sweeps out to the cue end.
+    """
+    words = _usable_words(cue, text) or [
+        (w["text"], w["start"], w["end"]) for w in even_split_words(text, start, end)
+    ]
+    # Clamp into the cue span and enforce monotonic starts so malformed
+    # persisted data can only mistime the sweep, never corrupt the script.
+    clamped: list[tuple[str, float]] = []
+    prev = start
+    for wtext, ws, _ in words:
+        ws = min(max(ws, prev), end)
+        clamped.append((wtext, ws))
+        prev = ws
+    parts: list[str] = []
+    lead = clamped[0][1] - start
+    if lead > 0.005:
+        parts.append(f"{{\\k{_cs(lead)}}}")
+    for i, (wtext, ws) in enumerate(clamped):
+        nxt = clamped[i + 1][1] if i + 1 < len(clamped) else end
+        sep = " " if i + 1 < len(clamped) else ""
+        parts.append(f"{{\\kf{_cs(max(nxt, ws) - ws)}}}{_ass_escape(wtext)}{sep}")
+    return "".join(parts)
+
+
+def build_ass(
+    cues: Sequence[dict],
+    *,
+    dual: bool = False,
+    play_res: tuple[int, int] = DEFAULT_PLAY_RES,
+) -> str:
+    """Build a karaoke ASS script from subtitle cues ({text, start, end, words?}).
+
+    One ``Default`` style; one Dialogue event per cue. ``dual`` exists for
+    signature parity with the line burn but dual-layout karaoke is out of
+    scope — callers must keep the SRT line burn for dual, so requesting it
+    here is a contract violation, not a rendering mode.
+    """
+    if dual:
+        raise ValueError(
+            "dual-layout karaoke is not supported; use the line (SRT) burn for dual subtitles"
+        )
+    res_x, res_y = play_res
+    lines = [_HEADER_TEMPLATE.format(res_x=int(res_x), res_y=int(res_y))]
+    for cue in cues or []:
+        text = _norm(cue.get("text"))
+        if not text:
+            continue
+        start = float(cue["start"])
+        end = float(cue["end"])
+        if end <= start:
+            end = start + 0.1
+        lines.append(
+            f"Dialogue: 0,{_ass_time(start)},{_ass_time(end)},Default,,0,0,0,,"
+            f"{_karaoke_text(cue, text, start, end)}"
+        )
+    return "\n".join(lines) + "\n"

backend/services/media_tools.py

@@ -154,6 +154,17 @@ def bundled_dir() -> str:
     return os.path.join(media_tools_dir(), f"ffbin-{_FFBIN_COMMIT[:12]}", _platform_key())
 
 
+def _publish_bundled_on_path() -> None:
+    """Make a newly validated bundle visible to bare-name subprocess calls."""
+    directory = os.path.abspath(bundled_dir())
+    current = os.environ.get("PATH", "")
+    entries = current.split(os.pathsep) if current else []
+    if os.path.normcase(directory) in {os.path.normcase(entry) for entry in entries if entry}:
+        return
+    os.environ["PATH"] = os.pathsep.join([directory, *entries])
+    logger.info("Published the acquired media-tool directory on PATH")
+
+
 def _exe(name: str) -> str:
     return f"{name}.exe" if sys.platform == "win32" else name
 
@@ -231,12 +242,21 @@ def acquire_bundled(wait: bool = False) -> dict:
         _ops["acquire"].update(state="running", progress=0.0, error=None)
 
     if all(bundled_tool_path(t) and _binary_runs(bundled_tool_path(t)) for t in TOOLS):
+        # The bundle may have arrived after startup's one-time PATH publish
+        # (first-run acquisition is asynchronous). Make it visible to pydub
+        # and other dependencies that launch ffmpeg/ffprobe by bare name now,
+        # without requiring a backend restart (#1677).
+        _publish_bundled_on_path()
         _set_op("acquire", state="done", progress=1.0)
         return _op_snapshot()["acquire"]
 
     def _worker():
         try:
             _do_acquire()
+            # Startup cannot publish binaries which do not exist yet. The
+            # background worker must complete that second half atomically with
+            # installation so the very next synthesis can use the tools.
+            _publish_bundled_on_path()
             _set_op("acquire", state="done", progress=1.0, error=None)
             logger.info("media-tools: bundled ffmpeg/ffprobe installed at %s", bundled_dir())
         except Exception as e:

backend/services/model_manager.py

@@ -404,7 +404,23 @@ _CONFIGURED_GPU_JOB_TIMEOUT_S = GPU_JOB_TIMEOUT_S
 # CPU synthesis is healthy but substantially slower than accelerated inference.
 # Keep a separate, bounded floor so a short render on CPU is not abandoned at
 # the GPU-oriented five-minute deadline (#1588).
+#
+# #1787 review fix: an explicit OMNIVOICE_CPU_GENERATE_TIMEOUT_S must ALWAYS
+# govern CPU dispatches, even when OMNIVOICE_GENERATE_TIMEOUT_S is ALSO
+# explicit. Before this flag existed, `universal_override` below treated any
+# explicit GENERATE_TIMEOUT_S as authoritative for CPU too, so the Settings
+# panel's "CPU budget" row could be saved and silently never apply whenever
+# the "Accelerated" row was also set — the exact defect (a control that looks
+# like it works and doesn't) issue #1787 exists to remove. Setting ONLY
+# OMNIVOICE_GENERATE_TIMEOUT_S keeps its historical "universal" behavior
+# unchanged (test_explicit_universal_generate_timeout_wins_on_cpu) — nobody
+# who already relies on that single-var override loses it. The only case that
+# changes is the previously-undocumented, previously-broken combination of
+# setting BOTH: the more specific (CPU) value now wins for CPU jobs, matching
+# what a user who filled in both Settings rows was told would happen.
+_CPU_GENERATE_TIMEOUT_EXPLICIT = "OMNIVOICE_CPU_GENERATE_TIMEOUT_S" in os.environ
 CPU_JOB_TIMEOUT_S = float(os.environ.get("OMNIVOICE_CPU_GENERATE_TIMEOUT_S", "600.0"))
+_CONFIGURED_CPU_JOB_TIMEOUT_S = CPU_JOB_TIMEOUT_S
 
 # Queue-wait budget — a SEPARATE, deliberately generous clock (#1190/#1202).
 # The execution bound above must never be spent waiting in line: a job queued
@@ -520,8 +536,11 @@ def generate_timeout_s(
     on long inputs. Lives here (not in a router) so every router shares it
     without importing generation.py.
 
-    Policy: floor at the configured OMNIVOICE_GENERATE_TIMEOUT_S, plus 1s per
-    40 characters past a 1200-character free allowance — generous enough for
+    Policy: floor at the configured OMNIVOICE_GENERATE_TIMEOUT_S (accelerated
+    hosts) or OMNIVOICE_CPU_GENERATE_TIMEOUT_S (CPU hosts — the latter wins
+    for CPU whenever it is itself explicit, even if the former also is; see
+    the #1787 comment on the module-level constants), plus 1s per 40
+    characters past a 1200-character free allowance — generous enough for
     CPU-class hardware, still bounded (a wedged job is caught in minutes, not
     hours).
     """
@@ -545,7 +564,14 @@ def generate_timeout_s(
             _GENERATE_TIMEOUT_EXPLICIT
             or GPU_JOB_TIMEOUT_S != _CONFIGURED_GPU_JOB_TIMEOUT_S
         )
-        if family == "cpu" and not universal_override:
+        # An explicit (env-set, or runtime-changed the same way tests do)
+        # CPU budget is more specific than the universal override and always
+        # wins for CPU dispatches — see the #1787 comment above.
+        cpu_explicit = (
+            _CPU_GENERATE_TIMEOUT_EXPLICIT
+            or CPU_JOB_TIMEOUT_S != _CONFIGURED_CPU_JOB_TIMEOUT_S
+        )
+        if family == "cpu" and (cpu_explicit or not universal_override):
             base = CPU_JOB_TIMEOUT_S
     except Exception:
         # Device probing is advisory here; the configured universal bound is
@@ -631,7 +657,8 @@ async def run_on_gpu_pool_guarded(fn, *, what: str = "GPU job",
                                   timeout: "float | None" = None,
                                   executor=None,
                                   queue_timeout: "float | None" = None,
-                                  min_vram_gb: float = 0.0):
+                                  min_vram_gb: float = 0.0,
+                                  on_abandon=None):
     """Run blocking ``fn`` on the GPU pool, bounding **execution** — not the
     wait for a free worker.
 
@@ -659,6 +686,12 @@ async def run_on_gpu_pool_guarded(fn, *, what: str = "GPU job",
     at 0 — the default, and correct for every non-TTS job on this pool
     (reference transcribe, watermarking, dub steps) — the under-provisioned-GPU
     wording is never used, because nothing measured says it applies (#1226).
+
+    ``on_abandon`` is called once, after a job whose caller stopped waiting can
+    no longer access its inputs. A queued job that is cancelled before it
+    starts calls it immediately; a running thread calls it from ``_job``'s
+    finalizer. Normal completion never calls it. This lets request-owned temp
+    files outlive abandoned workers without delaying ordinary requests (#1668).
     """
     loop = asyncio.get_running_loop()
     ex = executor if executor is not None else _get_gpu_pool()
@@ -673,6 +706,24 @@ async def run_on_gpu_pool_guarded(fn, *, what: str = "GPU job",
     # job's model-load heartbeats (#1367). A dict, not a nonlocal: the closure
     # runs on a pool thread while the waiter reads from the event loop.
     _ident_box: dict = {}
+    _abandon_lock = threading.Lock()
+    _abandon_state = {
+        "requested": False,
+        "finished": False,
+        "callback_called": False,
+    }
+
+    def _fire_abandon_callback() -> None:
+        if on_abandon is None:
+            return
+        with _abandon_lock:
+            if _abandon_state["callback_called"]:
+                return
+            _abandon_state["callback_called"] = True
+        try:
+            on_abandon()
+        except Exception:  # noqa: BLE001 — cleanup cannot hide the pool result
+            logger.exception("%s abandon cleanup failed", _log_safe(what))
 
     def _job():
         # First thing the worker does: tell the awaiting coroutine the
@@ -689,8 +740,26 @@ async def run_on_gpu_pool_guarded(fn, *, what: str = "GPU job",
             # Idents are reused by the OS; a stale heartbeat under this ident
             # must not vouch for some future job on the same thread.
             _MODEL_LOAD_ACTIVITY.pop(threading.get_ident(), None)
+            with _abandon_lock:
+                _abandon_state["finished"] = True
+                abandoned = _abandon_state["requested"]
+            if abandoned:
+                _fire_abandon_callback()
+
+    concurrent_fut = ex.submit(_job)
+    fut = asyncio.wrap_future(concurrent_fut, loop=loop)
+
+    def _abandon() -> None:
+        # Keep the concurrent future so we can distinguish a job cancelled out
+        # of the queue from a thread that Python cannot stop once it has begun.
+        cancelled_before_start = concurrent_fut.cancel()
+        with _abandon_lock:
+            _abandon_state["requested"] = True
+            finished = _abandon_state["finished"]
+        fut.cancel()
+        if cancelled_before_start or finished:
+            _fire_abandon_callback()
 
-    fut = loop.run_in_executor(ex, _job)
     waiter = asyncio.ensure_future(started.wait())
     try:
         # Phase 1 — queue wait. Watch the future too, so a job that fails or is
@@ -703,6 +772,7 @@ async def run_on_gpu_pool_guarded(fn, *, what: str = "GPU job",
         # Caller went away (client disconnect). We stop awaiting the job, so
         # make sure its eventual result/exception is consumed rather than
         # logged as "Future exception was never retrieved".
+        _abandon()
         fut.add_done_callback(_swallow_abandoned)
         raise
     finally:
@@ -712,7 +782,7 @@ async def run_on_gpu_pool_guarded(fn, *, what: str = "GPU job",
         # Never picked up: cancel it out of the queue (a not-yet-started
         # concurrent future cancels cleanly) and report saturation, NOT a
         # too-heavy job.
-        fut.cancel()
+        _abandon()
         fut.add_done_callback(_swallow_abandoned)
         stats = gpu_pool_stats(ex)
         logger.warning(
@@ -783,14 +853,14 @@ async def run_on_gpu_pool_guarded(fn, *, what: str = "GPU job",
         # Caller went away mid-execution. The old wait_for cancelled the
         # wrapper itself; asyncio.wait does not, so do both halves here or the
         # eventual result is logged as "Future exception was never retrieved".
-        fut.cancel()
+        _abandon()
         fut.add_done_callback(_swallow_abandoned)
         raise
     except asyncio.TimeoutError as timeout_exc:
         # Parity with the old wait_for semantics: cancel the asyncio wrapper;
         # the worker thread keeps going regardless. Consume whatever it
         # eventually produces.
-        fut.cancel()
+        _abandon()
         fut.add_done_callback(_swallow_abandoned)
         # Capture the stacks BEFORE reset(): reset() replaces the executor, and
         # once the wedged thread is no longer a pool worker we can no longer
@@ -2805,6 +2875,21 @@ async def preload_model():
     if model is not None:
         return  # already loaded
 
+    # On MPS the configured ``omnivoice`` id resolves to a crash-isolated
+    # sidecar. Warming the native singleton here would put the same fatal MPS
+    # allocator risk back into the API process before the isolated engine is
+    # ever asked to synthesize.
+    try:
+        from core.device_caps import detect_host_caps
+
+        if detect_host_caps().family == "mps":
+            logger.info(
+                "Native TTS preload skipped: OmniVoice uses crash isolation on this host."
+            )
+            return
+    except Exception:  # noqa: BLE001 -- preload selection must stay best-effort
+        logger.debug("effective TTS preload selection failed", exc_info=True)
+
     # A machine lending its GPU has no local user to warm the model FOR. This
     # preload exists to make the first /generate feel instant for the person
     # sitting in front of the app; on a headless node there is nobody sitting
@@ -2894,7 +2979,10 @@ async def preload_model():
                 "The TTS model could not be loaded. Settings → Logs → Backend "
                 "has the full error."
             )
-        _set_loading("failed", detail, error=detail)
+        # `sub_stage` is a public API enum and the frontend keys failure state
+        # off `error`. Keep the human-readable word "failed" in the detail,
+        # not in the state machine (#1695).
+        _set_loading("error", detail, error=detail)
 
 def get_model_status():
     is_loaded = model is not None

backend/services/segmentation.py

@@ -79,6 +79,7 @@ class Segment:
     text: str
     speaker_id: str = "Speaker 1"
     id: str = field(default_factory=lambda: str(uuid.uuid4())[:8])
+    extra: dict = field(default_factory=dict)
 
     @property
     def duration(self) -> float:
@@ -90,6 +91,7 @@ class Segment:
 
     def to_dict(self) -> dict:
         return {
+            **self.extra,
             "id": self.id,
             "start": round(self.start, 2),
             "end": round(self.end, 2),
@@ -98,6 +100,69 @@ class Segment:
         }
 
 
+def _serialize_words(words: Sequence[Word]) -> list[dict]:
+    """Word objects → the ``{text, start, end}`` dicts persisted on segments.
+
+    Per-word timing is kept on each segment (``Segment.extra["words"]``, so
+    ``to_dict`` carries it onto the job) to drive the karaoke hardsub export.
+    """
+    return [
+        {"text": w.text, "start": round(w.start, 3), "end": round(w.end, 3)}
+        for w in words
+    ]
+
+
+def _merge_segment_extra(target: Segment, incoming: Segment, *, prepend: bool) -> None:
+    """Preserve editor metadata when cleanup folds ``incoming`` into ``target``."""
+    # Word lists must CONCATENATE in text order (the setdefault below would
+    # otherwise adopt the incoming list wholesale when the target has none,
+    # then double it). Capture both sides before setdefault runs.
+    raw_target_words = target.extra.get("words")
+    raw_incoming_words = incoming.extra.get("words")
+    for key, value in incoming.extra.items():
+        target.extra.setdefault(key, value)
+    target_words = raw_target_words if isinstance(raw_target_words, list) else []
+    incoming_words = raw_incoming_words if isinstance(raw_incoming_words, list) else []
+    if target_words or incoming_words:
+        target.extra["words"] = (
+            incoming_words + target_words if prepend else target_words + incoming_words
+        )
+
+    def joined(left: object, right: object) -> str:
+        return _clean(f"{left or ''} {right or ''}")
+
+    target_original = target.extra.get("text_original")
+    incoming_original = incoming.extra.get("text_original")
+    if target_original is not None or incoming_original is not None:
+        target.extra["text_original"] = (
+            joined(incoming_original, target_original)
+            if prepend
+            else joined(target_original, incoming_original)
+        )
+
+    raw_target_translations = target.extra.get("translations")
+    raw_incoming_translations = incoming.extra.get("translations")
+    target_translations = raw_target_translations if isinstance(raw_target_translations, dict) else {}
+    incoming_translations = (
+        raw_incoming_translations if isinstance(raw_incoming_translations, dict) else {}
+    )
+    if target_translations or incoming_translations:
+        merged = {}
+        languages = {
+            *target_translations.keys(),
+            *incoming_translations.keys(),
+        }
+        for language in languages:
+            target_text = target_translations.get(language)
+            incoming_text = incoming_translations.get(language)
+            merged[language] = (
+                joined(incoming_text, target_text)
+                if prepend
+                else joined(target_text, incoming_text)
+            )
+        target.extra["translations"] = merged
+
+
 def _clean(text: str) -> str:
     return _WS.sub(" ", (text or "").strip())
 
@@ -191,7 +256,10 @@ def _build_segments_from_words(words: Sequence[Word]) -> List[Segment]:
         if not text:
             buf = []
             return
-        segments.append(Segment(start=buf_start, end=buf[-1].end, text=text))
+        segments.append(Segment(
+            start=buf_start, end=buf[-1].end, text=text,
+            extra={"words": _serialize_words(buf)},
+        ))
         buf = []
         if not force:
             buf_start = 0.0
@@ -249,6 +317,7 @@ def _build_segments_from_words(words: Sequence[Word]) -> List[Segment]:
                     start=buf_start,
                     end=left_buf[-1].end,
                     text=_clean(" ".join(x.text for x in left_buf)),
+                    extra={"words": _serialize_words(left_buf)},
                 ))
                 buf = list(right_buf)
                 buf_start = right_buf[0].start
@@ -317,12 +386,14 @@ def _merge_short(segments: List[Segment]) -> List[Segment]:
                 i += 1
                 continue
             if target is prev:
+                _merge_segment_extra(prev, s, prepend=False)
                 prev.text = _clean(prev.text + " " + s.text)
                 prev.end = max(prev.end, s.end)
                 segments.pop(i)
                 did_merge = True
                 continue
             if target is nxt:
+                _merge_segment_extra(nxt, s, prepend=True)
                 nxt.text = _clean(s.text + " " + nxt.text)
                 nxt.start = min(nxt.start, s.start)
                 segments.pop(i)
@@ -360,6 +431,7 @@ def _stitch_adjacent_shorts(segments: List[Segment]) -> List[Segment]:
                 and b.duration <= STITCH_DUR
                 and combined_dur <= MAX_DUR
             ):
+                _merge_segment_extra(a, b, prepend=False)
                 a.text = _clean(a.text + " " + b.text)
                 a.end = b.end
                 segments.pop(i + 1)
@@ -386,6 +458,11 @@ def clean_up_segments(segments: List[dict]) -> List[dict]:
                 text=_clean(str(s.get("text", ""))),
                 speaker_id=str(s.get("speaker_id") or "Speaker 1"),
                 id=str(s.get("id") or uuid.uuid4().hex[:8]),
+                extra={
+                    key: value
+                    for key, value in s.items()
+                    if key not in {"id", "start", "end", "text", "speaker_id"}
+                },
             ))
         except (TypeError, ValueError):
             continue
@@ -429,11 +506,23 @@ def _apply_scene_cuts(segments: List[Segment], scene_cuts: Iterable[float]) -> L
                 or (remaining.end - cut) < MIN_DUR
             ):
                 continue
+            # Segment text is the joined word texts, so a whitespace-boundary
+            # text split maps exactly onto a word-count split of the list.
+            words = remaining.extra.get("words")
+            left_extra: dict = {}
+            right_extra: dict = {}
+            if isinstance(words, list) and words:
+                n_left = len(left_text.split())
+                if n_left and len(words) > n_left:
+                    left_extra = {"words": words[:n_left]}
+                    right_extra = {"words": words[n_left:]}
             out.append(Segment(
                 start=remaining.start, end=cut, text=left_text, speaker_id=remaining.speaker_id,
+                extra=left_extra,
             ))
             remaining = Segment(
                 start=cut, end=remaining.end, text=right_text, speaker_id=remaining.speaker_id,
+                extra=right_extra,
             )
         out.append(remaining)
     return out
@@ -665,6 +754,10 @@ def _resplit_core(
             piece["text"] = text
             piece["start"] = s0 if k == 0 else ws[0].start
             piece["end"] = s1 if k == n_runs - 1 else ws[-1].end
+            # dict(seg) copied the WHOLE segment's word list into every piece;
+            # each piece keeps only its own run's words (karaoke burn-in).
+            if "words" in piece:
+                piece["words"] = _serialize_words(ws)
             if label:
                 piece["speaker_id"] = label
             if piece_no > 0:

backend/services/settings_store.py

@@ -254,6 +254,31 @@ def get_text(key: str, default: Optional[str] = None) -> Optional[str]:
         return default
 
 
+def get_text_state(key: str) -> tuple[bool, str]:
+    """Return ``(is_present, value)`` without hiding storage failures.
+
+    Rollback snapshots must distinguish a missing row from an unreadable
+    database.  ``get_text`` deliberately collapses those cases for ordinary
+    preference reads, so transactional callers use this strict variant.
+    """
+    if key == _TOKEN_KEY or key.startswith(_SECRET_PREFIX):
+        raise ValueError(
+            "get_text_state refuses to read an encrypted secret row; "
+            "use get_hf_token()/get_secret() for secrets"
+        )
+    from core.db import db_conn
+
+    with db_conn() as conn:
+        row = conn.execute(
+            "SELECT value FROM settings WHERE key = ?", (key,)
+        ).fetchone()
+    if row is None:
+        return False, ""
+    if row[0] is None:
+        return True, ""
+    return True, str(row[0])
+
+
 def set_text(key: str, value: str) -> None:
     """Persist a non-encrypted text value into the settings table.
 
@@ -274,6 +299,19 @@ def set_text(key: str, value: str) -> None:
         )
 
 
+def clear_text(key: str) -> None:
+    """Remove a non-encrypted text setting, preserving a missing-row default."""
+    if key == _TOKEN_KEY or key.startswith(_SECRET_PREFIX):
+        raise ValueError(
+            "clear_text refuses to delete an encrypted secret row; "
+            "use clear_hf_token()/clear_secret() for secrets"
+        )
+    from core.db import db_conn
+
+    with db_conn() as conn:
+        conn.execute("DELETE FROM settings WHERE key = ?", (key,))
+
+
 # ── Phase 4 Plan 04-01 (GGUF-04): per-engine quant override ────────────────
 #
 # Settings row "gguf_quant_override" holds either:

backend/services/sherpa_dictation.py

@@ -273,6 +273,13 @@ def sherpa_available() -> tuple[bool, str]:
         return False, f"sherpa-onnx unavailable ({type(e).__name__}): {e}"
 
 
+def _usable_model_file(path: str) -> bool:
+    try:
+        return os.path.isfile(path) and os.path.getsize(path) > 0
+    except OSError:
+        return False
+
+
 def _resolve_model_dir(spec: SherpaModelSpec, *, download: bool = True) -> str:
     """Return the local directory containing this model's ONNX assets.
 
@@ -281,41 +288,114 @@ def _resolve_model_dir(spec: SherpaModelSpec, *, download: bool = True) -> str:
     Restricts the fetch to the exact int8 assets we pin via ``allow_patterns``
     so we never pull the bundled fp32 weights or test wavs.
     """
-    from huggingface_hub import constants as hf_constants
     from huggingface_hub import snapshot_download
-    from services.hf_revisions import installed_revision, revision_for
+    from services.hf_revisions import revision_for
 
     wanted = list(spec.files.values())
+    cache_dir = _live_hub_cache_dir()
     # Probe the revision an existing installation actually resolved.  Older
     # releases followed ``main`` and may therefore have a different snapshot;
     # retaining it preserves offline upgrades.  Any network fetch still uses
     # the reviewed immutable pin.
-    installed = installed_revision(spec.repo_id, hf_constants.HF_HUB_CACHE)
-    try:
-        return snapshot_download(
-            repo_id=spec.repo_id,
-            revision=installed,
-            local_files_only=True,
-            allow_patterns=wanted,
-        )
-    except Exception:
-        if not download:
-            raise
+    installed = _installed_snapshot(spec)
+    if installed:
+        return installed
+    if not download:
+        raise FileNotFoundError(f"No complete cached snapshot for {spec.repo_id}")
+
+    # A Windows cache can retain a snapshot entry whose target blob vanished,
+    # or a zero-byte ONNX placeholder left by an interrupted download.  Hub may
+    # then treat that entry as already materialized and return the same broken
+    # snapshot.  Repair those entries before asking for another download so the
+    # recognizer never receives a path to a file that does not resolve (#1733).
+    from services.hf_cache_repair import (
+        find_dangling_entries,
+        repair_repo_cache,
+        repo_cache_dir,
+    )
+
+    if find_dangling_entries(repo_cache_dir(spec.repo_id, cache_dir)):
+        repair = repair_repo_cache(spec.repo_id, cache_dir)
+        installed = _installed_snapshot(spec)
+        if installed:
+            return installed
+        if not repair.get("ok"):
+            logger.warning(
+                "sherpa dictation: cache repair for %s failed: %s",
+                spec.repo_id,
+                repair.get("error") or repair.get("outcome") or "unknown error",
+            )
+
     logger.info("sherpa dictation: downloading %s on first use", spec.repo_id)
-    return snapshot_download(
+    snapshot = snapshot_download(
         repo_id=spec.repo_id,
         revision=revision_for(spec.repo_id),
         allow_patterns=wanted,
+        cache_dir=cache_dir,
+    )
+    missing = [
+        name for name in wanted
+        if not _usable_model_file(os.path.join(snapshot, name))
+    ]
+    if not missing:
+        return snapshot
+
+    # Verify after the Hub reports success.  This catches hosts where a broken
+    # snapshot entry short-circuits snapshot_download.  The generic repair
+    # removes only broken entries, preserves blobs, and retries the immutable
+    # installed revision.
+    repair = repair_repo_cache(spec.repo_id, cache_dir)
+    installed = _installed_snapshot(spec)
+    if installed:
+        return installed
+    detail = repair.get("error") or repair.get("outcome") or "repair did not restore them"
+    raise FileNotFoundError(
+        f"Sherpa model cache is incomplete for {spec.repo_id}; missing "
+        f"{', '.join(missing)}. Cache repair failed: {detail}. Reinstall this "
+        "model from Model Catalogue."
     )
 
 
+def _live_hub_cache_dir() -> str:
+    """The effective hub root, evaluated after Settings restores the env."""
+    direct = os.environ.get("HF_HUB_CACHE") or os.environ.get("HUGGINGFACE_HUB_CACHE")
+    if direct:
+        return os.path.expanduser(direct)
+    home = os.environ.get("HF_HOME") or os.path.expanduser("~/.cache/huggingface")
+    return os.path.join(os.path.expanduser(home), "hub")
+
+
+def _installed_snapshot(spec: SherpaModelSpec) -> str | None:
+    """Complete snapshot for the recorded revision in the live cache."""
+    from services.hf_revisions import installed_revision
+
+    cache_dir = _live_hub_cache_dir()
+    revision = installed_revision(spec.repo_id, cache_dir)
+    snapshot = os.path.join(
+        cache_dir,
+        "models--" + spec.repo_id.replace("/", "--"),
+        "snapshots",
+        revision,
+    )
+    if all(
+        _usable_model_file(os.path.join(snapshot, filename))
+        for filename in spec.files.values()
+    ):
+        return snapshot
+    return None
+
+
 def is_installed(spec: SherpaModelSpec) -> bool:
-    """True if every pinned asset is already present in the HF cache."""
-    try:
-        d = _resolve_model_dir(spec, download=False)
-    except Exception:
-        return False
-    return all(os.path.isfile(os.path.join(d, f)) for f in spec.files.values())
+    """True if the recorded cached snapshot contains every pinned asset.
+
+    Do not use ``snapshot_download(local_files_only=True)`` for this probe.
+    ``huggingface_hub.constants.HF_HUB_CACHE`` is fixed when that module is
+    first imported, while VoiceStudio can restore its cache directory later
+    from the durable user settings. Resolve the live root and the recorded
+    revision ourselves so readiness and loading cannot disagree after a cache
+    move, desktop relaunch, or stale snapshot (#1707).
+    """
+    return _installed_snapshot(spec) is not None
 
 
 # ── Recognizers ──────────────────────────────────────────────────────────────

backend/services/sidecar_install.py

@@ -60,6 +60,7 @@ from pathlib import Path
 from typing import Callable, Optional
 
 from core.config import DATA_DIR
+from core.contained_subprocess import OwnedPopen, WindowsJobPopen, spawn_owned
 
 logger = logging.getLogger("omnivoice.sidecar_install")
 
@@ -115,6 +116,11 @@ class SidecarSpec:
     weights_config_names: tuple[str, ...] = ("config.yaml",)
     docs_path: str = "docs/engines"         # where the manual-install fallback lives
     required_bytes: int = 12 * _GIB    # conservative source+venv+weights estimate for preflight
+    weights_bytes: Optional[int] = None
+    dependency_bytes: Optional[int] = None
+    potentially_shared_bytes: Optional[int] = None
+    temporary_free_bytes: Optional[int] = None
+    disk_confidence: str = "unknown"
     # Called after a successful install/uninstall so the engine's memoised
     # venv resolution re-probes (import inside the lambda — never at module load).
     invalidate: Callable[[], None] = field(default=lambda: None)
@@ -156,6 +162,11 @@ SPECS: dict[str, SidecarSpec] = {
         # ~6 GB weights. Deliberately conservative; the preflight subtracts
         # whatever a partial install already put on disk.
         required_bytes=12 * _GIB,
+        weights_bytes=6 * _GIB,
+        dependency_bytes=6 * _GIB,
+        potentially_shared_bytes=None,
+        temporary_free_bytes=12 * _GIB,
+        disk_confidence="estimated",
         invalidate=_indextts_invalidate,
         installed_probe=_indextts_installed,
     ),
@@ -311,6 +322,25 @@ def _dir_size_bytes(path: Path) -> int:
     return total
 
 
+def _preserved_install_bytes(spec: SidecarSpec, checkout: Path) -> tuple[int, int]:
+    """Return bytes preserved for the final install and dependency peak.
+
+    Resumable weights reduce the final download requirement, but they do not
+    reduce uv's separate environment-build peak. Only source and a usable
+    existing venv count against that peak.
+    """
+    if not _source_present(spec, checkout):
+        return 0, 0
+
+    weights_dir = checkout / spec.weights_subdir
+    weights = _dir_size_bytes(weights_dir) if spec.weights_repo_id else 0
+    venv_dir = checkout / ".venv"
+    venv = _dir_size_bytes(venv_dir)
+    source = max(0, _dir_size_bytes(checkout) - weights - venv)
+    usable_venv = venv if _venv_python(venv_dir).is_file() else 0
+    return source + usable_venv + weights, source + usable_venv
+
+
 def disk_free_bytes(path: Path) -> int:
     """Free bytes on the volume backing *path* (nearest existing ancestor).
     Never raises; 0 when the volume can't be probed."""
@@ -333,8 +363,17 @@ def disk_space_error(spec: SidecarSpec) -> Optional[str]:
     root = managed_root(spec)
     # A preserved predecessor is not a partial copy of the new install: the
     # upgrade needs its full space until the new sidecar is verified.
-    already = _dir_size_bytes(managed_checkout(spec))
-    remaining = max(0, spec.required_bytes - already)
+    checkout = managed_checkout(spec)
+    # Credit only bytes the later steps preserve. An invalid layout or revision
+    # marker makes _step_fetch_source delete the whole checkout.
+    preserved, dependency_peak_credit = _preserved_install_bytes(spec, checkout)
+    remaining = max(0, spec.required_bytes - preserved)
+    if spec.temporary_free_bytes is not None:
+        # Resumable model weights are unrelated to uv's dependency-build peak.
+        remaining = max(
+            remaining,
+            max(0, spec.temporary_free_bytes - dependency_peak_credit),
+        )
     free = disk_free_bytes(root)
     if free <= 0:
         return None  # can't probe → never block on missing information
@@ -998,6 +1037,11 @@ def _step_persist(spec: SidecarSpec, job: dict) -> None:
 # ── Subprocess runner with live log capture ────────────────────────────────
 
 
+def _install_containment_kwargs() -> dict:
+    """Nested process-group/Job ownership is supplied by ``spawn_owned``."""
+    return {}
+
+
 def _run_logged(job: dict, argv: list[str], *, timeout: float,
                 env: "dict[str, str] | None" = None) -> int:
     """Run *argv*, streaming combined stdout+stderr lines into the job log.
@@ -1012,13 +1056,12 @@ def _run_logged(job: dict, argv: list[str], *, timeout: float,
     killed child — a blocking ``for line in proc.stdout`` on this thread
     would hang past the timeout waiting for pipe EOF.
     """
-    popen_kwargs: dict = {}
-    if os.name == "posix":
-        # New session → we can kill the whole process group on timeout
-        # instead of only the direct child.
-        popen_kwargs["start_new_session"] = True
+    # ``spawn_owned`` creates the local timeout group/Job before the operation
+    # starts. POSIX links it to backend death through a control pipe; Windows
+    # retains a kill-on-close Job handle in this backend process.
+    popen_kwargs = _install_containment_kwargs()
     try:
-        proc = subprocess.Popen(
+        proc = spawn_owned(
             argv,
             stdout=subprocess.PIPE,
             stderr=subprocess.STDOUT,
@@ -1053,29 +1096,18 @@ def _run_logged(job: dict, argv: list[str], *, timeout: float,
 
 
 def _kill_tree(proc: "subprocess.Popen") -> None:
-    """Kill the child and its whole process tree, on every platform.
-
-    POSIX: the child was started in its own session, so SIGKILL the group.
-    Windows: ``proc.kill()`` only terminates the direct child — a git/uv
-    helper it spawned would keep running (and writing into the checkout)
-    past our timeout — so use ``taskkill /T`` to fell the tree.
-    """
-    if os.name == "posix":
-        import signal
+    """Kill an operation through its stable nested group/Job owner."""
+    if isinstance(proc, (OwnedPopen, WindowsJobPopen)):
+        # The retained supervisor/process-group or nested Job is the stable
+        # per-operation owner.  Do not fall back to a direct PID kill.
+        proc.kill()
         try:
-            os.killpg(proc.pid, signal.SIGKILL)
+            proc.wait(timeout=5)
+        except subprocess.TimeoutExpired:
             return
-        except (ProcessLookupError, PermissionError, OSError):
-            pass  # group already gone / not ours — fall through to plain kill
-    else:  # Windows
-        try:
-            subprocess.run(
-                ["taskkill", "/F", "/T", "/PID", str(proc.pid)],
-                capture_output=True, timeout=15,
-            )
-            return
-        except (OSError, subprocess.SubprocessError):
-            pass  # taskkill unavailable/failed — fall through to plain kill
+        return
+    # A test double or a legacy caller without the nested owner can only be
+    # stopped through its stable direct-process handle.
     try:
         proc.kill()
     except OSError:

backend/services/subprocess_backend.py

@@ -32,9 +32,9 @@ Threat-model summary (see Plan 02-01 frontmatter):
               AUTH-05 installed (``HFTokenRedactor``) on the root logger.
     T-02-04 — compromised sidecar emitting unexpected ops: parent allowlist
               ``PARENT_INBOUND_OPS`` rejects everything else.
-    T-02-05 — Tauri group-kill scope: ``start_new_session=True`` on Unix
-              and ``CREATE_NEW_PROCESS_GROUP`` on Windows isolate the
-              sidecar's process group.
+    T-02-05 — nested containment: a retained POSIX supervisor process group or
+              Windows Job owns each engine operation, while still permitting
+              independent timeout teardown and cleanup on backend death.
 """
 from __future__ import annotations
 
@@ -56,6 +56,7 @@ from typing import Optional
 import numpy as np
 import torch
 
+from core.contained_subprocess import spawn_owned
 from services.tts_backend import TTSBackend
 
 logger = logging.getLogger("omnivoice.subprocess_backend")
@@ -362,6 +363,7 @@ class SubprocessBackend(TTSBackend):
     # be a different class object from the one the subclass closed over.
     # A duck-typed marker survives that.
     _is_subprocess_isolated: bool = True
+    spawn_ready_timeout_s: float = SPAWN_READY_TIMEOUT_S
 
     # Generation happens in the sidecar: parent-side accelerator counters
     # can't see its allocations (see TTSBackend.runs_out_of_process).
@@ -470,13 +472,6 @@ class SubprocessBackend(TTSBackend):
             "env": env,
             "bufsize": 0,  # unbuffered binary pipes
         }
-        # Process-group isolation so the Tauri lib.rs group-kill in shutdown
-        # doesn't escape into other children. See T-02-05.
-        if sys.platform == "win32":
-            kwargs["creationflags"] = subprocess.CREATE_NEW_PROCESS_GROUP
-        else:
-            kwargs["start_new_session"] = True
-
         # `venv_python()` resolves the engine's interpreter, and on a cold
         # first run that is not cheap: it spawns each candidate to import the
         # engine (bounded, but tens of seconds on a slow disk), and if none is
@@ -509,7 +504,7 @@ class SubprocessBackend(TTSBackend):
             self.id, Path(python_path).name, Path(script_path).name,
         )
         try:
-            self._proc = subprocess.Popen([python_path, script_path], **kwargs)
+            self._proc = spawn_owned([python_path, script_path], **kwargs)
         except OSError as exc:
             raise InvalidBinaryError(
                 python_path,
@@ -530,7 +525,7 @@ class SubprocessBackend(TTSBackend):
         # Block on the ready handshake. A sidecar that fails to emit ready
         # within SPAWN_READY_TIMEOUT_S is killed and the failure is raised.
         try:
-            frame = self._recv_with_timeout(SPAWN_READY_TIMEOUT_S)
+            frame = self._recv_with_timeout(self.spawn_ready_timeout_s)
         except Exception:
             self._force_kill()
             raise

backend/services/tts_backend.py

@@ -407,6 +407,13 @@ class TTSBackend(ABC):
     # entirely (it drives the shared model_manager singleton).
     _MODEL_ATTRS: tuple[str, ...] = ("_model", "_tts")
 
+    def execution_evidence_loaded(self) -> bool:
+        """Whether this instance has live model state worth reporting."""
+        if self.runs_out_of_process:
+            proc = getattr(self, "_proc", None)
+            return proc is not None and proc.poll() is None
+        return any(getattr(self, attr, None) is not None for attr in self._MODEL_ATTRS)
+
     def unload(self) -> None:
         """Release the heavy model this backend holds, and free device caches.
 
@@ -669,7 +676,7 @@ class OmniVoiceBackend(TTSBackend):
 
     id = "omnivoice"
     display_name = "VoiceStudio (k2-fsa/OmniVoice, 600+ languages)"
-    gpu_compat = ("cuda", "mps", "cpu")
+    gpu_compat = ("cuda", "rocm", "mps", "cpu")
     # Derived from the pool's own per-job budget (_GPU_VRAM_PER_JOB_GB = 5.0 in
     # model_manager, itself measured from the ~1.6 GB forward + autoregressive
     # decode and the co-loaded WhisperX on the clone path), plus room for the
@@ -2392,12 +2399,15 @@ def list_backends() -> list[dict]:
     # Routing is host-aware but the host caps are constant per process, so probe
     # ONCE here and resolve each engine's effective device against the same caps.
     from core.device_caps import detect_host_caps
+    from services.engine_disk_usage import disk_summary_for
+    from services.engine_evidence import snapshot as execution_snapshot
     from services.engine_routing import routing_fields
     caps = detect_host_caps()
     installable = _sidecar_installable_ids()
 
     out: list[dict] = []
     for bid, cls in _REGISTRY.items():
+        cls = _effective_backend_class(bid, cls, caps.family)
         try:
             ok, msg = cls.is_available()
         except Exception:
@@ -2424,6 +2434,12 @@ def list_backends() -> list[dict]:
         # descriptor, not a bool, so report None (= model-dependent) there
         # instead of an always-truthy false positive.
         _clone = getattr(cls, "supports_cloning", True)
+        routing = routing_fields(gpu_compat, caps, getattr(cls, "min_vram_gb", 0.0))
+        loaded_instance = None
+        if _active_instance_id == bid:
+            loaded_instance = _active_instance
+        if loaded_instance is None:
+            loaded_instance = _ENGINE_INSTANCES.get(cls)
         out.append({
             "id": bid,
             "display_name": cls.display_name,
@@ -2443,6 +2459,7 @@ def list_backends() -> list[dict]:
             # in-app (Settings renders an Install button instead of leading
             # with the manual setup snippet).
             "one_click_install": bid in installable,
+            "disk_usage": disk_summary_for(bid),
             "last_error": _LAST_ERRORS.get(bid),
             "isolation_mode": isolation,
             "gpu_compat": list(gpu_compat),
@@ -2450,7 +2467,14 @@ def list_backends() -> list[dict]:
             "min_vram_gb": getattr(cls, "min_vram_gb", 0.0) or None,
             # effective_device / routing_status / routing_reason (scrubbed);
             # the reason now also carries the under-provisioned-GPU caveat.
-            **routing_fields(gpu_compat, caps, getattr(cls, "min_vram_gb", 0.0)),
+            **routing,
+            "execution_evidence": execution_snapshot(
+                engine_id=bid,
+                engine_cls=cls,
+                instance=loaded_instance,
+                routing=routing,
+                caps=caps,
+            ),
         })
         # #981: mlx-audio multiplexes 7+ curated models behind one backend id
         # — surface the roster + the currently-active pick so Settings can
@@ -2471,10 +2495,29 @@ def list_backends() -> list[dict]:
     return out
 
 
+def _effective_backend_class(
+    backend_id: str,
+    backend_cls: type[TTSBackend],
+    host_family: str | None = None,
+) -> type[TTSBackend]:
+    """Resolve host-specific containment without changing the configured id."""
+    if backend_id != "omnivoice":
+        return backend_cls
+    if host_family is None:
+        from core.device_caps import detect_host_caps
+
+        host_family = detect_host_caps().family
+    if host_family != "mps":
+        return backend_cls
+    from engines.omnivoice_subprocess import OmniVoiceMPSSubprocessBackend
+
+    return OmniVoiceMPSSubprocessBackend
+
+
 def get_backend_class(backend_id: str) -> type[TTSBackend]:
     if backend_id not in _REGISTRY:
         raise ValueError(f"Unknown TTS backend: {backend_id!r}. Known: {list(_REGISTRY)}")
-    return _REGISTRY[backend_id]
+    return _effective_backend_class(backend_id, _REGISTRY[backend_id])
 
 
 def cloning_capable_engine_ids() -> list[str]:

backend/speech_client/__init__.py

@@ -0,0 +1 @@
+"""Dependency-free client for VoiceStudio's local speech platform."""

backend/speech_client/__main__.py

@@ -0,0 +1,278 @@
+"""CLI/module bridge for terminals, editor extensions, and agent hooks.
+
+The desktop app must be running for native dictation control. Batch
+transcription can also target a standalone or remote VoiceStudio backend.
+"""
+from __future__ import annotations
+
+import argparse
+import ipaddress
+import json
+import mimetypes
+import os
+from pathlib import Path
+import secrets
+import sys
+from typing import Any
+from urllib import error, request
+from urllib.parse import urlsplit
+
+DEFAULT_CONTROL_URL = "http://127.0.0.1:3902"
+DEFAULT_ENGINE_URL = "http://127.0.0.1:3900"
+
+
+class SpeechClientError(RuntimeError):
+    pass
+
+
+class _RejectCredentialRedirect(request.HTTPRedirectHandler):
+    def redirect_request(self, req, fp, code, msg, headers, newurl):  # noqa: ARG002
+        raise SpeechClientError("VoiceStudio refused a credentialed redirect")
+
+
+def _join_url(base_url: str, path: str) -> str:
+    return f"{base_url.rstrip('/')}/{path.lstrip('/')}"
+
+
+def _decode_error(exc: error.HTTPError) -> str:
+    try:
+        body = exc.read().decode("utf-8", errors="replace")
+    except Exception:
+        body = ""
+    try:
+        detail = json.loads(body)
+    except (TypeError, json.JSONDecodeError):
+        detail = body.strip()
+    return f"HTTP {exc.code}: {detail or exc.reason}"
+
+
+def _is_loopback_host(host: str | None) -> bool:
+    if not host:
+        return False
+    if host.lower() == "localhost":
+        return True
+    try:
+        return ipaddress.ip_address(host).is_loopback
+    except ValueError:
+        return False
+
+
+def _open(req: request.Request, timeout: float = 300.0) -> tuple[bytes, str]:
+    target = urlsplit(req.full_url)
+    scheme = target.scheme.lower()
+    if scheme not in {"http", "https"}:
+        raise SpeechClientError("VoiceStudio URLs must use http:// or https://")
+    credentialed = bool(req.get_header("Authorization"))
+    if credentialed and scheme != "https" and not _is_loopback_host(target.hostname):
+        raise SpeechClientError("Remote VoiceStudio credentials require https://")
+    try:
+        opener = (
+            request.build_opener(_RejectCredentialRedirect())
+            if credentialed
+            else request.build_opener()
+        )
+        with opener.open(req, timeout=timeout) as response:  # noqa: S310
+            return response.read(), response.headers.get("Content-Type", "")
+    except error.HTTPError as exc:
+        raise SpeechClientError(_decode_error(exc)) from exc
+    except error.URLError as exc:
+        raise SpeechClientError(f"VoiceStudio is unavailable: {exc.reason}") from exc
+
+
+def _json_request(method: str, url: str, payload: Any | None = None) -> Any:
+    data = None if payload is None else json.dumps(payload).encode("utf-8")
+    headers = {"Accept": "application/json"}
+    if data is not None:
+        headers["Content-Type"] = "application/json"
+    body, _ = _open(request.Request(url, data=data, headers=headers, method=method), timeout=10.0)
+    try:
+        return json.loads(body)
+    except json.JSONDecodeError as exc:
+        raise SpeechClientError("VoiceStudio returned invalid JSON") from exc
+
+
+def _encode_multipart(
+    *,
+    filename: str,
+    audio: bytes,
+    fields: dict[str, str],
+    boundary: str | None = None,
+) -> tuple[bytes, str]:
+    boundary = boundary or f"voicestudio-{secrets.token_hex(16)}"
+    marker = boundary.encode("ascii")
+    parts: list[bytes] = []
+    for name, value in fields.items():
+        parts.extend(
+            [
+                b"--" + marker + b"\r\n",
+                f'Content-Disposition: form-data; name="{name}"\r\n\r\n'.encode(),
+                value.encode("utf-8"),
+                b"\r\n",
+            ]
+        )
+    safe_filename = Path(filename).name.replace('"', "") or "audio.wav"
+    content_type = mimetypes.guess_type(safe_filename)[0] or "application/octet-stream"
+    if Path(safe_filename).suffix.lower() in {".wav", ".wave"}:
+        content_type = "audio/wav"
+    parts.extend(
+        [
+            b"--" + marker + b"\r\n",
+            (
+                'Content-Disposition: form-data; name="file"; '
+                f'filename="{safe_filename}"\r\n'
+            ).encode(),
+            f"Content-Type: {content_type}\r\n\r\n".encode(),
+            audio,
+            b"\r\n--" + marker + b"--\r\n",
+        ]
+    )
+    return b"".join(parts), f"multipart/form-data; boundary={boundary}"
+
+
+def _control(args: argparse.Namespace, action: str) -> int:
+    method = "GET" if action in {"status", "capabilities"} else "POST"
+    path = {
+        "status": "/v1/status",
+        "capabilities": "/v1/capabilities",
+        "start": "/v1/dictation/start",
+        "stop": "/v1/dictation/stop",
+        "toggle": "/v1/dictation/toggle",
+    }[action]
+    result = _json_request(method, _join_url(args.control_url, path))
+    print(json.dumps(result, ensure_ascii=False, indent=2))
+    return 0
+
+
+def _read_audio(path: str, stdin_filename: str) -> tuple[bytes, str]:
+    if path == "-":
+        return sys.stdin.buffer.read(), stdin_filename
+    audio_path = Path(path)
+    try:
+        return audio_path.read_bytes(), audio_path.name
+    except OSError as exc:
+        display_name = path.replace("\\", "/").rsplit("/", 1)[-1] or "audio input"
+        reason = exc.strerror or type(exc).__name__
+        raise SpeechClientError(f"could not read '{display_name}': {reason}") from exc
+
+
+def _response_text(body: bytes, content_type: str) -> str:
+    decoded = body.decode("utf-8", errors="replace")
+    if "json" not in content_type.lower():
+        return decoded
+    try:
+        payload = json.loads(decoded)
+    except json.JSONDecodeError:
+        return decoded
+    if isinstance(payload, dict) and isinstance(payload.get("text"), str):
+        return payload["text"]
+    return decoded
+
+
+def _transcribe(args: argparse.Namespace) -> int:
+    audio, filename = _read_audio(args.audio, args.stdin_filename)
+    fields = {
+        "model": args.model,
+        "response_format": args.response_format,
+    }
+    if args.language:
+        fields["language"] = args.language
+    body, content_type = _encode_multipart(filename=filename, audio=audio, fields=fields)
+    headers = {"Content-Type": content_type, "Accept": "application/json, text/plain"}
+    api_key = os.environ.get("OMNIVOICE_API_KEY", "").strip()
+    if api_key:
+        headers["Authorization"] = f"Bearer {api_key}"
+
+    output_session_id = None
+    if args.insert:
+        session = _json_request(
+            "POST", _join_url(args.control_url, "/v1/output/sessions")
+        )
+        output_session_id = session["session_id"]
+
+    session_needs_cleanup = output_session_id is not None
+    try:
+        response_body, response_type = _open(
+            request.Request(
+                _join_url(args.engine_url, "/v1/audio/transcriptions"),
+                data=body,
+                headers=headers,
+                method="POST",
+            )
+        )
+        if output_session_id is not None:
+            _json_request(
+                "POST",
+                _join_url(
+                    args.control_url,
+                    f"/v1/output/sessions/{output_session_id}/insert",
+                ),
+                {"text": _response_text(response_body, response_type)},
+            )
+            session_needs_cleanup = False
+    finally:
+        if session_needs_cleanup:
+            try:
+                _json_request(
+                    "DELETE",
+                    _join_url(args.control_url, f"/v1/output/sessions/{output_session_id}"),
+                )
+            except Exception:  # noqa: BLE001
+                # Best-effort cleanup must not replace the original failure or
+                # KeyboardInterrupt that brought control into this finally.
+                pass
+
+    sys.stdout.buffer.write(response_body)
+    if response_body and not response_body.endswith(b"\n"):
+        sys.stdout.buffer.write(b"\n")
+    return 0
+
+
+def _parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(
+        prog="voicestudio-speech",
+        description="Control and consume VoiceStudio's local speech platform.",
+    )
+    parser.add_argument(
+        "--control-url",
+        default=os.environ.get("VOICESTUDIO_SPEECH_URL", DEFAULT_CONTROL_URL),
+    )
+    parser.add_argument(
+        "--engine-url",
+        default=os.environ.get("VOICESTUDIO_URL", DEFAULT_ENGINE_URL),
+    )
+    subparsers = parser.add_subparsers(dest="command", required=True)
+    for command in ("status", "capabilities", "start", "stop", "toggle"):
+        subparsers.add_parser(command)
+
+    transcribe = subparsers.add_parser("transcribe")
+    transcribe.add_argument("audio", help="audio file, or - for stdin")
+    transcribe.add_argument("--stdin-filename", default="audio.wav")
+    transcribe.add_argument("--model", default="whisper-1")
+    transcribe.add_argument("--language")
+    transcribe.add_argument(
+        "--format",
+        dest="response_format",
+        choices=("json", "text", "verbose_json", "srt", "vtt"),
+        default="text",
+    )
+    transcribe.add_argument(
+        "--insert",
+        action="store_true",
+        help="insert the result into the app focused when this command starts",
+    )
+    return parser
+
+
+def main(argv: list[str] | None = None) -> int:
+    args = _parser().parse_args(argv)
+    try:
+        if args.command == "transcribe":
+            return _transcribe(args)
+        return _control(args, args.command)
+    except (SpeechClientError, KeyError) as exc:
+        print(f"voicestudio-speech: {exc}", file=sys.stderr)
+        return 2
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

backend/tests/test_asr_transcribe_timeout.py

@@ -11,6 +11,7 @@ that the error message tells the user what to do.
 import asyncio
 import os
 import sys
+import threading
 import time
 
 import pytest
@@ -75,17 +76,71 @@ def test_fast_transcribe_passes_through():
     pool.shutdown(wait=True)
 
 
+def test_timeout_defers_abandon_cleanup_until_running_worker_finishes():
+    """A timed-out native worker may still be reading request-owned inputs."""
+    pool = ThreadPoolExecutor(max_workers=1)
+    started = threading.Event()
+    finish = threading.Event()
+    cleaned = threading.Event()
+
+    def _slow():
+        started.set()
+        finish.wait(timeout=5)
+        assert not cleaned.is_set()
+        return "done"
+
+    async def _go():
+        with pytest.raises(ASRTimeoutError):
+            await run_transcribe_guarded(
+                pool,
+                _slow,
+                what="Convert",
+                timeout=0.05,
+                on_abandon=cleaned.set,
+            )
+        assert started.is_set()
+        assert not cleaned.is_set()
+        finish.set()
+        await asyncio.to_thread(cleaned.wait, 2)
+        assert cleaned.is_set()
+
+    try:
+        asyncio.run(_go())
+    finally:
+        finish.set()
+        pool.shutdown(wait=True)
+
+
+def test_normal_completion_keeps_abandon_cleanup_with_caller():
+    pool = ThreadPoolExecutor(max_workers=1)
+    cleaned = threading.Event()
+
+    async def _go():
+        result = await run_transcribe_guarded(
+            pool,
+            lambda: "done",
+            what="Convert",
+            timeout=5,
+            on_abandon=cleaned.set,
+        )
+        assert result == "done"
+        assert not cleaned.is_set()
+
+    try:
+        asyncio.run(_go())
+    finally:
+        pool.shutdown(wait=True)
+
+
 def test_timeout_error_is_a_timeouterror_subclass():
     # Routers that catch broad TimeoutError (openai_compat) must also catch ours.
     assert issubclass(ASRTimeoutError, TimeoutError)
 
 
-def test_timeout_resets_a_resilient_pool_to_restore_capacity():
-    # #730: a wedged transcribe holds its GPU-pool worker forever; with a 1-2
-    # worker pool that starves TTS generate and surfaces as "can't reach
-    # backend". On timeout, run_transcribe_guarded must reset() a pool that
-    # supports it (the real _ResilientGpuPool) so the next submit gets a fresh
-    # worker — capacity restored without an app restart.
+def test_timeout_does_not_overlap_an_in_process_native_worker():
+    # #1669: reset() cannot kill the old native thread. A fresh pool let the
+    # retry enter the same whisperx/CTranslate2 model concurrently and the
+    # process died with 0xC0000005. Keep the old worker accounted for instead.
     class _FakePool(ThreadPoolExecutor):
         def __init__(self):
             super().__init__(max_workers=1)
@@ -105,7 +160,7 @@ def test_timeout_resets_a_resilient_pool_to_restore_capacity():
             await run_transcribe_guarded(pool, _hang, what="Dub", timeout=0.2)
 
     asyncio.run(_go())
-    assert pool.reset_calls == 1
+    assert pool.reset_calls == 0
     pool.shutdown(wait=False)
 
 

backend/tests/test_batch.py

@@ -112,6 +112,42 @@ class TestEnqueue:
         job = client.get(f"/batch/jobs/{job_id}").json()
         assert job["filename"] == "test.mp4"
 
+    @pytest.mark.asyncio
+    async def test_upload_is_persisted_in_bounded_chunks(self, batch, tmp_path):
+        class RecordingUpload:
+            def __init__(self):
+                self.read_sizes = []
+                self.remaining = b"video"
+
+            async def read(self, size):
+                self.read_sizes.append(size)
+                chunk, self.remaining = self.remaining[:size], self.remaining[size:]
+                return chunk
+
+        upload = RecordingUpload()
+        destination = tmp_path / "video.mp4"
+        await batch._save_upload(upload, str(destination))
+
+        assert destination.read_bytes() == b"video"
+        assert upload.read_sizes == [batch._UPLOAD_CHUNK_BYTES, batch._UPLOAD_CHUNK_BYTES]
+
+    @pytest.mark.asyncio
+    async def test_failed_upload_removes_partial_file(self, batch, tmp_path):
+        class FailingUpload:
+            calls = 0
+
+            async def read(self, _size):
+                self.calls += 1
+                if self.calls == 1:
+                    return b"partial"
+                raise OSError("upload interrupted")
+
+        destination = tmp_path / "video.mp4"
+        with pytest.raises(OSError, match="upload interrupted"):
+            await batch._save_upload(FailingUpload(), str(destination))
+
+        assert not destination.exists()
+
 
 class TestListJobs:
     def test_empty(self, client):

backend/tests/test_contained_subprocess.py

@@ -0,0 +1,346 @@
+"""Stable nested operation ownership (model-free, cross-platform seams)."""
+import ctypes
+import builtins
+import os
+import runpy
+import subprocess
+import sys
+import threading
+import time
+import types
+from ctypes import wintypes
+from pathlib import Path
+
+import pytest
+
+from core import contained_subprocess as owned
+
+
+class _Call:
+    def __init__(self, fn):
+        self.fn = fn
+
+    def __call__(self, *args):
+        return self.fn(*args)
+
+
+def test_supervisor_argv_uses_entry_module_for_source_and_frozen_binary(monkeypatch):
+    monkeypatch.delattr(owned.sys, "frozen", raising=False)
+    source = owned._supervisor_argv(3, 4, ["operation"])
+    assert source[:2] == [sys.executable, str(Path(owned.__file__).parents[1] / "main.py")]
+    assert source[2:] == ["--supervise", "3", "4", "--", "operation"]
+
+    monkeypatch.setattr(owned.sys, "frozen", True, raising=False)
+    frozen = owned._supervisor_argv(3, 4, ["operation"])
+    assert frozen == [sys.executable, "--supervise", "3", "4", "--", "operation"]
+
+
+def test_source_main_dispatches_supervisor_before_heavy_imports(monkeypatch):
+    calls = []
+    fake = types.ModuleType("core.contained_subprocess")
+    fake.supervisor_main = lambda args: calls.append(args) or 23
+    monkeypatch.setitem(sys.modules, "core.contained_subprocess", fake)
+    main_path = Path(owned.__file__).parents[1] / "main.py"
+    monkeypatch.setattr(
+        sys,
+        "argv",
+        [str(main_path), "--supervise", "3", "4", "--", "operation"],
+    )
+    original_import = builtins.__import__
+
+    def guard_heavy_import(name, *args, **kwargs):
+        if name == "math":
+            raise AssertionError("supervisor dispatch reached application imports")
+        return original_import(name, *args, **kwargs)
+
+    monkeypatch.setattr(builtins, "__import__", guard_heavy_import)
+    with pytest.raises(SystemExit, match="23"):
+        runpy.run_path(str(main_path), run_name="__main__")
+    assert calls == [["--supervise", "3", "4", "--", "operation"]]
+
+
+@pytest.mark.skipif(os.name != "posix", reason="Unix drain pipe contract")
+def test_drain_fd_is_explicitly_inherited_by_wrapper_but_not_operation(monkeypatch):
+    drain_read, drain_write = os.pipe()
+    monkeypatch.setenv("OMNIVOICE_DESKTOP_CONTAINED", "1")
+    monkeypatch.setenv("OMNIVOICE_DESKTOP_DRAIN_FD", str(drain_write))
+    owned.secure_backend_drain_fd()
+    assert not os.get_inheritable(drain_write)
+    implicit_probe = subprocess.check_output(
+        [
+            sys.executable,
+            "-c",
+            "import os; "
+            "fd=int(os.environ['OMNIVOICE_DESKTOP_DRAIN_FD']); "
+            "\ntry: os.fstat(fd); print('leaked')"
+            "\nexcept OSError: print('closed')",
+        ],
+        close_fds=False,
+        text=True,
+    )
+    assert implicit_probe.strip() == "closed"
+    script = (
+        "import os,time; token=os.environ.get('OMNIVOICE_DESKTOP_DRAIN_FD'); "
+        "marker=os.environ.get('OMNIVOICE_DESKTOP_CONTAINED'); "
+        "\nif token is None and marker is None: state='stripped'"
+        "\nelse:"
+        "\n try: os.fstat(int(token)); state='leaked'"
+        "\n except OSError: state='closed'"
+        "\nprint(state, flush=True); time.sleep(60)"
+    )
+    proc = owned.spawn_owned(
+        [sys.executable, "-c", script],
+        stdout=subprocess.PIPE,
+        text=True,
+    )
+    try:
+        assert proc.stdout.readline().strip() == "stripped"
+        os.close(drain_write)
+        drain_write = -1
+        os.set_blocking(drain_read, False)
+        with pytest.raises(BlockingIOError):
+            os.read(drain_read, 1)  # wrapper still holds the only writer
+        proc.kill()
+        proc.wait(timeout=5)
+        deadline = time.monotonic() + 2
+        while time.monotonic() < deadline:
+            try:
+                if os.read(drain_read, 1) == b"":
+                    break
+            except BlockingIOError:
+                time.sleep(0.01)
+        else:
+            pytest.fail("wrapper exit did not close the desktop drain writer")
+    finally:
+        if drain_write >= 0:
+            os.close(drain_write)
+        os.close(drain_read)
+        if proc.poll() is None:
+            proc.kill()
+            proc.wait(timeout=5)
+
+
+def test_invalid_or_missing_desktop_drain_fd_fails_safe(monkeypatch):
+    monkeypatch.setenv("OMNIVOICE_DESKTOP_CONTAINED", "1")
+    monkeypatch.setenv("OMNIVOICE_DESKTOP_DRAIN_FD", "not-an-fd")
+    with pytest.raises(RuntimeError, match="missing its live.*drain descriptor"):
+        owned.spawn_owned([sys.executable, "-c", "print('unsafe')"])
+
+    monkeypatch.delenv("OMNIVOICE_DESKTOP_DRAIN_FD")
+    with pytest.raises(RuntimeError, match="missing its live.*drain descriptor"):
+        owned.secure_backend_drain_fd()
+
+    monkeypatch.delenv("OMNIVOICE_DESKTOP_CONTAINED")
+    assert owned.backend_drain_fd(required=True) is None
+    proc = owned.spawn_owned(
+        [sys.executable, "-c", "print('standalone')"],
+        stdout=subprocess.PIPE,
+        text=True,
+    )
+    assert proc.stdout.readline().strip() == "standalone"
+    assert proc.wait(timeout=5) == 0
+
+
+def test_windows_operation_is_in_kill_on_close_job_before_resume(monkeypatch):
+    """The child gets no instruction before stable nested Job assignment."""
+    events = []
+    job_closed = threading.Event()
+    job = 99
+
+    def close_handle(handle):
+        value = getattr(handle, "value", handle)
+        events.append(("close", value))
+        if value == job:
+            job_closed.set()
+        return True
+
+    kernel = type("Kernel", (), {})()
+    kernel.AssignProcessToJobObject = _Call(
+        lambda assigned_job, process: events.append(("assign", assigned_job, process)) or True
+    )
+    kernel.TerminateJobObject = _Call(
+        lambda assigned_job, code: events.append(("terminate", assigned_job, code)) or True
+    )
+    kernel.WriteFile = _Call(
+        lambda handle, payload, size, written, overlap: events.append(("write", size)) or True
+    )
+    kernel.CloseHandle = _Call(close_handle)
+
+    def read_control(*_args):
+        job_closed.wait(2)
+        return False
+
+    kernel.ReadFile = _Call(read_control)
+    monkeypatch.setattr(owned, "_windows_job", lambda: (job, kernel, wintypes))
+    monkeypatch.setattr(
+        owned,
+        "_resume_windows_process",
+        lambda _kernel, _types, pid: events.append(("resume", pid)),
+    )
+
+    class Child:
+        _handle = 77
+        pid = 123
+
+        def wait(self, timeout=None):
+            events.append(("wait", timeout))
+            return 0
+
+    monkeypatch.setattr(
+        owned.subprocess,
+        "Popen",
+        lambda *args, **kwargs: events.append(("spawn", kwargs["creationflags"])) or Child(),
+    )
+
+    assert owned._supervise_windows(11, 12, ["operation.exe"]) == 0
+    assert job_closed.wait(1)
+
+    names = [event[0] for event in events]
+    assert names.index("assign") < names.index("resume") < names.index("wait")
+    assert names.index("wait") < names.index("terminate") < names.index("write")
+
+
+def test_windows_assignment_failure_kills_suspended_unowned_child(monkeypatch):
+    """A child outside the nested Job must be killed through its stable handle."""
+    events = []
+    job_closed = threading.Event()
+    job = 99
+
+    def close_handle(handle):
+        value = getattr(handle, "value", handle)
+        events.append(("close", value))
+        if value == job:
+            job_closed.set()
+        return True
+
+    kernel = type("Kernel", (), {})()
+    kernel.AssignProcessToJobObject = _Call(
+        lambda assigned_job, process: events.append(("assign", assigned_job, process))
+        or False
+    )
+    kernel.TerminateJobObject = _Call(
+        lambda assigned_job, code: events.append(("terminate", assigned_job, code)) or True
+    )
+    kernel.WriteFile = _Call(
+        lambda handle, payload, size, written, overlap: events.append(("write", size)) or True
+    )
+    kernel.CloseHandle = _Call(close_handle)
+
+    def read_control(*_args):
+        job_closed.wait(2)
+        return False
+
+    kernel.ReadFile = _Call(read_control)
+    monkeypatch.setattr(owned, "_windows_job", lambda: (job, kernel, wintypes))
+    monkeypatch.setattr(ctypes, "get_last_error", lambda: 5, raising=False)
+
+    class Child:
+        _handle = 77
+        pid = 123
+
+        def kill(self):
+            events.append(("kill",))
+
+        def wait(self, timeout=None):
+            events.append(("wait", timeout))
+            return 1
+
+    monkeypatch.setattr(
+        owned.subprocess,
+        "Popen",
+        lambda *args, **kwargs: events.append(("spawn", kwargs["creationflags"])) or Child(),
+    )
+
+    assert owned._supervise_windows(11, 12, ["operation.exe"]) == 127
+    assert job_closed.wait(1)
+
+    names = [event[0] for event in events]
+    assert names.index("assign") < names.index("terminate") < names.index("kill")
+    assert names.index("kill") < names.index("wait") < names.index("write")
+
+
+def test_windows_direct_job_owner_assigns_before_resume(monkeypatch):
+    """Windows skips the extra Python wrapper but retains pre-start Job ownership."""
+    events = []
+    job = 99
+
+    kernel = type("Kernel", (), {})()
+    kernel.AssignProcessToJobObject = _Call(
+        lambda assigned_job, process: events.append(("assign", assigned_job, process)) or True
+    )
+    kernel.TerminateJobObject = _Call(
+        lambda assigned_job, code: events.append(("terminate", assigned_job, code)) or True
+    )
+    kernel.CloseHandle = _Call(
+        lambda handle: events.append(("close", getattr(handle, "value", handle))) or True
+    )
+    monkeypatch.setattr(owned, "_windows_job", lambda: (job, kernel, wintypes))
+    monkeypatch.setattr(
+        owned,
+        "_resume_windows_process",
+        lambda _kernel, _types, pid: events.append(("resume", pid)),
+    )
+
+    class Child:
+        _handle = 77
+        pid = 123
+        args = ["operation.exe"]
+        stdin = None
+        stdout = object()
+        stderr = object()
+        returncode = None
+
+        def poll(self):
+            return self.returncode
+
+        def wait(self, timeout=None):
+            events.append(("wait", timeout))
+            return self.returncode
+
+        def kill(self):
+            events.append(("kill",))
+
+    child = Child()
+
+    def fake_popen(argv, **kwargs):
+        events.append(("spawn", argv, kwargs))
+        return child
+
+    monkeypatch.setattr(owned.subprocess, "Popen", fake_popen)
+    proc = owned._spawn_windows_owned(
+        ["operation.exe"],
+        {
+            "env": {
+                "KEEP": "yes",
+                "OMNIVOICE_DESKTOP_CONTAINED": "1",
+                "OMNIVOICE_DESKTOP_DRAIN_FD": "42",
+            },
+            "creationflags": 0x00000200,
+        },
+    )
+
+    names = [event[0] for event in events]
+    assert names[:3] == ["spawn", "assign", "resume"]
+    spawn_argv, spawn_kwargs = events[0][1:]
+    assert spawn_argv == ["operation.exe"]
+    assert spawn_kwargs["creationflags"] == 0x08000204
+    assert spawn_kwargs["env"] == {"KEEP": "yes"}
+    assert proc.stdout is child.stdout
+
+    child.returncode = 0
+    assert proc.poll() == 0
+    assert [event[0] for event in events][-2:] == ["terminate", "close"]
+
+
+def test_spawn_owned_selects_direct_windows_job_path(monkeypatch):
+    sentinel = object()
+    calls = []
+    monkeypatch.setattr(owned.os, "name", "nt")
+    monkeypatch.setattr(
+        owned,
+        "_spawn_windows_owned",
+        lambda argv, kwargs: calls.append((argv, kwargs)) or sentinel,
+    )
+
+    assert owned.spawn_owned(["sidecar.exe"], text=True) is sentinel
+    assert calls == [(["sidecar.exe"], {"text": True})]

backend/tests/test_contained_subprocess_waitid_fallback.py

@@ -0,0 +1,126 @@
+"""macOS fallback for the os.waitid probe (#1656).
+
+CPython on macOS does not expose os.waitid, so OwnedPopen's WNOWAIT dance
+crashed with AttributeError on every poll after the first spawn. These tests
+simulate that platform (monkeypatch os.waitid away) and pin the fallback:
+poll/wait/kill must work, exit codes must be real, and an already-reaped
+leader must be refused (ChildProcessError path), never signalled blind.
+"""
+import os
+import subprocess
+import sys
+import time
+
+import pytest
+
+from core import contained_subprocess as owned
+
+
+def _make_owned(argv):
+    cr, cw = os.pipe()
+    rr, rw = os.pipe()
+    proc = subprocess.Popen(argv, start_new_session=True)
+    os.close(cw)
+    os.close(rw)  # result writer gone: _read_result falls back to wrapper rc
+    return owned.OwnedPopen(proc, cr, rr), proc
+
+
+@pytest.fixture()
+def no_waitid(monkeypatch):
+    monkeypatch.delattr(os, "waitid", raising=False)
+
+
+def test_poll_running_then_exited_without_waitid(no_waitid):
+    h, _ = _make_owned([sys.executable, "-c", "import time; time.sleep(1.5)"])
+    try:
+        assert h.poll() is None, "running child must poll None"
+        h._proc.wait()
+        deadline = time.monotonic() + 5
+        rc = None
+        while rc is None and time.monotonic() < deadline:
+            rc = h.poll()
+            time.sleep(0.05)
+        assert rc == 0
+        assert h.poll() == 0
+    finally:
+        h._close_control()
+        if h._result_fd is not None:
+            os.close(h._result_fd)
+
+
+def test_poll_reports_real_exit_code_without_waitid(no_waitid):
+    h, _ = _make_owned([sys.executable, "-c", "raise SystemExit(3)"])
+    try:
+        deadline = time.monotonic() + 5
+        while h.poll() is None and time.monotonic() < deadline:
+            time.sleep(0.05)
+        assert h.poll() == 3
+    finally:
+        h._close_control()
+        if h._result_fd is not None:
+            os.close(h._result_fd)
+
+
+def test_wait_returns_after_kill_without_waitid(no_waitid):
+    h, _ = _make_owned([sys.executable, "-c", "import time; time.sleep(30)"])
+    try:
+        h.kill()
+        rc = h.wait(timeout=5)
+        assert rc != 0
+    finally:
+        h._close_control()
+        if h._result_fd is not None:
+            os.close(h._result_fd)
+
+
+def test_reaped_by_own_popen_reports_code_without_waitid(no_waitid):
+    h, proc = _make_owned([sys.executable, "-c", "pass"])
+    try:
+        proc.wait()  # reaped through OUR handle: known code, not a refusal
+        assert h.poll() == 0
+    finally:
+        h._close_control()
+        if h._result_fd is not None:
+            os.close(h._result_fd)
+
+
+def test_foreign_reaped_leader_is_refused_without_waitid(no_waitid):
+    h, proc = _make_owned([sys.executable, "-c", "pass"])
+    try:
+        # Reap OUTSIDE this handle: Popen never learns the code, so poll must
+        # refuse (None) rather than guess or signal a maybe-reused group.
+        while True:
+            pid, _ = os.waitpid(proc.pid, os.WNOHANG)
+            if pid == proc.pid:
+                break
+            time.sleep(0.05)
+        assert h.poll() is None
+    finally:
+        h._close_control()
+        if h._result_fd is not None:
+            os.close(h._result_fd)
+
+
+def test_kill_after_pid_reuse_does_not_signal_without_waitid(no_waitid, monkeypatch):
+    """A foreign-reaped leader's reused numeric pid must not authorize killpg."""
+    import signal as _signal
+
+    h, proc = _make_owned([sys.executable, "-c", "pass"])
+    try:
+        while True:
+            pid, _ = os.waitpid(proc.pid, os.WNOHANG)
+            if pid == proc.pid:
+                break
+            time.sleep(0.05)
+        # Model the numeric pid being reused: kill(pid, 0) would succeed even
+        # though waitpid still reports that the original child is no longer
+        # ours. The old guard therefore reached killpg and fails this test.
+        monkeypatch.setattr(os, "kill", lambda _pid, _sig: None)
+        signalled = []
+        monkeypatch.setattr(os, "killpg", lambda pid, sig: signalled.append((pid, sig)))
+        h._signal_owned_group(_signal.SIGKILL)
+        assert signalled == []
+    finally:
+        h._close_control()
+        if h._result_fd is not None:
+            os.close(h._result_fd)

backend/tests/test_dub_preview_upload.py

@@ -0,0 +1,37 @@
+from __future__ import annotations
+
+import asyncio
+import io
+import threading
+
+import pytest
+from fastapi import UploadFile
+
+
+@pytest.mark.asyncio
+async def test_preview_ffmpeg_does_not_block_event_loop(monkeypatch, tmp_path):
+    from api.routers import dub_core
+
+    loop = asyncio.get_running_loop()
+    started = asyncio.Event()
+    release = threading.Event()
+
+    def slow_ffmpeg(*_args, **_kwargs):
+        loop.call_soon_threadsafe(started.set)
+        assert release.wait(timeout=2)
+
+    monkeypatch.setattr(dub_core, "PREVIEW_DIR", str(tmp_path))
+    monkeypatch.setattr(dub_core, "find_ffmpeg", lambda: "ffmpeg")
+    monkeypatch.setattr(dub_core.subprocess, "run", slow_ffmpeg)
+    upload = UploadFile(filename="preview.mp4", file=io.BytesIO(b"video"))
+
+    before = loop.time()
+    task = asyncio.create_task(dub_core.preview_upload(upload))
+    try:
+        await asyncio.wait_for(started.wait(), timeout=1)
+        assert loop.time() - before < 0.5
+    finally:
+        release.set()
+
+    result = await task
+    assert result["audioUrl"].endswith(".wav")

backend/tests/test_generation_adhoc_ref_cleanup.py

@@ -0,0 +1,71 @@
+from __future__ import annotations
+
+import asyncio
+import threading
+from concurrent.futures import ThreadPoolExecutor
+
+import pytest
+
+
+@pytest.mark.asyncio
+async def test_abandoned_reader_keeps_adhoc_reference_until_worker_finishes(tmp_path):
+    from api.routers.generation import (
+        _TempReferenceLease,
+        _run_with_reference_lease,
+    )
+    from services.model_manager import run_on_gpu_pool_guarded
+
+    reference = tmp_path / "reference.wav"
+    reference.write_bytes(b"voice")
+    lease = _TempReferenceLease(str(reference))
+    started = threading.Event()
+    release_worker = threading.Event()
+    worker_read = threading.Event()
+
+    def read_reference():
+        started.set()
+        assert release_worker.wait(timeout=2)
+        assert reference.read_bytes() == b"voice"
+        worker_read.set()
+
+    with ThreadPoolExecutor(max_workers=1) as executor:
+        task = asyncio.create_task(
+            _run_with_reference_lease(
+                lease,
+                lambda on_abandon: run_on_gpu_pool_guarded(
+                    read_reference,
+                    executor=executor,
+                    timeout=1,
+                    on_abandon=on_abandon,
+                ),
+            )
+        )
+        assert await asyncio.to_thread(started.wait, 1)
+        task.cancel()
+        cancelled = await asyncio.gather(task, return_exceptions=True)
+        assert isinstance(cancelled[0], asyncio.CancelledError)
+
+        lease.finish_request()
+        assert reference.exists()
+        release_worker.set()
+        assert await asyncio.to_thread(worker_read.wait, 1)
+
+    for _ in range(100):
+        if not reference.exists():
+            break
+        await asyncio.sleep(0.01)
+    assert not reference.exists()
+
+
+def test_normal_request_deletes_adhoc_reference_immediately(tmp_path):
+    from api.routers.generation import _TempReferenceLease
+
+    reference = tmp_path / "reference.wav"
+    reference.write_bytes(b"voice")
+    lease = _TempReferenceLease(str(reference))
+
+    release = lease.acquire()
+    release()
+    lease.finish_request()
+
+    assert not reference.exists()

backend/tests/test_gpu_pool_abandon_cleanup.py

@@ -0,0 +1,70 @@
+from __future__ import annotations
+
+import asyncio
+import threading
+from concurrent.futures import ThreadPoolExecutor
+
+import pytest
+
+
+@pytest.mark.asyncio
+async def test_abandon_callback_waits_for_running_worker_to_finish():
+    from services.model_manager import run_on_gpu_pool_guarded
+
+    started = threading.Event()
+    release = threading.Event()
+    cleaned = threading.Event()
+
+    def job():
+        started.set()
+        assert release.wait(timeout=2)
+
+    with ThreadPoolExecutor(max_workers=1) as executor:
+        task = asyncio.create_task(
+            run_on_gpu_pool_guarded(
+                job,
+                executor=executor,
+                timeout=1,
+                on_abandon=cleaned.set,
+            )
+        )
+        assert await asyncio.to_thread(started.wait, 1)
+        task.cancel()
+        cancelled = await asyncio.gather(task, return_exceptions=True)
+        assert isinstance(cancelled[0], asyncio.CancelledError)
+
+        assert not cleaned.is_set()
+        release.set()
+        assert await asyncio.to_thread(cleaned.wait, 1)
+
+
+@pytest.mark.asyncio
+async def test_queued_cancellation_releases_without_running_job():
+    from services.model_manager import GpuPoolBusyError, run_on_gpu_pool_guarded
+
+    hog_started = threading.Event()
+    release_hog = threading.Event()
+    cleaned = threading.Event()
+    queued_job_ran = threading.Event()
+
+    def hog():
+        hog_started.set()
+        assert release_hog.wait(timeout=2)
+
+    with ThreadPoolExecutor(max_workers=1) as executor:
+        hog_future = executor.submit(hog)
+        assert hog_started.wait(timeout=1)
+        try:
+            with pytest.raises(GpuPoolBusyError):
+                await run_on_gpu_pool_guarded(
+                    queued_job_ran.set,
+                    executor=executor,
+                    timeout=1,
+                    queue_timeout=0.05,
+                    on_abandon=cleaned.set,
+                )
+            assert cleaned.is_set()
+            assert not queued_job_ran.is_set()
+        finally:
+            release_hog.set()
+            hog_future.result(timeout=1)

backend/tests/test_omnivoice_subprocess.py

@@ -17,18 +17,31 @@ import json
 import math
 import array
 import base64
+import io
+import os
+import subprocess
+import sys
+import time
+import asyncio
+from pathlib import Path
 
 import pytest
 
-from services.subprocess_backend import SubprocessBackend, RECV_TIMEOUT_S
-from services.tts_backend import get_backend_class
-from engines.omnivoice_subprocess import OmniVoiceSubprocessBackend
+from services.subprocess_backend import (
+    RECV_TIMEOUT_S,
+    SubprocessBackend,
+)
+from services.tts_backend import OmniVoiceBackend, get_backend_class, list_backends
+from engines.omnivoice_subprocess import (
+    OmniVoiceMPSSubprocessBackend,
+    OmniVoiceSubprocessBackend,
+)
 
 
 # ── stub sidecar (model-free) ──────────────────────────────────────────────
 
 STUB_SIDECAR = r'''
-import sys, json, struct, time, math, array, base64
+import sys, os, json, struct, time, math, array, base64, subprocess
 
 def _send(o):
     b = json.dumps(o, separators=(",", ":")).encode()
@@ -60,9 +73,20 @@ while True:
         sys.exit(0)
     elif op == "synthesize":
         t = m.get("text", "")
+        if t == "CRASH":
+            os._exit(137)
         if t == "HANG":
             while True:  # wedge forever; the parent must hard-kill us
                 time.sleep(1)
+        if t == "HANG_CHILD":
+            subprocess.Popen([
+                sys.executable,
+                "-c",
+                "import os,time; time.sleep(1); "
+                "open(os.environ['OMNIVOICE_TIMEOUT_MARKER'], 'w').write('bad')",
+            ])
+            while True:
+                time.sleep(1)
         # Emit progress frames before the audio when asked, to exercise the
         # parent's progress-consuming recv loop (the cold-load fix).
         if t.startswith("PROG:"):
@@ -98,6 +122,80 @@ def test_registry_resolves_to_subprocess_backend():
     assert get_backend_class("omnivoice-subprocess") is OmniVoiceSubprocessBackend
 
 
+@pytest.mark.parametrize(
+    ("family", "expected_name"),
+    [("mps", "OmniVoiceMPSSubprocessBackend"), ("cuda", "OmniVoiceBackend"),
+     ("cpu", "OmniVoiceBackend")],
+)
+def test_omnivoice_is_crash_isolated_only_on_mps(monkeypatch, family, expected_name):
+    from core.device_caps import HostCaps
+
+    available = (family, "cpu") if family != "cpu" else ("cpu",)
+    monkeypatch.setattr(
+        "core.device_caps.detect_host_caps",
+        lambda: HostCaps(family=family, available_families=available),
+    )
+
+    resolved = get_backend_class("omnivoice")
+    assert resolved.__name__ == expected_name
+    if family != "mps":
+        assert resolved is OmniVoiceBackend
+
+
+def test_engine_catalogue_reports_effective_mps_isolation(monkeypatch):
+    from core.device_caps import HostCaps
+    from services import tts_backend
+
+    monkeypatch.setattr(tts_backend, "_REGISTRY", {"omnivoice": OmniVoiceBackend})
+    monkeypatch.setattr(
+        "core.device_caps.detect_host_caps",
+        lambda: HostCaps(family="mps", available_families=("mps", "cpu")),
+    )
+    monkeypatch.setattr(
+        "engines.omnivoice_subprocess.OmniVoiceSubprocessBackend.is_available",
+        classmethod(lambda cls: (True, "ready")),
+    )
+
+    row = next(item for item in list_backends() if item["id"] == "omnivoice")
+    assert row["isolation_mode"] == "subprocess"
+
+
+def test_mps_startup_does_not_preload_native_model(monkeypatch):
+    from core.device_caps import HostCaps
+    from services import model_manager
+
+    monkeypatch.setattr(
+        "core.device_caps.detect_host_caps",
+        lambda: HostCaps(family="mps", available_families=("mps", "cpu")),
+    )
+    monkeypatch.setenv("OMNIVOICE_TTS_BACKEND", "omnivoice")
+    monkeypatch.setattr(model_manager, "model", None)
+
+    async def fail_load():
+        raise AssertionError("native OmniVoice must not load in the API process on MPS")
+
+    monkeypatch.setattr(model_manager, "_load_model_with_timeout", fail_load)
+    asyncio.run(model_manager.preload_model())
+
+
+def test_streaming_mps_path_does_not_load_native_model(monkeypatch):
+    from api.routers.tts_stream import _resolve_stream_backend
+    from services import model_manager, tts_backend
+
+    sentinel = object()
+    monkeypatch.setattr(tts_backend, "active_backend_id", lambda: "omnivoice")
+    monkeypatch.setattr(
+        tts_backend, "get_backend_class", lambda _id: OmniVoiceMPSSubprocessBackend,
+    )
+    monkeypatch.setattr(tts_backend, "get_active_tts_backend", lambda: sentinel)
+
+    async def fail_load():
+        raise AssertionError("streaming must not load native OmniVoice on MPS")
+
+    monkeypatch.setattr(model_manager, "get_model", fail_load)
+    assert asyncio.run(_resolve_stream_backend(None)) is sentinel
+
+
 def test_is_marked_subprocess_isolated():
     # list_backends() detects isolation via this duck-typed marker, not issubclass.
     assert getattr(OmniVoiceSubprocessBackend, "_is_subprocess_isolated", False) is True
@@ -136,11 +234,81 @@ def test_base_default_recv_timeout_is_60s():
     assert _PlainBackend().recv_timeout_s == 60.0
 
 
+def test_sidecar_spawn_delegates_all_containment_to_nested_owner(monkeypatch, tmp_path):
+    from services import subprocess_backend as backend_module
+
+    captured = {}
+
+    class StubProcess:
+        stderr = io.BytesIO()
+
+        @staticmethod
+        def poll():
+            return None
+
+    def fake_spawn(argv, **kwargs):
+        captured.update(kwargs)
+        return StubProcess()
+
+    monkeypatch.setattr(_PlainBackend, "venv_python", classmethod(lambda cls: Path(sys.executable)))
+    monkeypatch.setattr(
+        _PlainBackend,
+        "sidecar_script",
+        classmethod(lambda cls: tmp_path / "stub.py"),
+    )
+    monkeypatch.setattr(backend_module, "spawn_owned", fake_spawn)
+    monkeypatch.setattr(backend_module, "_ensure_reaper_running", lambda: None)
+    backend = _PlainBackend()
+    monkeypatch.setattr(backend, "_recv_with_timeout", lambda _timeout: {"op": "ready"})
+
+    try:
+        backend._spawn()
+        assert not ({"start_new_session", "creationflags", "preexec_fn"} & captured.keys())
+    finally:
+        backend._proc = None
+
+
 def test_omnivoice_subprocess_recv_timeout_overrides_default():
     b = OmniVoiceSubprocessBackend()
     assert b.recv_timeout_s == 300.0  # aligns with the generate budget
 
 
+def test_omnivoice_subprocess_has_longer_spawn_budget_than_other_sidecars():
+    assert _PlainBackend.spawn_ready_timeout_s == 30.0
+    assert OmniVoiceSubprocessBackend.spawn_ready_timeout_s == 120.0
+
+
+def test_spawn_uses_backend_specific_ready_timeout(monkeypatch, tmp_path):
+    _use_stub(monkeypatch, tmp_path / "unused.py")
+    backend = OmniVoiceSubprocessBackend()
+    observed = []
+
+    class StubProcess:
+        stderr = io.BytesIO()
+
+        @staticmethod
+        def poll():
+            return None
+
+    monkeypatch.setattr(
+        "services.subprocess_backend.spawn_owned",
+        lambda *_args, **_kwargs: StubProcess(),
+    )
+    monkeypatch.setattr(
+        backend,
+        "_recv_with_timeout",
+        lambda timeout: observed.append(timeout) or {"op": "ready"},
+    )
+    monkeypatch.setattr("services.subprocess_backend._ensure_reaper_running", lambda: None)
+
+    try:
+        backend._spawn()
+    finally:
+        backend._proc = None
+
+    assert observed == [120.0]
+
+
 def test_omnivoice_subprocess_recv_timeout_env_override(monkeypatch):
     monkeypatch.setenv("OMNIVOICE_SIDECAR_RECV_TIMEOUT_S", "120")
     assert OmniVoiceSubprocessBackend().recv_timeout_s == 120.0
@@ -205,6 +373,48 @@ def test_wedged_sidecar_is_hard_killed_and_recovers(stub_sidecar, monkeypatch):
         b.shutdown()
 
 
+def test_mps_proxy_survives_fatal_child_exit_and_recovers(stub_sidecar, monkeypatch):
+    _use_stub(monkeypatch, stub_sidecar)
+    monkeypatch.setattr(
+        "services.model_manager.make_room_before_generate", lambda: None,
+    )
+    b = OmniVoiceMPSSubprocessBackend()
+    try:
+        with pytest.raises(RuntimeError, match="backend is still running"):
+            b.generate("CRASH")
+        assert b._proc is not None and b._proc.poll() is not None
+        assert b.generate("ok").shape[1] == 24000
+    finally:
+        b.shutdown()
+
+
+def test_desktop_timeout_kills_engine_subtree_before_late_mutation(
+    stub_sidecar, monkeypatch, tmp_path
+):
+    marker = tmp_path / "late-engine-mutation"
+    monkeypatch.setenv("OMNIVOICE_DESKTOP_CONTAINED", "1")
+    drain_read, drain_write = os.pipe()
+    monkeypatch.setenv("OMNIVOICE_DESKTOP_DRAIN_FD", str(drain_write))
+    monkeypatch.setenv("OMNIVOICE_TIMEOUT_MARKER", str(marker))
+    _use_stub(monkeypatch, stub_sidecar)
+    monkeypatch.setattr(
+        OmniVoiceSubprocessBackend,
+        "recv_timeout_s",
+        property(lambda self: 0.3),
+    )
+    b = OmniVoiceSubprocessBackend()
+    try:
+        with pytest.raises(RuntimeError):
+            b.generate("HANG_CHILD")
+        time.sleep(1.2)
+        assert not marker.exists()
+        assert b.generate("ok").shape[1] == 24000
+    finally:
+        b.shutdown()
+        os.close(drain_write)
+        os.close(drain_read)
+
+
 def test_generate_does_not_deadlock_when_called_on_gpu_pool_worker(stub_sidecar, monkeypatch):
     # Regression: /v1/audio/speech and /generate dispatch backend.generate() via
     # run_on_gpu_pool_guarded, i.e. ON a gpu-pool worker. generate() must NOT
@@ -222,3 +432,94 @@ def test_generate_does_not_deadlock_when_called_on_gpu_pool_worker(stub_sidecar,
         assert tensor.shape[1] == 24000
     finally:
         b.shutdown()
+
+
+def test_sidecar_forwards_native_controls_and_applies_seed(monkeypatch):
+    import torch
+    from engines.omnivoice_subprocess import main as sidecar
+
+    calls = []
+    seeds = []
+    frames = []
+
+    class FakeModel:
+        sampling_rate = 24000
+
+        def generate(self, **kwargs):
+            calls.append(kwargs)
+            return [torch.zeros(1, 16)]
+
+    monkeypatch.setattr(sidecar, "_load_model", lambda _stdout: FakeModel())
+    monkeypatch.setattr(sidecar, "_send", lambda _stdout, frame: frames.append(frame))
+    real_manual_seed = torch.manual_seed
+    monkeypatch.setattr(
+        torch, "manual_seed", lambda seed: (seeds.append(seed), real_manual_seed(seed))[1],
+    )
+
+    sidecar._handle_synthesize({
+        "text": "hello",
+        "seed": 123,
+        "t_shift": 0.4,
+        "layer_penalty_factor": 0.2,
+        "position_temperature": 0.7,
+        "class_temperature": 0.8,
+        "audio_chunk_duration": 10,
+        "audio_chunk_threshold": 0.6,
+    }, object())
+
+    assert seeds == [123]
+    assert calls == [{
+        "text": "hello",
+        "ref_audio": None,
+        "ref_text": None,
+        "t_shift": 0.4,
+        "layer_penalty_factor": 0.2,
+        "position_temperature": 0.7,
+        "class_temperature": 0.8,
+        "audio_chunk_duration": 10,
+        "audio_chunk_threshold": 0.6,
+    }]
+    assert frames[-1]["op"] == "audio"
+
+
+def test_generation_proxy_forwards_native_controls_and_seed():
+    import torch
+    from api.routers.generation import _run_backend_inference
+
+    calls = []
+
+    class Proxy:
+        id = "omnivoice"
+        display_name = "OmniVoice"
+        sample_rate = 24000
+        applies_own_mastering = True
+        supports_native_omnivoice_controls = True
+
+        def generate(self, text, **kwargs):
+            calls.append((text, kwargs))
+            return torch.zeros(1, 240)
+
+    _run_backend_inference(
+        Proxy(), "hello", "en", None, None, None, None,
+        16, 2.0, 1.0, False, False, 321,
+        t_shift=0.4, layer_penalty_factor=0.2,
+        position_temperature=0.7, class_temperature=0.8,
+    )
+
+    assert calls == [("hello", {
+        "duration": None,
+        "language": "en",
+        "ref_audio": None,
+        "ref_text": None,
+        "instruct": None,
+        "num_step": 16,
+        "guidance_scale": 2.0,
+        "speed": 1.0,
+        "denoise": False,
+        "postprocess_output": False,
+        "t_shift": 0.4,
+        "layer_penalty_factor": 0.2,
+        "position_temperature": 0.7,
+        "class_temperature": 0.8,
+        "seed": 321,
+    })]

backend/worker/async_utils.py

@@ -0,0 +1,61 @@
+"""Cancellation helpers for work that cannot be stopped mid-call."""
+
+from __future__ import annotations
+
+import asyncio
+from collections.abc import Callable
+from typing import Any, TypeVar
+
+_Result = TypeVar("_Result")
+
+
+async def drain_task(task: asyncio.Task[Any]) -> None:
+    """Wait for ``task`` even if the waiter is cancelled again."""
+    while not task.done():
+        try:
+            await asyncio.shield(task)
+        except asyncio.CancelledError:
+            continue
+        except BaseException:
+            break
+    if task.done():
+        try:
+            task.result()
+        except BaseException:
+            pass
+
+
+async def to_thread_and_drain_on_cancel(
+    function: Callable[..., _Result], /, *args: Any
+) -> _Result:
+    """Run a blocking call without detaching it when its waiter is cancelled."""
+    thread_task = asyncio.create_task(asyncio.to_thread(function, *args))
+    try:
+        return await asyncio.shield(thread_task)
+    except asyncio.CancelledError:
+        await drain_task(thread_task)
+        raise
+
+
+async def to_thread_and_defer_cancellation(
+    function: Callable[..., _Result], /, *args: Any
+) -> tuple[_Result, bool]:
+    """Finish a durable call and report cancellation after its result is known.
+
+    Authority writes need their event-loop publication even when the HTTP
+    caller disappears while SQLite is committing. Returning the cancellation
+    flag lets the caller publish that result first, then propagate cancellation.
+    """
+    thread_task = asyncio.create_task(asyncio.to_thread(function, *args))
+    try:
+        return await asyncio.shield(thread_task), False
+    except asyncio.CancelledError:
+        await drain_task(thread_task)
+        return thread_task.result(), True
+
+
+__all__ = [
+    "drain_task",
+    "to_thread_and_defer_cancellation",
+    "to_thread_and_drain_on_cancel",
+]

backend/worker/capacity.py

@@ -59,10 +59,18 @@ _VRAM_PER_JOB_BYTES = 5 * 1024**3
 #   cpu     — oversubscription just thrashes
 _ALWAYS_SERIAL = frozenset({"mps", "mlx", "cpu", ""})
 
-# Absolute ceiling regardless of how much memory a card reports. Beyond this
-# the bottleneck stops being VRAM and starts being scheduler overhead and
-# host-side I/O contention.
-_MAX_DERIVED = 4
+# Absolute protocol ceiling regardless of how much memory a peer reports.
+# Beyond this the bottleneck stops being VRAM and starts being scheduler
+# overhead and host-side I/O contention.  It is public because every wire
+# boundary must clamp to the same number; a UINT32_MAX heartbeat must not grow
+# a scheduler queue that local derivation would never create.
+MAX_CONCURRENT_TASKS = 4
+
+
+def clamp_concurrency(value: int, *, allow_zero: bool = False) -> int:
+    """Bound an advertised concurrency value to the server's safe range."""
+    minimum = 0 if allow_zero else 1
+    return max(minimum, min(MAX_CONCURRENT_TASKS, int(value)))
 
 # Bounds on how long a parked slot is held. The caller passes the timed-out
 # job's own execution budget — the longest its thread can still legitimately be
@@ -104,7 +112,7 @@ def derive_concurrency(
     budget = max(min_model_bytes, _VRAM_PER_JOB_BYTES)
     if budget <= 0:
         return 1
-    return max(1, min(_MAX_DERIVED, int(free_memory_bytes // budget)))
+    return clamp_concurrency(int(free_memory_bytes // budget))
 
 
 @dataclass
@@ -149,6 +157,9 @@ class WorkerCapacity:
     resident_models: set[str] = field(default_factory=set)
     slots: dict[str, ModelSlot] = field(default_factory=dict)
 
+    def __post_init__(self) -> None:
+        self.max_concurrent_tasks = clamp_concurrency(self.max_concurrent_tasks)
+
     @staticmethod
     def slot_key(engine: str, model_id: str) -> str:
         return f"{engine}:{model_id}"
@@ -198,6 +209,21 @@ class WorkerCapacity:
             )
         slot.active += 1
 
+    def reserve_unknown(self) -> None:
+        """Consume worker-wide capacity for claimed work we cannot classify.
+
+        Reconciliation will tell the peer to cancel a terminal or unknown
+        attempt, but until that cancellation lands it is still using the GPU.
+        """
+        self.active_tasks += 1
+
+    def release_unknown(self) -> bool:
+        """Release one exact reconciled claim with no model-slot identity."""
+        if self.active_tasks <= 0:
+            return False
+        self.active_tasks -= 1
+        return True
+
     def release(
         self,
         engine: str,
@@ -267,8 +293,12 @@ class WorkerCapacity:
     ) -> None:
         """Adopt a heartbeat snapshot. The worker is the source of truth for
         what it is actually running."""
-        self.active_tasks = max(0, active_tasks)
-        reported_ceiling = self.active_tasks + max(0, available_slots)
+        self.active_tasks = clamp_concurrency(active_tasks, allow_zero=True)
+        bounded_available = clamp_concurrency(available_slots, allow_zero=True)
+        bounded_available = min(
+            bounded_available, MAX_CONCURRENT_TASKS - self.active_tasks
+        )
+        reported_ceiling = self.active_tasks + bounded_available
         if reported_ceiling > 0:
             # Adopted, not merely grown. The worker computes this as its own
             # ``max_concurrent_tasks``, so a ceiling we refuse to lower is one
@@ -305,4 +335,10 @@ class WorkerCapacity:
         }
 
 
-__all__ = ["ModelSlot", "WorkerCapacity", "derive_concurrency"]
+__all__ = [
+    "MAX_CONCURRENT_TASKS",
+    "ModelSlot",
+    "WorkerCapacity",
+    "clamp_concurrency",
+    "derive_concurrency",
+]

backend/worker/executor.py

@@ -17,15 +17,19 @@ from __future__ import annotations
 
 import asyncio
 import base64
+import errno
 import hashlib
+import io
 import json
 import logging
 import os
-import io
-import zipfile
+import threading
+import time
 import uuid
+import zipfile
 from typing import Any, Awaitable, Callable, Optional
 
+from worker.async_utils import drain_task
 from worker.errors import ErrorClass, WorkerError
 
 logger = logging.getLogger("omnivoice.worker")
@@ -44,6 +48,18 @@ INPUT_ERRORS_PARAM = "input_errors"
 # leak on the worker that unpurged artifacts were on the control plane.
 INPUT_CACHE_LIMIT_BYTES = 2 * 1024 * 1024 * 1024
 _FALLBACK_INPUT_FETCH_SECONDS = 600.0
+_STALE_INPUT_PARTIAL_SECONDS = 60 * 60.0
+
+# Pruning runs in worker threads and every executor instance shares the same
+# on-disk cache, so active-path leases are process-wide and thread-safe.
+_INPUT_CACHE_LEASE_LOCK = threading.Lock()
+_INPUT_CACHE_LEASES: dict[str, int] = {}
+_INPUT_CACHE_FETCH_LEASES: dict[str, int] = {}
+_INPUT_CACHE_MUTATIONS: set[str] = set()
+# Concurrent fetches keep distinct partial files but serialize the instant a
+# verified generation is published at its content address.
+_INPUT_CACHE_FETCH_LOCKS: dict[str, asyncio.Lock] = {}
+_INPUT_CACHE_FETCH_USERS: dict[str, int] = {}
 
 #   on_progress(fraction: float, stage: str)
 #   on_model_loading(fraction: float, detail: str)
@@ -88,6 +104,7 @@ class TaskExecutor:
         self._on_model_loading = on_model_loading
         self._fetch_input = fetch_input
         self._input_dir = input_dir
+        self._blocking_tasks: set[asyncio.Task] = set()
 
     async def execute(
         self,
@@ -110,33 +127,35 @@ class TaskExecutor:
         """
         operation = (assignment.operation or "tts").lower()
         params = _parse_params(assignment.params_json)
-        params = await self._materialize_inputs(
+        params, leased_inputs = await self._materialize_inputs(
             assignment, params, fetch_input or self._fetch_input
         )
-
-        handler = {
-            "tts": self._run_tts,
-            "clone": self._run_tts,
-            "audiobook": self._run_audiobook,
-            "dub_segments": self._run_dub_segments,
-        }.get(operation)
-        if handler is None:
-            raise TaskFailure(
-                WorkerError(
-                    error_class=ErrorClass.CAPABILITY,
-                    code="OPERATION_UNSUPPORTED",
-                    message=f"This worker cannot run '{operation}' tasks.",
-                    hint="Run this task locally, or use a worker that supports it.",
+        try:
+            handler = {
+                "tts": self._run_tts,
+                "clone": self._run_tts,
+                "audiobook": self._run_audiobook,
+                "dub_segments": self._run_dub_segments,
+            }.get(operation)
+            if handler is None:
+                raise TaskFailure(
+                    WorkerError(
+                        error_class=ErrorClass.CAPABILITY,
+                        code="OPERATION_UNSUPPORTED",
+                        message=f"This worker cannot run '{operation}' tasks.",
+                        hint="Run this task locally, or use a worker that supports it.",
+                    )
                 )
+            return await handler(
+                assignment,
+                params,
+                _Reporters(
+                    on_progress or self._on_progress,
+                    on_model_loading or self._on_model_loading,
+                ),
             )
-        return await handler(
-            assignment,
-            params,
-            _Reporters(
-                on_progress or self._on_progress,
-                on_model_loading or self._on_model_loading,
-            ),
-        )
+        finally:
+            self._release_inputs_after_active_work(leased_inputs)
 
     async def _run_dub_segments(self, assignment, params: dict, report: "_Reporters") -> dict:
         """Render every requested dub line under one lease and return one bundle."""
@@ -150,8 +169,9 @@ class TaskExecutor:
             ))
         load_budget, run_budget = _budgets(assignment)
         await report.loading(0.0, f"preparing {assignment.engine}")
-        backend = await self._bounded(
-            asyncio.to_thread(self._load_backend, assignment.engine),
+        backend = await self._bounded_thread(
+            self._load_backend,
+            assignment.engine,
             timeout=load_budget, code="MODEL_LOAD_TIMEOUT", what=f"Loading '{assignment.engine}'",
         )
         await report.loading(1.0, "model ready")
@@ -159,11 +179,15 @@ class TaskExecutor:
         for index, row in enumerate(rows):
             row = dict(row)
             row["ref_audio"] = refs[index] if index < len(refs) else None
-            audio = await self._bounded(
-                asyncio.to_thread(self._synthesize_dub_segment, backend, row),
+            audio = await self._bounded_thread(
+                self._synthesize_dub_segment,
+                backend,
+                row,
                 timeout=run_budget, code="EXECUTION_TIMEOUT", what=f"Dubbing segment {index + 1}",
             )
-            payload, _meta = await asyncio.to_thread(self._encode, audio, row, backend)
+            payload, _meta = await self._thread_call(
+                self._encode, audio, row, backend
+            )
             rendered.append((int(row.get("index", index)), payload))
             await report.progress((index + 1) / len(rows), f"segment {index + 1} of {len(rows)}")
 
@@ -184,9 +208,11 @@ class TaskExecutor:
         from services.text_normalization import normalize_for_tts
 
         text = normalize_for_tts(row.get("text") or "", row.get("language"))
+        seed = None
         if row.get("seed") is not None:
             import torch
-            torch.manual_seed(int(row["seed"]))
+            seed = int(row["seed"])
+            torch.manual_seed(seed)
         kwargs = {
             "language": row.get("language") if row.get("language") != "Auto" else None,
             "ref_audio": row.get("ref_audio"), "ref_text": row.get("ref_text"),
@@ -197,6 +223,11 @@ class TaskExecutor:
             "speed": float(row.get("speed") or 1.0), "denoise": True,
             "postprocess_output": True,
         }
+        if (
+            getattr(backend, "supports_native_omnivoice_controls", False)
+            and seed is not None
+        ):
+            kwargs["seed"] = seed
         audio = backend.generate(text=text, **kwargs)
         preset = row.get("effect_preset") or "broadcast"
         if preset != "raw":
@@ -225,8 +256,9 @@ class TaskExecutor:
         load_budget, run_budget = _budgets(assignment)
 
         await report.loading(0.0, f"preparing {assignment.engine}")
-        backend = await self._bounded(
-            asyncio.to_thread(self._load_backend, assignment.engine),
+        backend = await self._bounded_thread(
+            self._load_backend,
+            assignment.engine,
             timeout=load_budget,
             code="MODEL_LOAD_TIMEOUT",
             what=f"Loading '{assignment.engine}'",
@@ -234,16 +266,22 @@ class TaskExecutor:
         await report.loading(1.0, "model ready")
 
         await report.progress(0.05, "synthesising")
-        audio = await self._bounded(
-            asyncio.to_thread(self._synthesize, backend, text, params),
+        audio = await self._bounded_thread(
+            self._synthesize,
+            backend,
+            text,
+            params,
             timeout=run_budget,
             code="EXECUTION_TIMEOUT",
             what="Synthesis",
         )
         await report.progress(0.9, "encoding")
 
-        payload, meta = await self._bounded(
-            asyncio.to_thread(self._encode, audio, params, backend),
+        payload, meta = await self._bounded_thread(
+            self._encode,
+            audio,
+            params,
+            backend,
             timeout=run_budget,
             code="EXECUTION_TIMEOUT",
             what="Encoding",
@@ -264,19 +302,26 @@ class TaskExecutor:
             ))
         load_budget, run_budget = _budgets(assignment)
         await report.loading(0.0, f"preparing {assignment.engine}")
-        backend = await self._bounded(
-            asyncio.to_thread(self._load_backend, assignment.engine),
+        backend = await self._bounded_thread(
+            self._load_backend,
+            assignment.engine,
             timeout=load_budget, code="MODEL_LOAD_TIMEOUT",
             what=f"Loading '{assignment.engine}'",
         )
         await report.loading(1.0, "model ready")
         await report.progress(0.05, "synthesising chapter")
-        audio = await self._bounded(
-            asyncio.to_thread(self._synthesize_audiobook, backend, spans, voices, params),
+        audio = await self._bounded_thread(
+            self._synthesize_audiobook,
+            backend,
+            spans,
+            voices,
+            params,
             timeout=run_budget, code="EXECUTION_TIMEOUT", what="Audiobook chapter",
         )
         await report.progress(0.9, "encoding")
-        payload, meta = await asyncio.to_thread(self._encode, audio, params, backend)
+        payload, meta = await self._thread_call(
+            self._encode, audio, params, backend
+        )
         await report.progress(1.0, "done")
         return {"meta": meta, "payload": payload}
 
@@ -294,7 +339,10 @@ class TaskExecutor:
             key: value for key, value in opts.to_manifest().items()
             if value is not None and key not in ("seed", "vary_repeats")
         }
-        if isinstance(backend, OmniVoiceBackend):
+        native_proxy = bool(
+            getattr(backend, "supports_native_omnivoice_controls", False)
+        )
+        if isinstance(backend, OmniVoiceBackend) or native_proxy:
             extra.setdefault("num_step", 32)
             extra.setdefault("guidance_scale", 2.0)
             for key in ("emo_vector", "emo_text", "emo_alpha"):
@@ -303,11 +351,13 @@ class TaskExecutor:
         def synth(text, index, speed=None):
             voice = voices[int(index)]
             base_seed = opts.seed if opts.seed is not None else voice.get("seed")
+            seed = None
             if base_seed is not None:
                 import torch
                 nonce = occurrence["value"] if opts.vary_repeats else 0
                 occurrence["value"] += 1
-                torch.manual_seed(segment_seed(base_seed, text, nonce))
+                seed = segment_seed(base_seed, text, nonce)
+                torch.manual_seed(seed)
             kwargs = {
                 "language": language,
                 "ref_audio": voice.get("ref_audio"),
@@ -316,6 +366,8 @@ class TaskExecutor:
                 "speed": float(speed) if speed else 1.0,
                 **extra,
             }
+            if native_proxy and seed is not None:
+                kwargs["seed"] = seed
             return backend.generate(text, **kwargs)
 
         spans = [Span(voice_id=str(i), text=row.get("text", ""),
@@ -329,7 +381,9 @@ class TaskExecutor:
 
     # ── Inputs ────────────────────────────────────────────────────────────
 
-    async def _materialize_inputs(self, assignment, params: dict, fetch) -> dict:
+    async def _materialize_inputs(
+        self, assignment, params: dict, fetch
+    ) -> tuple[dict, list[str]]:
         """Turn declared inputs into local files, then point the params at them.
 
         The control plane sends artifact ids, never paths — its own paths mean
@@ -352,7 +406,7 @@ class TaskExecutor:
 
         refs = [ref for ref in (getattr(assignment, "inputs", None) or []) if ref.artifact_id]
         if not refs:
-            return params
+            return params, []
         if fetch is None:
             raise TaskFailure(
                 WorkerError(
@@ -365,16 +419,24 @@ class TaskExecutor:
 
         _, run_budget = _budgets(assignment)
         local: dict[str, str] = {}
-        for ref in refs:
-            local[ref.artifact_id] = await self._bounded(
-                self._fetch_one(ref, fetch),
-                timeout=min(run_budget, _FALLBACK_INPUT_FETCH_SECONDS),
-                code="INPUT_FETCH_TIMEOUT",
-                what=f"Fetching '{ref.filename or ref.artifact_id}'",
-            )
-        return _rewrite_params(params, local)
+        leased: list[str] = []
+        try:
+            for ref in refs:
+                path = await self._bounded(
+                    self._fetch_one(ref, fetch, retain=True),
+                    timeout=min(run_budget, _FALLBACK_INPUT_FETCH_SECONDS),
+                    code="INPUT_FETCH_TIMEOUT",
+                    what=f"Fetching '{ref.filename or ref.artifact_id}'",
+                )
+                local[ref.artifact_id] = path
+                leased.append(path)
+            return _rewrite_params(params, local), leased
+        except BaseException:
+            for path in leased:
+                _release_input_cache_path(path)
+            raise
 
-    async def _fetch_one(self, ref, fetch) -> str:
+    async def _fetch_one(self, ref, fetch, *, retain: bool = False) -> str:
         """The local copy of one input, downloaded only if we lack it.
 
         Content-addressed: the name is the hash the control plane computed, so
@@ -382,37 +444,127 @@ class TaskExecutor:
         worker — costs no transfer at all.
         """
         directory = self._input_dir or default_input_dir()
-        os.makedirs(directory, exist_ok=True)
+        await self._thread_call(_durable_makedirs, directory)
         destination = os.path.join(directory, _cache_name(ref))
-        if _already_held(destination, ref):
-            _touch(destination)
-            return destination
+        return await self._fetch_one_owned(
+            ref,
+            fetch,
+            directory=directory,
+            destination=destination,
+            retain=retain,
+        )
 
-        partial = f"{destination}.{uuid.uuid4().hex}.part"
+    async def _fetch_one_owned(
+        self,
+        ref,
+        fetch,
+        *,
+        directory: str,
+        destination: str,
+        retain: bool,
+    ) -> str:
+        """Validate, fetch, and safely publish one content address."""
+        await _acquire_input_cache_path(destination, fetching=True)
+        leased_result = destination
+        succeeded = False
         try:
-            await fetch(ref, partial)
-        except TaskFailure:
-            raise
-        except Exception as exc:
-            _discard(partial)
-            raise TaskFailure(
-                WorkerError(
-                    # Transient on purpose: an id we cannot resolve now is far
-                    # more often a dropped stream than a permanently missing
-                    # file, and one wasted retry beats failing real work.
-                    error_class=ErrorClass.TRANSIENT,
-                    code="INPUT_FETCH_FAILED",
-                    message=f"Could not fetch '{ref.filename or ref.artifact_id}': {exc}",
-                    hint="The control plane may have restarted; the task will be retried.",
-                )
-            ) from exc
+            # Cache hits still hash the advertised content identity. Filename
+            # plus size is not proof after disk corruption or external edits.
+            if await self._thread_call(_already_held, destination, ref):
+                await self._thread_call(_touch, destination)
+                succeeded = True
+                return destination
 
-        # Off the loop: hashing a source video on the event loop thread would
-        # stall every heartbeat this worker owes the control plane.
-        await asyncio.to_thread(_verify, partial, ref)
-        os.replace(partial, destination)
-        await asyncio.to_thread(_prune_input_cache, directory)
-        return destination
+            partial = f"{destination}.{uuid.uuid4().hex}.part"
+            _lease_input_cache_path(partial)
+            finalized = False
+            try:
+                try:
+                    await fetch(ref, partial)
+                except TaskFailure:
+                    raise
+                except Exception as exc:
+                    raise _input_fetch_failure(ref, exc) from exc
+
+                # Hashing and durability barriers can both block on a large
+                # source or slow disk. Keep them off the loop and drain before
+                # cleanup so Windows never unlinks a file still in use.
+                await self._thread_call(_verify, partial, ref)
+                gate_key = _cache_path_key(destination)
+                gate = _INPUT_CACHE_FETCH_LOCKS.setdefault(
+                    gate_key, asyncio.Lock()
+                )
+                _INPUT_CACHE_FETCH_USERS[gate_key] = (
+                    _INPUT_CACHE_FETCH_USERS.get(gate_key, 0) + 1
+                )
+                try:
+                    async with gate:
+                        # A concurrent fetch may have published these exact
+                        # bytes while this one was downloading its own partial.
+                        if await self._thread_call(
+                            _already_held, destination, ref
+                        ):
+                            await self._thread_call(_discard, partial)
+                            finalized = True
+                        elif _claim_input_cache_mutation(destination):
+                            try:
+                                await self._thread_call(
+                                    _durable_replace, partial, destination
+                                )
+                            finally:
+                                _finish_input_cache_mutation(destination)
+                            finalized = True
+                        else:
+                            # Another execution is actively reading the
+                            # canonical generation. Never unlink or replace
+                            # bytes underneath it; publish this verified fetch
+                            # under a leased sibling path and let a later
+                            # unshared fetch repair canonical.
+                            stem, suffix = os.path.splitext(destination)
+                            alternate = (
+                                f"{stem}.{uuid.uuid4().hex}.generation{suffix}"
+                            )
+                            await _acquire_input_cache_path(
+                                alternate, fetching=True
+                            )
+                            try:
+                                await self._thread_call(
+                                    _durable_replace, partial, alternate
+                                )
+                            except BaseException:
+                                _release_input_cache_path(
+                                    alternate, fetching=True
+                                )
+                                await self._thread_call(_discard, alternate)
+                                raise
+                            finalized = True
+                            _release_input_cache_path(
+                                destination, fetching=True
+                            )
+                            leased_result = alternate
+                except OSError as exc:
+                    raise _input_fetch_failure(ref, exc) from exc
+                finally:
+                    remaining = _INPUT_CACHE_FETCH_USERS[gate_key] - 1
+                    if remaining:
+                        _INPUT_CACHE_FETCH_USERS[gate_key] = remaining
+                    else:
+                        _INPUT_CACHE_FETCH_USERS.pop(gate_key, None)
+                        if _INPUT_CACHE_FETCH_LOCKS.get(gate_key) is gate:
+                            _INPUT_CACHE_FETCH_LOCKS.pop(gate_key, None)
+            finally:
+                if not finalized:
+                    await self._thread_call(_discard, partial)
+                _release_input_cache_path(partial)
+
+            await self._thread_call(_prune_input_cache, directory)
+            succeeded = True
+            return leased_result
+        finally:
+            if retain and succeeded:
+                _promote_input_cache_lease(leased_result)
+            else:
+                _release_input_cache_path(leased_result, fetching=True)
 
     # ── Engine plumbing ───────────────────────────────────────────────────
 
@@ -508,6 +660,10 @@ class TaskExecutor:
                     duration, num_step, guidance_scale, speed, denoise,
                     postprocess_output, used_seed, effect_preset,
                     max_chunk_chars, crossfade_ms,
+                    t_shift=params.get("t_shift"),
+                    layer_penalty_factor=params.get("layer_penalty_factor"),
+                    position_temperature=params.get("position_temperature"),
+                    class_temperature=params.get("class_temperature"),
                 )
         except Exception as exc:
             from worker import errors as worker_errors  # noqa: PLC0415
@@ -551,6 +707,92 @@ class TaskExecutor:
 
     # ── Bounding ──────────────────────────────────────────────────────────
 
+    def _release_inputs_after_active_work(self, paths: list[str]) -> None:
+        """Keep files leased while a timed-out engine thread still owns them."""
+        pending = [task for task in self._blocking_tasks if not task.done()]
+        if not pending:
+            for path in paths:
+                _release_input_cache_path(path)
+            return
+        remaining = {"count": len(pending)}
+
+        def finished(_task: asyncio.Task) -> None:
+            remaining["count"] -= 1
+            if remaining["count"] == 0:
+                for path in paths:
+                    _release_input_cache_path(path)
+
+        for task in pending:
+            task.add_done_callback(finished)
+
+    def _start_thread(self, function, /, *args) -> asyncio.Task:
+        task = asyncio.create_task(asyncio.to_thread(function, *args))
+        self._blocking_tasks.add(task)
+
+        def finished(completed: asyncio.Task) -> None:
+            self._blocking_tasks.discard(completed)
+            if not completed.cancelled():
+                # Timed-out calls intentionally finish in the background. Read
+                # their exception so asyncio never reports an unowned task.
+                completed.exception()
+
+        task.add_done_callback(finished)
+        return task
+
+    async def _thread_call(self, function, /, *args):
+        task = self._start_thread(function, *args)
+        try:
+            return await asyncio.shield(task)
+        except asyncio.CancelledError:
+            await drain_task(task)
+            raise
+
+    async def drain_active_work(self) -> None:
+        """Wait until every blocking engine call has relinquished the process."""
+        cancelled = bool(
+            (current := asyncio.current_task()) is not None and current.cancelling()
+        )
+        while self._blocking_tasks:
+            for task in list(self._blocking_tasks):
+                try:
+                    await asyncio.shield(task)
+                except asyncio.CancelledError:
+                    # Cancellation cannot make a Python GPU thread stop. Hold
+                    # authority until it really exits, then propagate the
+                    # cancellation so callers never publish a false free slot.
+                    cancelled = True
+                    await drain_task(task)
+                except BaseException:
+                    # The owner reports/classifies the engine exception. This
+                    # barrier only establishes that the thread has finished.
+                    pass
+        if cancelled:
+            raise asyncio.CancelledError
+
+    async def _bounded_thread(
+        self, function, /, *args, timeout: float, code: str, what: str
+    ):
+        """Bound a blocking call without losing ownership of its live thread."""
+        task = self._start_thread(function, *args)
+        try:
+            done, _pending = await asyncio.wait({task}, timeout=timeout)
+        except asyncio.CancelledError:
+            await drain_task(task)
+            raise
+        if done:
+            return task.result()
+        # A GPU call cannot be killed. Return the timeout so the scheduler can
+        # park its slot, but retain the task above so terminal authority loss
+        # can drain it before claiming this worker has stopped.
+        raise TaskFailure(
+            WorkerError(
+                error_class=ErrorClass.TIMEOUT,
+                code=code,
+                message=f"{what} exceeded the {timeout:g}s budget for this task.",
+                hint="Try a shorter input, or a worker with more headroom.",
+            )
+        )
+
     @staticmethod
     async def _bounded(coro, *, timeout: float, code: str, what: str):
         """Run ``coro`` under the server's budget for this phase.
@@ -636,6 +878,158 @@ def default_input_dir() -> str:
         return os.path.join(tempfile.gettempdir(), "omnivoice-worker-inputs")
 
 
+def _cache_path_key(path: str) -> str:
+    return os.path.normcase(os.path.abspath(path))
+
+
+def _lease_input_cache_path(path: str, *, fetching: bool = False) -> bool:
+    key = _cache_path_key(path)
+    with _INPUT_CACHE_LEASE_LOCK:
+        if key in _INPUT_CACHE_MUTATIONS:
+            return False
+        _INPUT_CACHE_LEASES[key] = _INPUT_CACHE_LEASES.get(key, 0) + 1
+        if fetching:
+            _INPUT_CACHE_FETCH_LEASES[key] = (
+                _INPUT_CACHE_FETCH_LEASES.get(key, 0) + 1
+            )
+        return True
+
+
+async def _acquire_input_cache_path(
+    path: str, *, fetching: bool = False
+) -> None:
+    while not _lease_input_cache_path(path, fetching=fetching):
+        await asyncio.sleep(0.01)
+
+
+def _release_input_cache_path(path: str, *, fetching: bool = False) -> None:
+    key = _cache_path_key(path)
+    with _INPUT_CACHE_LEASE_LOCK:
+        if fetching:
+            fetch_remaining = _INPUT_CACHE_FETCH_LEASES.get(key, 0) - 1
+            if fetch_remaining > 0:
+                _INPUT_CACHE_FETCH_LEASES[key] = fetch_remaining
+            else:
+                _INPUT_CACHE_FETCH_LEASES.pop(key, None)
+        remaining = _INPUT_CACHE_LEASES.get(key, 0) - 1
+        if remaining > 0:
+            _INPUT_CACHE_LEASES[key] = remaining
+        else:
+            _INPUT_CACHE_LEASES.pop(key, None)
+
+
+def _promote_input_cache_lease(path: str) -> None:
+    """Turn a fetcher's lease into the active execution lease it returns."""
+    key = _cache_path_key(path)
+    with _INPUT_CACHE_LEASE_LOCK:
+        remaining = _INPUT_CACHE_FETCH_LEASES.get(key, 0) - 1
+        if remaining > 0:
+            _INPUT_CACHE_FETCH_LEASES[key] = remaining
+        else:
+            _INPUT_CACHE_FETCH_LEASES.pop(key, None)
+
+
+def _leased_input_cache_paths() -> set[str]:
+    with _INPUT_CACHE_LEASE_LOCK:
+        return set(_INPUT_CACHE_LEASES)
+
+
+def _claim_input_cache_mutation(path: str) -> bool:
+    key = _cache_path_key(path)
+    with _INPUT_CACHE_LEASE_LOCK:
+        if key in _INPUT_CACHE_MUTATIONS:
+            return False
+        active_leases = _INPUT_CACHE_LEASES.get(
+            key, 0
+        ) - _INPUT_CACHE_FETCH_LEASES.get(key, 0)
+        # Fetchers can safely converge under the publication gate. A lease
+        # already promoted to an execution may have this exact pathname open.
+        if active_leases > 0:
+            return False
+        _INPUT_CACHE_MUTATIONS.add(key)
+        return True
+
+
+def _finish_input_cache_mutation(path: str) -> None:
+    key = _cache_path_key(path)
+    with _INPUT_CACHE_LEASE_LOCK:
+        _INPUT_CACHE_MUTATIONS.discard(key)
+
+
+def _fsync_file(path: str) -> None:
+    with open(path, "r+b") as handle:
+        os.fsync(handle.fileno())
+
+
+def _fsync_parent_directory(directory: str) -> None:
+    directory_flag = getattr(os, "O_DIRECTORY", None)
+    if directory_flag is None:
+        return
+    unsupported = {
+        errno.EINVAL,
+        getattr(errno, "ENOTSUP", errno.EINVAL),
+        getattr(errno, "EOPNOTSUPP", errno.EINVAL),
+    }
+    try:
+        descriptor = os.open(directory, os.O_RDONLY | directory_flag)
+    except OSError as exc:
+        if exc.errno in unsupported:
+            return
+        raise
+    try:
+        os.fsync(descriptor)
+    except OSError as exc:
+        if exc.errno not in unsupported:
+            raise
+    finally:
+        os.close(descriptor)
+
+
+def _durable_makedirs(directory: str) -> None:
+    target = os.path.abspath(directory)
+    missing: list[str] = []
+    current = target
+    while not os.path.isdir(current):
+        if os.path.exists(current):
+            if os.path.isdir(current):
+                break
+            raise NotADirectoryError(current)
+        missing.append(current)
+        parent = os.path.dirname(current)
+        if parent == current:
+            break
+        current = parent
+    for path in reversed(missing):
+        try:
+            os.mkdir(path)
+        except FileExistsError:
+            if not os.path.isdir(path):
+                raise
+        _fsync_parent_directory(os.path.dirname(path) or ".")
+    if not missing:
+        _fsync_parent_directory(os.path.dirname(target) or ".")
+
+
+def _durable_replace(source: str, destination: str) -> None:
+    _fsync_file(source)
+    os.replace(source, destination)
+    _fsync_parent_directory(os.path.dirname(destination) or ".")
+
+
+def _input_fetch_failure(ref, error: BaseException) -> TaskFailure:
+    return TaskFailure(
+        WorkerError(
+            # Transient on purpose: an id we cannot resolve now is far more
+            # often a dropped stream/disk barrier than a permanently missing
+            # file, and one wasted retry beats failing real work.
+            error_class=ErrorClass.TRANSIENT,
+            code="INPUT_FETCH_FAILED",
+            message=f"Could not fetch '{ref.filename or ref.artifact_id}': {error}",
+            hint="The control plane may have restarted; the task will be retried.",
+        )
+    )
+
+
 def _cache_name(ref) -> str:
     """A safe, content-addressed local name for one input.
 
@@ -656,13 +1050,25 @@ def _cache_name(ref) -> str:
 def _already_held(path: str, ref) -> bool:
     """Do we already have this exact input?
 
-    Size alone: the name is the content hash and the only writer is an atomic
-    rename, so a file of the right size at this name cannot be different bytes.
+    The filename is content-addressed, but disks and external edits can still
+    change bytes at that name. Re-hash the advertised identity before reuse.
     """
     try:
         expected = int(getattr(ref, "size_bytes", 0) or 0)
-        return os.path.isfile(path) and (not expected or os.path.getsize(path) == expected)
-    except OSError:  # pragma: no cover
+        if not os.path.isfile(path):
+            return False
+        if expected and os.path.getsize(path) != expected:
+            return False
+        expected_hash = (getattr(ref, "sha256", "") or "").strip().lower()
+        if expected_hash:
+            digest = hashlib.sha256()
+            with open(path, "rb") as handle:
+                for block in iter(lambda: handle.read(1024 * 1024), b""):
+                    digest.update(block)
+            if digest.hexdigest() != expected_hash:
+                return False
+        return True
+    except OSError:
         return False
 
 
@@ -723,22 +1129,45 @@ def _verify(path: str, ref) -> None:
         )
 
 
-def _prune_input_cache(directory: str, limit_bytes: int = INPUT_CACHE_LIMIT_BYTES) -> None:
-    """Keep the input cache under its ceiling, oldest first."""
+def _prune_input_cache(
+    directory: str,
+    limit_bytes: int = INPUT_CACHE_LIMIT_BYTES,
+    now: Optional[float] = None,
+) -> None:
+    """Keep the cache bounded without deleting inputs a task is still using."""
     try:
         entries = []
         total = 0
+        stamp = time.time() if now is None else now
+        leased = _leased_input_cache_paths()
         for name in os.listdir(directory):
             path = os.path.join(directory, name)
-            if name.endswith(".part") or not os.path.isfile(path):
+            if not os.path.isfile(path):
                 continue
             stat = os.stat(path)
-            entries.append((stat.st_mtime, stat.st_size, path))
+            key = _cache_path_key(path)
+            is_partial = name.endswith(".part")
+            if (
+                is_partial
+                and key not in leased
+                and stamp - stat.st_mtime >= _STALE_INPUT_PARTIAL_SECONDS
+            ):
+                os.remove(path)
+                continue
             total += stat.st_size
+            # Active finals and partial transfers count toward the ceiling but
+            # cannot be evicted. Young unleased .part files may belong to a
+            # process that has not yet rebuilt its in-memory lease after fork;
+            # the age sweep will remove them if they are crash leftovers.
+            if key not in leased and not is_partial:
+                entries.append((stat.st_mtime, stat.st_size, path))
         for _mtime, size, path in sorted(entries):
             if total <= limit_bytes:
                 break
-            os.remove(path)
+            try:
+                os.remove(path)
+            except FileNotFoundError:
+                continue
             total -= size
     except OSError:  # pragma: no cover — a full cache is not a failed task
         logger.debug("Could not prune the worker input cache", exc_info=True)

backend/worker/identity.py

@@ -25,6 +25,7 @@ in the dialog that shows it once.
 from __future__ import annotations
 
 import base64
+import errno
 import hashlib
 import hmac
 import json
@@ -301,10 +302,31 @@ def save_worker_key(path: str, keypair: WorkerKeypair) -> None:
     tmp = f"{path}.tmp"
     fd = os.open(tmp, os.O_WRONLY | os.O_CREAT | os.O_TRUNC, 0o600)
     try:
-        os.write(fd, keypair.private_bytes())
-    finally:
+        remaining = memoryview(keypair.private_bytes())
+        while remaining:
+            written = os.write(fd, remaining)
+            if written <= 0:
+                raise OSError("could not finish writing the worker identity key")
+            remaining = remaining[written:]
+        os.fsync(fd)
+    except Exception:
         os.close(fd)
-    os.replace(tmp, path)
+        try:
+            os.unlink(tmp)
+        except FileNotFoundError:
+            pass
+        raise
+    else:
+        os.close(fd)
+    try:
+        os.replace(tmp, path)
+    except Exception:
+        try:
+            os.unlink(tmp)
+        except FileNotFoundError:
+            pass
+        raise
+    _fsync_parent_directory(directory)
     try:
         os.chmod(path, 0o600)
     except OSError:
@@ -313,6 +335,30 @@ def save_worker_key(path: str, keypair: WorkerKeypair) -> None:
         pass
 
 
+def _fsync_parent_directory(directory: str) -> None:
+    directory_flag = getattr(os, "O_DIRECTORY", None)
+    if directory_flag is None:
+        return
+    unsupported = {
+        errno.EINVAL,
+        getattr(errno, "ENOTSUP", errno.EINVAL),
+        getattr(errno, "EOPNOTSUPP", errno.EINVAL),
+    }
+    try:
+        descriptor = os.open(directory, os.O_RDONLY | directory_flag)
+    except OSError as exc:
+        if exc.errno in unsupported:
+            return
+        raise
+    try:
+        os.fsync(descriptor)
+    except OSError as exc:
+        if exc.errno not in unsupported:
+            raise
+    finally:
+        os.close(descriptor)
+
+
 def load_worker_key(path: str) -> Optional[WorkerKeypair]:
     try:
         with open(path, "rb") as fh:

backend/worker/inbound/connector.py

@@ -22,20 +22,106 @@ import logging
 import os
 import socket
 import ssl
-from typing import Optional
+from typing import BinaryIO, Optional
 
 import grpc
 
 from worker import identity, registry, tls
+from worker.async_utils import to_thread_and_drain_on_cancel
 from worker.inbound.connection_string import Connection
 from worker.inbound.listener import KEY_METADATA_KEY
 from worker.protocol.gen import worker_v1_pb2 as pb
 from worker.protocol.gen import worker_v1_pb2_grpc as pb_grpc
-from worker.transport.client import MAX_MESSAGE_BYTES, backoff_delay
+from worker.transport.client import (
+    MAX_MESSAGE_BYTES,
+    TerminalRegistrationError,
+    backoff_delay,
+)
 
 logger = logging.getLogger(__name__)
 
 _PUSH_CHUNK_BYTES = 1024 * 1024
+# Register remains provisional until the node confirms that it durably saved
+# the panel-assigned identity.  Match the control plane's provisional-session
+# lifetime so a peer that stops after Register cannot strand this connector.
+_REGISTRATION_CONFIRMATION_TIMEOUT_SECONDS = 30.0
+_REMOTE_SHUTDOWN_TIMEOUT_SECONDS = 30.0
+
+_FileVersion = tuple[int, int, int, int, int]
+
+
+def _write_all(handle: BinaryIO, payload: bytes) -> None:
+    remaining = memoryview(payload)
+    while remaining:
+        written = handle.write(remaining)
+        if not written:
+            raise OSError("result write made no progress")
+        remaining = remaining[written:]
+
+
+def _remove_quietly(path: str) -> None:
+    with contextlib.suppress(OSError):
+        os.remove(path)
+
+
+class InboundConnectionError(RuntimeError):
+    """A pasted inbound connection could not be validated or activated."""
+
+
+class InboundConnectionRollbackError(InboundConnectionError):
+    """A failed connection change could not restore its prior generation."""
+
+
+class RemoteShutdownUnavailable(InboundConnectionError):
+    """The node may retain work, but no live stream can revoke it safely."""
+
+
+def _file_version(stat: os.stat_result) -> _FileVersion:
+    """Fields that identify both a staged path and the bytes hashed from it."""
+    return (
+        int(stat.st_dev),
+        int(stat.st_ino),
+        int(stat.st_size),
+        int(stat.st_mtime_ns),
+        int(stat.st_ctime_ns),
+    )
+
+
+def _hash_staged_input(path: str) -> tuple[int, str, _FileVersion]:
+    """Hash one stable generation without ever allocating the whole file."""
+    digest = hashlib.sha256()
+    received = 0
+    with open(path, "rb") as handle:
+        before = _file_version(os.fstat(handle.fileno()))
+        while True:
+            block = handle.read(_PUSH_CHUNK_BYTES)
+            if not block:
+                break
+            received += len(block)
+            digest.update(block)
+        after = _file_version(os.fstat(handle.fileno()))
+    if before != after or received != before[2]:
+        raise RuntimeError("the staged task input changed while it was being hashed")
+    return received, digest.hexdigest(), before
+
+
+def _validate_staged_input(path: str, expected: _FileVersion) -> None:
+    """Reject a replacement or in-place edit between hashing and streaming."""
+    try:
+        current = _file_version(os.stat(path))
+    except OSError as exc:
+        raise RuntimeError("the staged task input is no longer available") from exc
+    if current != expected:
+        raise RuntimeError("the staged task input changed before it could be sent")
+
+
+def _validate_open_staged_input(
+    handle: BinaryIO, path: str, expected: _FileVersion
+) -> None:
+    """The open generation and its path must still be the bytes we hashed."""
+    if _file_version(os.fstat(handle.fileno())) != expected:
+        raise RuntimeError("the staged task input changed before it could be sent")
+    _validate_staged_input(path, expected)
 
 
 def _fetch_pinned_certificate(
@@ -70,9 +156,15 @@ class NodeConnection:
         self._connection = connection
         self._label = label or connection.host
         self._outbox: asyncio.Queue[pb.ServerMessage] = asyncio.Queue()
+        self._active_session = None
         self._stub: Optional[pb_grpc.NodeServiceStub] = None
         self._worker_id = ""
         self._stop = asyncio.Event()
+        self._session_closed = asyncio.Event()
+        self._session_closed.set()
+        self._shutdown_confirmed = asyncio.Event()
+        self._registration_ready = asyncio.Event()
+        self._remote_protocol_retained = False
         self._last_error = ""
 
     @property
@@ -105,6 +197,10 @@ class NodeConnection:
                 attempt = 0
             except asyncio.CancelledError:
                 raise
+            except TerminalRegistrationError as exc:
+                self._remote_protocol_retained = False
+                self._last_error = str(exc)
+                raise
             except Exception:
                 attempt += 1
                 self._last_error = "Connection failed; check the backend log for details."
@@ -114,8 +210,143 @@ class NodeConnection:
                     await asyncio.wait_for(self._stop.wait(), timeout=delay)
 
     async def stop(self) -> None:
+        if self._stop.is_set():
+            return
+        if self._shutdown_confirmed.is_set() and not self._remote_protocol_retained:
+            self._stop.set()
+            return
+        if self._active_session is None:
+            if self._remote_protocol_retained:
+                raise RemoteShutdownUnavailable(
+                    "That GPU machine is offline and may still be running work. "
+                    "Reconnect it, then remove the connection again."
+                )
+            self._stop.set()
+            return
+        # EOF is indistinguishable from a network blip and deliberately keeps
+        # node execution alive for reconnect.  Send an explicit terminal frame
+        # and wait for the node to drain before removal reports success.
+        self._shutdown_confirmed.clear()
+        await self._outbox.put(
+            pb.ServerMessage(
+                shutdown=pb.Shutdown(reason="This GPU-machine connection was removed.")
+            )
+        )
+        confirmed = asyncio.create_task(self._shutdown_confirmed.wait())
+        disconnected = asyncio.create_task(self._session_closed.wait())
+        try:
+            done, _pending = await asyncio.wait(
+                {confirmed, disconnected},
+                timeout=_REMOTE_SHUTDOWN_TIMEOUT_SECONDS,
+                return_when=asyncio.FIRST_COMPLETED,
+            )
+            if confirmed not in done and not self._shutdown_confirmed.is_set():
+                raise RemoteShutdownUnavailable(
+                    "The GPU machine disconnected before it confirmed shutdown. "
+                    "Reconnect it, then remove the connection again."
+                )
+        finally:
+            confirmed.cancel()
+            disconnected.cancel()
+            await asyncio.gather(confirmed, disconnected, return_exceptions=True)
+        self._remote_protocol_retained = False
         self._stop.set()
 
+    async def close(self) -> None:
+        """End this process without revoking reconnectable remote work."""
+        self._stop.set()
+
+    def confirm_remote_shutdown(self, session) -> None:
+        if self._active_session is not session:
+            return
+        self._remote_protocol_retained = False
+        self._shutdown_confirmed.set()
+
+    def confirm_registration(self, session) -> None:
+        """Publish readiness only after the shared servicer activated the session."""
+        if self._active_session is session:
+            self._registration_ready.set()
+
+    async def wait_until_registered(
+        self, task: asyncio.Task, *, timeout: float = 30.0
+    ) -> None:
+        """Wait for activation or surface a terminal/background dial failure."""
+        ready = asyncio.create_task(self._registration_ready.wait())
+        try:
+            done, _pending = await asyncio.wait(
+                {ready, task}, timeout=timeout, return_when=asyncio.FIRST_COMPLETED
+            )
+            if ready in done:
+                return
+            if task in done:
+                if task.cancelled():
+                    raise InboundConnectionError(
+                        "The GPU-machine connection stopped before it became ready."
+                    )
+                exc = task.exception()
+                if exc is not None:
+                    raise InboundConnectionError(str(exc)) from exc
+            raise InboundConnectionError(
+                "That GPU machine did not finish connecting in time."
+            )
+        finally:
+            ready.cancel()
+            await asyncio.gather(ready, return_exceptions=True)
+
+    async def probe(self) -> None:
+        """Authenticate a replacement paste without publishing a worker session."""
+        try:
+            certificate_pem = await asyncio.to_thread(
+                _fetch_pinned_certificate, self._connection
+            )
+            async with self._channel(certificate_pem) as channel:
+                stub = pb_grpc.NodeServiceStub(channel)
+                metadata = ((KEY_METADATA_KEY, self._connection.secret),)
+                stream = stub.Attach(self._outbound(), metadata=metadata)
+                try:
+                    first = await asyncio.wait_for(
+                        stream.read(),
+                        timeout=_REGISTRATION_CONFIRMATION_TIMEOUT_SECONDS,
+                    )
+                finally:
+                    stream.cancel()
+        except InboundConnectionError:
+            raise
+        except grpc.aio.AioRpcError as exc:
+            detail = exc.details() or "The GPU machine rejected this connection."
+            raise InboundConnectionError(detail) from exc
+        except asyncio.TimeoutError as exc:
+            raise InboundConnectionError(
+                "That GPU machine did not answer in time."
+            ) from exc
+        except Exception as exc:
+            raise InboundConnectionError(str(exc)) from exc
+
+        if first == grpc.aio.EOF or first.WhichOneof("payload") != "register":
+            raise InboundConnectionError(
+                "That machine answered, but not as a VoiceStudio GPU node."
+            )
+        request = first.register
+        validate = getattr(self._servicer, "validate_inbound_request", None)
+        refusal = validate(request) if callable(validate) else None
+        if refusal is not None and refusal.error.code:
+            raise InboundConnectionError(
+                f"{refusal.error.code}: {refusal.error.message}"
+            )
+        public_key = bytes(request.public_key)
+        if len(public_key) != 32:
+            raise InboundConnectionError("That machine sent no usable identity.")
+        key_id = identity.key_id_for(public_key)
+        if registry.is_revoked(key_id):
+            raise InboundConnectionError(
+                "This GPU machine was removed from this app. Add it again to use it."
+            )
+        known = registry.get_by_key_id(key_id)
+        if known is not None and not self._proves_key_possession(request, known):
+            raise InboundConnectionError(
+                "That machine could not prove its saved identity."
+            )
+
     async def _connect_once(self) -> None:
         # A fresh outbox per attempt. The queue used to be built once and
         # reused, so anything a dying session left behind became the NEXT
@@ -123,7 +354,10 @@ class NodeConnection:
         # registration it requires first, aborted the call, and the pair span
         # at full speed: on hardware this reached session epoch 2445 inside a
         # second, with the log reading "Locally aborted" over and over.
+        self._worker_id = ""
+        self._stub = None
         self._outbox = asyncio.Queue()
+        self._active_session = None
         certificate_pem = await asyncio.to_thread(
             _fetch_pinned_certificate, self._connection
         )
@@ -140,33 +374,98 @@ class NodeConnection:
                     "That machine answered, but not as a VoiceStudio GPU node."
                 )
 
-            response = self._register(first.register)
+            response = await self._register(first.register)
             if response.error.code:
                 # A refusal here is a decision, not a blip: the node is a
                 # different machine than the one this key was trusted for, or
                 # its version cannot work with ours. Reconnecting cannot fix
-                # either, so surface it rather than looping.
-                raise RuntimeError(f"{response.error.code}: {response.error.message}")
+                # either. Deliver the verdict before surfacing it locally so
+                # the node can retire work retained across the dead stream;
+                # closing first strands that executor with nobody left able to
+                # cancel it.
+                await self._outbox.put(pb.ServerMessage(registered=response))
+                try:
+                    await asyncio.wait_for(
+                        stream.read(),
+                        timeout=_REGISTRATION_CONFIRMATION_TIMEOUT_SECONDS,
+                    )
+                except (asyncio.TimeoutError, grpc.aio.AioRpcError):
+                    pass
+                raise TerminalRegistrationError(
+                    f"{response.error.code}: {response.error.message}"
+                )
 
-            self._worker_id = response.worker_id
-            self._stub = stub
-            self._last_error = ""
             await self._outbox.put(pb.ServerMessage(registered=response))
-
-            session = self._servicer.session_for(self._worker_id)
-            if session is None:
-                raise RuntimeError("the session went away before the stream opened")
-
-            pump = asyncio.create_task(self._pump_outbound(session))
             try:
-                await self._servicer.run_inbound_stream(session, _Frames(stream), self)
+                await self._complete_registration(stream, response, stub)
             finally:
-                pump.cancel()
-                with contextlib.suppress(asyncio.CancelledError, Exception):
-                    await pump
-                self._stub = None
+                # Idempotent after activation; essential before it. A user can
+                # remove this connection while the node is still persisting
+                # identity, and cancellation must release the old worker's
+                # scheduling gate immediately rather than wait for expiry.
+                self._servicer.discard_unopened_session(
+                    response.worker_id, session_token=response.session_token
+                )
 
-    def _register(self, request: pb.RegisterRequest) -> pb.RegisterResponse:
+    async def _complete_registration(self, stream, response, stub) -> None:
+        """Validate durable acceptance, then run the exact issued session."""
+        try:
+            confirmation = await asyncio.wait_for(
+                stream.read(), timeout=_REGISTRATION_CONFIRMATION_TIMEOUT_SECONDS
+            )
+        except asyncio.TimeoutError as exc:
+            raise RuntimeError(
+                "That GPU machine did not confirm registration in time."
+            ) from exc
+        except grpc.aio.AioRpcError as exc:
+            detail = exc.details() or ""
+            error_code = detail.partition(":")[0].strip()
+            if exc.code() == grpc.StatusCode.FAILED_PRECONDITION and error_code in {
+                "AUTH_FAILED",
+                "LOCAL_STATE",
+                "UPGRADE_REQUIRED",
+            }:
+                raise TerminalRegistrationError(detail) from exc
+            raise
+        if confirmation == grpc.aio.EOF:
+            raise RuntimeError(
+                "That GPU machine disconnected before confirming registration."
+            )
+        if confirmation.WhichOneof("payload") != "heartbeat":
+            raise RuntimeError(
+                "That GPU machine sent an invalid registration confirmation."
+            )
+
+        session = self._servicer.session_for(
+            response.worker_id, session_token=response.session_token
+        )
+        if session is None:
+            raise RuntimeError("the session went away before the stream opened")
+
+        self._worker_id = response.worker_id
+        self._stub = stub
+        self._last_error = ""
+        self._active_session = session
+        self._remote_protocol_retained = True
+        self._shutdown_confirmed.clear()
+        self._session_closed.clear()
+
+        pump = asyncio.create_task(self._pump_outbound(session))
+        try:
+            await self._servicer.run_inbound_stream(
+                session, _Frames(stream, first=confirmation), self
+            )
+        finally:
+            pump.cancel()
+            with contextlib.suppress(asyncio.CancelledError, Exception):
+                await pump
+            self._stub = None
+            self._worker_id = ""
+            if self._active_session is session:
+                self._active_session = None
+            self._session_closed.set()
+
+    async def _register(self, request: pb.RegisterRequest) -> pb.RegisterResponse:
         """Trust on first sight, then require the same key forever after.
 
         Pasting the connection string is the consent — the user went to the
@@ -175,14 +474,28 @@ class NodeConnection:
         is a licence for a different machine to answer at that address later,
         which is why the key is bound on first contact.
         """
+        refusal = self._servicer.validate_inbound_request(request)
+        if refusal is not None:
+            return refusal
+        worker, refusal = await to_thread_and_drain_on_cancel(
+            self._authenticate_registration, request
+        )
+        if refusal is not None:
+            return refusal
+        return await self._servicer.establish_session(
+            worker, request, address=self._connection.endpoint
+        )
+
+    def _authenticate_registration(self, request: pb.RegisterRequest):
+        """Resolve inbound identity without running SQLite on the app loop."""
         public_key = bytes(request.public_key)
         if len(public_key) != 32:
-            return self._servicer._refuse(
+            return None, self._servicer._refuse(
                 "AUTH_FAILED", "That machine sent no usable identity."
             )
         key_id = identity.key_id_for(public_key)
         if registry.is_revoked(key_id):
-            return self._servicer._refuse(
+            return None, self._servicer._refuse(
                 "AUTH_FAILED",
                 "This GPU machine was removed from this app. Add it again to use it.",
             )
@@ -219,14 +532,12 @@ class NodeConnection:
                 )
                 worker = known
             if worker is None:
-                return self._servicer._refuse(
+                return None, self._servicer._refuse(
                     "AUTH_FAILED",
                     "That machine could not prove it is the one this key was added for.",
                 )
 
-        return self._servicer.register_inbound(
-            worker, request, address=self._connection.endpoint
-        )
+        return worker, None
 
     @staticmethod
     def _proves_key_possession(request: pb.RegisterRequest, known) -> bool:
@@ -249,9 +560,34 @@ class NodeConnection:
             public_key, message, bytes(request.challenge_signature)
         )
 
-    async def _outbound(self):
+    def _outbound(self):
+        # grpc closes request iterators itself when the peer ends a stream. An
+        # async generator can still be suspended in ``Queue.get`` at that
+        # point, making its concurrent ``aclose`` fail and leak teardown into
+        # the next channel. A plain async iterator has no generator-finalizer
+        # race and keeps the same one-frame-at-a-time backpressure.
+        return _OutboundFrames(self)
+
+    def fence_session_egress(self, session) -> None:
+        """Drop frames copied before a replacement generation activated."""
+        if self._active_session is not session:
+            return
         while True:
-            yield await self._outbox.get()
+            try:
+                self._outbox.get_nowait()
+            except asyncio.QueueEmpty:
+                break
+
+    def revoke_session(self, session) -> None:
+        """Synchronously fence frames already copied into the request queue."""
+        if self._active_session is not session:
+            return
+        while True:
+            try:
+                self._outbox.get_nowait()
+            except asyncio.QueueEmpty:
+                break
+        self._outbox.put_nowait(None)
 
     async def _pump_outbound(self, session) -> None:
         """Move the servicer's per-session outbox onto the dialled stream.
@@ -260,8 +596,18 @@ class NodeConnection:
         to cross into the request generator instead, because this side is the
         caller.
         """
-        while True:
-            await self._outbox.put(await session.outbox.get())
+        task = asyncio.current_task()
+        if task is not None:
+            session.egress_tasks.add(task)
+        try:
+            while not session.revoked and not getattr(session, "egress_fenced", False):
+                message = await session.outbox.get()
+                if session.revoked or getattr(session, "egress_fenced", False):
+                    return
+                await self._outbox.put(message)
+        finally:
+            if task is not None:
+                session.egress_tasks.discard(task)
 
     # ── Artifacts ─────────────────────────────────────────────────────────
 
@@ -277,32 +623,52 @@ class NodeConnection:
         if stub is None:
             raise RuntimeError("that GPU machine is not connected")
 
-        size = os.path.getsize(path)
-        digest = hashlib.sha256()
-        with open(path, "rb") as handle:
-            digest.update(handle.read())
+        size, digest, version = await to_thread_and_drain_on_cancel(
+            _hash_staged_input, path
+        )
+        # Hashing and the gRPC request are separate operations. Re-resolve the
+        # path immediately before handing the iterator to gRPC so a replaced
+        # staging file is never described by the old generation's digest.
+        await to_thread_and_drain_on_cancel(_validate_staged_input, path, version)
 
         declared = pb.ArtifactRef()
         declared.CopyFrom(ref)
         declared.size_bytes = size
-        declared.sha256 = digest.hexdigest()
+        declared.sha256 = digest
         if not declared.filename:
             declared.filename = os.path.basename(path)
 
         async def chunks():
             offset = 0
-            with open(path, "rb") as handle:
-                while True:
-                    data = handle.read(_PUSH_CHUNK_BYTES)
+            handle = await to_thread_and_drain_on_cancel(open, path, "rb")
+            try:
+                await to_thread_and_drain_on_cancel(
+                    _validate_open_staged_input, handle, path, version
+                )
+                while offset < size:
+                    data = await to_thread_and_drain_on_cancel(
+                        handle.read, min(_PUSH_CHUNK_BYTES, size - offset)
+                    )
                     if not data:
-                        break
+                        raise RuntimeError(
+                            "the staged task input changed before it could be sent"
+                        )
                     offset += len(data)
+                    last = offset == size
+                    if last:
+                        # Do not publish the terminal frame until both the open
+                        # generation and its path still match what was hashed.
+                        await to_thread_and_drain_on_cancel(
+                            _validate_open_staged_input, handle, path, version
+                        )
                     yield pb.ArtifactChunk(
                         ref=declared,
                         offset=offset - len(data),
                         data=data,
-                        last=offset >= size,
+                        last=last,
                     )
+            finally:
+                await to_thread_and_drain_on_cancel(handle.close)
 
         ack = await stub.PushInput(
             chunks(), metadata=((KEY_METADATA_KEY, self._connection.secret),)
@@ -327,7 +693,11 @@ class NodeConnection:
         offset = 0
         complete = False
         try:
-            with open(destination, "wb") as handle:
+            handle = None
+            try:
+                handle = await to_thread_and_drain_on_cancel(
+                    open, destination, "wb"
+                )
                 async for chunk in stub.FetchResult(
                     request, metadata=((KEY_METADATA_KEY, self._connection.secret),)
                 ):
@@ -339,27 +709,31 @@ class NodeConnection:
                         raise RuntimeError(
                             "the result is larger than the control plane accepts"
                         )
-                    handle.write(chunk.data)
-                    digest.update(chunk.data)
-                    offset += len(chunk.data)
+                    data = bytes(chunk.data)
+                    await to_thread_and_drain_on_cancel(_write_all, handle, data)
+                    digest.update(data)
+                    offset += len(data)
                     if chunk.last:
                         complete = True
                         break
+            finally:
+                if handle is not None:
+                    await to_thread_and_drain_on_cancel(handle.close)
+        except asyncio.CancelledError:
+            await to_thread_and_drain_on_cancel(_remove_quietly, destination)
+            raise
         except Exception:
-            with contextlib.suppress(OSError):
-                os.remove(destination)
+            await to_thread_and_drain_on_cancel(_remove_quietly, destination)
             raise
 
         # A truncated file that is renamed into place and called done is the
         # exact failure the upload path was hardened against; the pull
         # direction gets the same treatment.
         if not complete:
-            with contextlib.suppress(OSError):
-                os.remove(destination)
+            await to_thread_and_drain_on_cancel(_remove_quietly, destination)
             raise RuntimeError("the result ended before its final chunk")
         if ref.sha256 and digest.hexdigest() != ref.sha256:
-            with contextlib.suppress(OSError):
-                os.remove(destination)
+            await to_thread_and_drain_on_cancel(_remove_quietly, destination)
             raise RuntimeError(
                 "the result did not match the checksum that machine declared"
             )
@@ -368,14 +742,46 @@ class NodeConnection:
 class _Frames:
     """Adapts a gRPC client stream to the ``async for`` the read loop expects."""
 
-    def __init__(self, stream) -> None:
+    def __init__(self, stream, *, first=None) -> None:
         self._stream = stream
+        self._first = first
 
     def __aiter__(self):
         return self
 
     async def __anext__(self):
+        if self._first is not None:
+            message = self._first
+            self._first = None
+            return message
         message = await self._stream.read()
         if message == grpc.aio.EOF:
             raise StopAsyncIteration
         return message
+
+
+class _OutboundFrames:
+    """Cancellation-safe request iterator for the inverted Attach stream."""
+
+    def __init__(self, connection: NodeConnection) -> None:
+        self._connection = connection
+
+    def __aiter__(self):
+        return self
+
+    async def __anext__(self):
+        connection = self._connection
+        while True:
+            message = await connection._outbox.get()
+            if message is None:
+                raise StopAsyncIteration
+            session = connection._active_session
+            if session is not None:
+                if session.revoked:
+                    raise StopAsyncIteration
+                if (
+                    getattr(session, "egress_fenced", False)
+                    and message.WhichOneof("payload") != "shutdown"
+                ):
+                    continue
+            return message

backend/worker/inbound/keys.py

@@ -15,6 +15,7 @@ settings store.
 
 from __future__ import annotations
 
+import errno
 import json
 import logging
 import os
@@ -45,6 +46,20 @@ _MAX_FAILURES = 5
 _LOCKOUT_SECONDS = 60.0
 _FAILURE_WINDOW_SECONDS = 300.0
 
+# ``Attach`` is the only RPC that records presence.  Persisting on every
+# reconnect lets an authenticated peer turn harmless telemetry into an fsync
+# storm on the gRPC event loop, so coalesce it to a useful reporting cadence.
+_LAST_SEEN_PERSIST_INTERVAL_SECONDS = 60.0
+
+# Authentication deliberately scans every stored hash in constant time. Keep
+# that work and the JSON credential file bounded even if an administrator
+# repeatedly issues replacements.
+MAX_PANEL_KEYS = 256
+
+
+class KeyLimitExceeded(RuntimeError):
+    """No additional panel credential can be retained safely."""
+
 
 @dataclass
 class PanelKey:
@@ -89,6 +104,31 @@ def _peer_host(peer: str) -> str:
     return peer
 
 
+def _fsync_parent_directory(directory: str) -> None:
+    """Make a preceding directory-entry replacement durable when supported."""
+    directory_flag = getattr(os, "O_DIRECTORY", None)
+    if directory_flag is None:
+        return
+    unsupported = {
+        errno.EINVAL,
+        getattr(errno, "ENOTSUP", errno.EINVAL),
+        getattr(errno, "EOPNOTSUPP", errno.EINVAL),
+    }
+    try:
+        descriptor = os.open(directory, os.O_RDONLY | directory_flag)
+    except OSError as exc:
+        if exc.errno in unsupported:
+            return
+        raise
+    try:
+        os.fsync(descriptor)
+    except OSError as exc:
+        if exc.errno not in unsupported:
+            raise
+    finally:
+        os.close(descriptor)
+
+
 @dataclass
 class IssuedKey:
     """The one and only time the plaintext exists outside the caller's hands."""
@@ -113,6 +153,10 @@ class KeyStore:
         self._connection_secrets: dict[str, str] = {}
         self._connection_fingerprints: dict[str, str] = {}
         self._failures: dict[str, _Failures] = {}
+        # A failed persistence attempt must remain denied in this process but
+        # still be retryable.  Keeping this separate from PanelKey.revoked lets
+        # the next DELETE attempt write the durable transition again.
+        self._pending_revocations: set[str] = set()
         self._load()
 
     # ── Persistence ───────────────────────────────────────────────────────
@@ -170,10 +214,31 @@ class KeyStore:
         # `identity.save_worker_key` uses for the Ed25519 private key.
         fd = os.open(tmp, os.O_WRONLY | os.O_CREAT | os.O_TRUNC, 0o600)
         try:
-            os.write(fd, payload)
-        finally:
+            remaining = memoryview(payload)
+            while remaining:
+                written = os.write(fd, remaining)
+                if written <= 0:
+                    raise OSError("could not finish writing the inbound key file")
+                remaining = remaining[written:]
+            os.fsync(fd)
+        except Exception:
             os.close(fd)
-        os.replace(tmp, self._path)
+            try:
+                os.unlink(tmp)
+            except FileNotFoundError:
+                pass
+            raise
+        else:
+            os.close(fd)
+        try:
+            os.replace(tmp, self._path)
+        except Exception:
+            try:
+                os.unlink(tmp)
+            except FileNotFoundError:
+                pass
+            raise
+        _fsync_parent_directory(directory)
         try:
             os.chmod(self._path, 0o600)
         except OSError:
@@ -196,8 +261,37 @@ class KeyStore:
             created_at=now,
         )
         with self._lock:
+            previous = self._keys.get(key.key_id)
+            pruned: dict[str, PanelKey] = {}
+            if previous is None and len(self._keys) >= MAX_PANEL_KEYS:
+                revoked = sorted(
+                    (
+                        stored
+                        for stored in self._keys.values()
+                        if stored.revoked
+                        and stored.key_id not in self._pending_revocations
+                    ),
+                    key=lambda stored: stored.created_at,
+                )
+                while len(self._keys) >= MAX_PANEL_KEYS and revoked:
+                    stale = revoked.pop(0)
+                    pruned[stale.key_id] = self._keys.pop(stale.key_id)
+                if len(self._keys) >= MAX_PANEL_KEYS:
+                    self._keys.update(pruned)
+                    raise KeyLimitExceeded(
+                        "This GPU machine already has as many panel keys as it accepts. "
+                        "Revoke an unused key, then try again."
+                    )
             self._keys[key.key_id] = key
-            self._save_locked()
+            try:
+                self._save_locked()
+            except Exception:
+                if previous is None:
+                    self._keys.pop(key.key_id, None)
+                else:
+                    self._keys[key.key_id] = previous
+                self._keys.update(pruned)
+                raise
         return IssuedKey(key=key, secret=secret)
 
     def revoke(self, key_id: str) -> bool:
@@ -206,8 +300,14 @@ class KeyStore:
             key = self._keys.get(key_id)
             if key is None or key.revoked:
                 return False
+            self._pending_revocations.add(key_id)
             key.revoked = True
-            self._save_locked()
+            try:
+                self._save_locked()
+            except Exception:
+                key.revoked = False
+                raise
+            self._pending_revocations.discard(key_id)
         return True
 
     def remember_worker_id(self, key_id: str, worker_id: str) -> None:
@@ -216,15 +316,46 @@ class KeyStore:
             return
         with self._lock:
             key = self._keys.get(key_id)
-            if key is None or key.worker_id == worker_id:
+            if (
+                key is None
+                or key.revoked
+                or key_id in self._pending_revocations
+            ):
+                raise PermissionError("the panel key was revoked during registration")
+            if key.worker_id == worker_id:
                 return
+            previous_worker_id = key.worker_id
             key.worker_id = worker_id
-            self._save_locked()
+            try:
+                self._save_locked()
+            except Exception:
+                # A callback retry must attempt the durable write again. If
+                # the failed value remains in memory, the equality fast path
+                # above accepts it as saved and the node reconnects with an id
+                # that disappears on process restart.
+                key.worker_id = previous_worker_id
+                raise
 
     def worker_id_for(self, key_id: str) -> str:
         with self._lock:
             key = self._keys.get(key_id)
-            return key.worker_id if key is not None else ""
+            return (
+                key.worker_id
+                if key is not None
+                and not key.revoked
+                and key_id not in self._pending_revocations
+                else ""
+            )
+
+    def is_active(self, key_id: str) -> bool:
+        """Whether this key still has authority to use an existing session."""
+        with self._lock:
+            key = self._keys.get(key_id)
+            return (
+                key is not None
+                and not key.revoked
+                and key_id not in self._pending_revocations
+            )
 
     def list_keys(self) -> list[dict]:
         with self._lock:
@@ -232,7 +363,10 @@ class KeyStore:
 
     def any_active(self) -> bool:
         with self._lock:
-            return any(not k.revoked for k in self._keys.values())
+            return any(
+                not key.revoked and key.key_id not in self._pending_revocations
+                for key in self._keys.values()
+            )
 
     # ── Panel-side connection credentials ───────────────────────────────
 
@@ -241,10 +375,23 @@ class KeyStore:
     ) -> None:
         """Persist a pasted node secret outside the UI-readable settings store."""
         with self._lock:
+            previous_secret = self._connection_secrets.get(endpoint)
+            previous_fingerprint = self._connection_fingerprints.get(endpoint)
             self._connection_secrets[endpoint] = secret
             if fingerprint:
                 self._connection_fingerprints[endpoint] = fingerprint
-            self._save_locked()
+            try:
+                self._save_locked()
+            except Exception:
+                if previous_secret is None:
+                    self._connection_secrets.pop(endpoint, None)
+                else:
+                    self._connection_secrets[endpoint] = previous_secret
+                if previous_fingerprint is None:
+                    self._connection_fingerprints.pop(endpoint, None)
+                else:
+                    self._connection_fingerprints[endpoint] = previous_fingerprint
+                raise
 
     def connection_secret(self, endpoint: str) -> str:
         with self._lock:
@@ -256,9 +403,19 @@ class KeyStore:
 
     def forget_connection_secret(self, endpoint: str) -> None:
         with self._lock:
-            if self._connection_secrets.pop(endpoint, None) is not None:
-                self._connection_fingerprints.pop(endpoint, None)
+            previous_secret = self._connection_secrets.get(endpoint)
+            if previous_secret is None:
+                return
+            previous_fingerprint = self._connection_fingerprints.get(endpoint)
+            self._connection_secrets.pop(endpoint, None)
+            self._connection_fingerprints.pop(endpoint, None)
+            try:
                 self._save_locked()
+            except Exception:
+                self._connection_secrets[endpoint] = previous_secret
+                if previous_fingerprint is not None:
+                    self._connection_fingerprints[endpoint] = previous_fingerprint
+                raise
 
     # ── Authentication ────────────────────────────────────────────────────
 
@@ -268,7 +425,9 @@ class KeyStore:
             record = self._failures.get(peer)
             return record is not None and record.locked_until > self._now()
 
-    def authenticate(self, secret: str, *, peer: str = "") -> Optional[PanelKey]:
+    def authenticate(
+        self, secret: str, *, peer: str = "", record_seen: bool = True
+    ) -> Optional[PanelKey]:
         """Return the matching live key, or None.
 
         Compares against every stored key in constant time and does not stop at
@@ -286,7 +445,11 @@ class KeyStore:
             candidate = hash_secret(secret) if secret else ""
             matched: Optional[PanelKey] = None
             for key in self._keys.values():
-                if key.revoked or not candidate:
+                if (
+                    key.revoked
+                    or key.key_id in self._pending_revocations
+                    or not candidate
+                ):
                     continue
                 if constant_time_equals(key.secret_hash, candidate):
                     matched = key
@@ -296,9 +459,21 @@ class KeyStore:
                 return None
 
             self._failures.pop(peer_host, None)
-            matched.last_seen_at = now
-            matched.last_seen_peer = peer
-            self._save_locked()
+            if record_seen and (
+                matched.last_seen_at <= 0.0
+                or now - matched.last_seen_at
+                >= _LAST_SEEN_PERSIST_INTERVAL_SECONDS
+            ):
+                previous_at = matched.last_seen_at
+                previous_peer = matched.last_seen_peer
+                matched.last_seen_at = now
+                matched.last_seen_peer = peer
+                try:
+                    self._save_locked()
+                except Exception:
+                    matched.last_seen_at = previous_at
+                    matched.last_seen_peer = previous_peer
+                    raise
             return matched
 
     def _record_failure_locked(self, peer: str, now: float) -> None:

backend/worker/inbound/service.py

@@ -14,11 +14,13 @@ second box — which is why neither implies the other.
 from __future__ import annotations
 
 import asyncio
+import ipaddress
 import logging
 import os
 from typing import Optional
 from urllib.parse import urlsplit
 
+from worker.async_utils import drain_task, to_thread_and_drain_on_cancel
 from worker.inbound.artifacts import ArtifactStore, KeyedArtifactTransport
 from worker.inbound.connection_log import ConnectionLog
 from worker.inbound.connection_string import (
@@ -27,6 +29,7 @@ from worker.inbound.connection_string import (
     format_connection,
     parse_connection,
 )
+from worker.inbound.connector import InboundConnectionRollbackError
 from worker.inbound.keys import KeyStore
 from worker.inbound.listener import DEFAULT_BIND, DEFAULT_PORT, NodeListener
 
@@ -38,6 +41,41 @@ _PORT_KEY = "inbound_node_port"
 _SAVED_KEY = "inbound_saved_nodes"
 
 
+async def _finish_rollback(rollback, *, description: str) -> None:
+    """Finish a lifecycle rollback even if its caller is cancelled again."""
+    task = asyncio.create_task(rollback, name="inbound-connection-rollback")
+    try:
+        await asyncio.shield(task)
+    except BaseException:
+        await drain_task(task)
+    if task.cancelled():
+        raise InboundConnectionRollbackError(
+            f"Could not {description}: rollback was cancelled."
+        )
+    return task.result()
+
+
+def _normalise_listener_host(value: str) -> str:
+    """Return the bare identity gRPC and X.509 expect for an IP literal."""
+    candidate = (value or "").strip()
+    inner = (
+        candidate[1:-1]
+        if len(candidate) >= 2
+        and candidate.startswith("[")
+        and candidate.endswith("]")
+        else candidate
+    )
+    try:
+        return str(ipaddress.ip_address(inner))
+    except ValueError:
+        return candidate
+
+
+def normalise_bind_host(value: str) -> str:
+    """Canonicalise a requested listener host before comparing or saving it."""
+    return _normalise_listener_host(value)
+
+
 def _setting(name: str, default: str = "") -> str:
     try:
         from services import settings_store  # noqa: PLC0415
@@ -86,13 +124,15 @@ def bind_host() -> str:
     point. Reaching this node from another machine should be a decision
     somebody made, not a side effect of turning the feature on.
     """
-    return (
-        os.environ.get("OMNIVOICE_INBOUND_BIND") or _setting(_BIND_KEY) or DEFAULT_BIND
+    return _normalise_listener_host(
+        os.environ.get("OMNIVOICE_INBOUND_BIND")
+        or _setting(_BIND_KEY)
+        or DEFAULT_BIND
     )
 
 
 def set_bind_host(value: str) -> None:
-    _set_setting(_BIND_KEY, (value or "").strip() or DEFAULT_BIND)
+    _set_setting(_BIND_KEY, _normalise_listener_host(value) or DEFAULT_BIND)
 
 
 def bind_port() -> int:
@@ -148,7 +188,9 @@ def is_exposed(host: Optional[str] = None) -> bool:
     connection string then admits clients beyond this machine. Transport
     remains pinned TLS (docs/adr/inbound-node-mode.md).
     """
-    return (host if host is not None else bind_host()) not in (
+    return _normalise_listener_host(
+        host if host is not None else bind_host()
+    ).lower() not in (
         "127.0.0.1",
         "localhost",
         "::1",
@@ -189,6 +231,7 @@ class InboundNode:
         self._keys: Optional[KeyStore] = None
         self._log = ConnectionLog()
         self._idle_sweep: Optional[asyncio.Task] = None
+        self._lifecycle_lock = asyncio.Lock()
         self.startup_error: Optional[str] = None
 
     @property
@@ -209,18 +252,12 @@ class InboundNode:
     def port(self) -> int:
         return self._listener.port if self._listener else 0
 
-    def _client_factory(self, artifacts: KeyedArtifactTransport, key_id: str):
-        # Imported here so a machine that never accepts connections does not
-        # pay for the executor or grpc at startup.
+    def _prepare_client(self, key_id: str) -> dict:
+        """Probe keys, host and accelerators away from the listener loop."""
         from worker import capabilities  # noqa: PLC0415
         from worker.agent import _paths as agent_paths  # noqa: PLC0415
-        from worker.executor import TaskExecutor  # noqa: PLC0415
         from worker.identity import load_or_create_worker_key  # noqa: PLC0415
-        from worker.transport.client import (  # noqa: PLC0415
-            WorkerClient,
-            WorkerConfig,
-            describe_host,
-        )
+        from worker.transport.client import describe_host  # noqa: PLC0415
 
         locations = agent_paths()
         os.makedirs(locations["root"], exist_ok=True)
@@ -228,6 +265,29 @@ class InboundNode:
         discovered = capabilities.discover(include_unavailable=True)
         host = describe_host()
         host["gpus"] = capabilities.describe_gpus()
+        return {
+            "keypair": keypair,
+            "discovered": discovered,
+            "host": host,
+            "worker_id": self.keys.worker_id_for(key_id),
+            "max_concurrent_tasks": capabilities.max_concurrent_tasks(discovered),
+        }
+
+    async def _client_factory(
+        self, artifacts: KeyedArtifactTransport, key_id: str
+    ):
+        # Imported here so a machine that never accepts connections does not
+        # pay for the executor or grpc at startup.
+        from worker import capabilities  # noqa: PLC0415
+        from worker.executor import TaskExecutor  # noqa: PLC0415
+        from worker.transport.client import (  # noqa: PLC0415
+            WorkerClient,
+            WorkerConfig,
+        )
+
+        prepared = await to_thread_and_drain_on_cancel(
+            self._prepare_client, key_id
+        )
 
         executor = TaskExecutor()
         return WorkerClient(
@@ -235,23 +295,28 @@ class InboundNode:
                 endpoint="",
                 cert_fingerprint="",
                 certificate_pem=b"",
-                keypair=keypair,
+                keypair=prepared["keypair"],
                 # Per panel key, not per node: each panel keeps its own
                 # registry, so the same machine is a different worker id to
                 # each of them, and the node signs its challenge over that id.
-                worker_id=self.keys.worker_id_for(key_id),
+                worker_id=prepared["worker_id"],
                 enrollment_token="",
-                max_concurrent_tasks=capabilities.max_concurrent_tasks(discovered),
-                capabilities=discovered,
-                host=host,
+                max_concurrent_tasks=prepared["max_concurrent_tasks"],
+                capabilities=prepared["discovered"],
+                host=prepared["host"],
             ),
             execute=executor.execute,
             capability_probe=lambda: capabilities.discover(include_unavailable=True),
             on_registered=lambda wid: self.keys.remember_worker_id(key_id, wid),
             artifacts=artifacts,
+            drain_active_work=executor.drain_active_work,
         )
 
     async def start(self) -> None:
+        async with self._lifecycle_lock:
+            await self._start()
+
+    async def _start(self) -> None:
         if self._listener is not None:
             return
         self.startup_error = None
@@ -264,9 +329,18 @@ class InboundNode:
         )
         try:
             await listener.start(host=bind_host(), port=bind_port())
+        except asyncio.CancelledError:
+            # NodeListener cleans a partially bound server before returning.
+            # If that cleanup itself failed it retains the handle; publish it
+            # here so a later stop can retry rather than losing a live socket.
+            if listener.running:
+                self._listener = listener
+            raise
         except Exception as exc:
             # A node that cannot listen must say so in the UI rather than look
             # enabled and quietly accept nothing.
+            if listener.running:
+                self._listener = listener
             self.startup_error = str(exc)
             logger.error("Could not start the inbound listener: %s", exc)
             return
@@ -282,12 +356,48 @@ class InboundNode:
         )
 
     async def stop(self) -> None:
-        sweep, self._idle_sweep = self._idle_sweep, None
-        if sweep is not None:
-            sweep.cancel()
-        listener, self._listener = self._listener, None
-        if listener is not None:
-            await listener.stop()
+        async with self._lifecycle_lock:
+            await self._stop()
+
+    async def _stop(self) -> None:
+        sweep = self._idle_sweep
+        listener = self._listener
+
+        async def shutdown() -> None:
+            if sweep is not None:
+                sweep.cancel()
+                await asyncio.gather(sweep, return_exceptions=True)
+            if listener is not None:
+                await listener.stop()
+
+        stopping = asyncio.create_task(shutdown(), name="inbound-node-stop")
+        try:
+            await asyncio.shield(stopping)
+        except asyncio.CancelledError:
+            await drain_task(stopping)
+            if stopping.cancelled():
+                raise
+            failure = stopping.exception()
+            if failure is not None:
+                raise failure
+            if self._idle_sweep is sweep:
+                self._idle_sweep = None
+            if self._listener is listener:
+                self._listener = None
+            raise
+        except BaseException:
+            await drain_task(stopping)
+            raise
+        if self._idle_sweep is sweep:
+            self._idle_sweep = None
+        if self._listener is listener:
+            self._listener = None
+
+    async def revoke_key(self, key_id: str) -> bool:
+        """Durably revoke one panel and withdraw all of its live sessions."""
+        if self._listener is not None:
+            return await self._listener.revoke_key_and_wait(key_id)
+        return self.keys.revoke(key_id)
 
     def connection_string(self, secret: str, *, host: Optional[str] = None) -> str:
         """The one artifact a user copies to another machine.
@@ -330,6 +440,8 @@ class OutboundNodes:
         self._connections: dict[str, object] = {}
         self._tasks: dict[str, asyncio.Task] = {}
         self._credentials = credentials
+        self._lifecycle_lock = asyncio.Lock()
+        self._servicer = None
 
     @property
     def credentials(self) -> KeyStore:
@@ -367,67 +479,337 @@ class OutboundNodes:
     async def add(self, text: str, servicer) -> Connection:
         """Parse, save and dial. Raises InvalidConnectionString on a bad paste."""
         connection = parse_connection(text)
-        entries = self.saved()
-        # Keyed by endpoint: re-pasting a rotated key for the same machine
-        # replaces it rather than leaving a dead entry that retries forever.
-        entries = [e for e in entries if _endpoint_of(e) != connection.endpoint]
-        entries.append(connection.endpoint)
-        self.credentials.remember_connection_secret(
-            connection.endpoint, connection.secret, connection.fingerprint
-        )
-        self._save(entries)
+        async with self._lifecycle_lock:
+            return await self._add(connection, servicer)
 
-        # Tear down any live session to this machine BEFORE dialling. Without
-        # this, re-pasting for an already-connected machine saved the new key
-        # and then short-circuited on the existing connection — so a wrong key
-        # reported success, kept working on the old session, and only failed
-        # after a restart, by which time nothing pointed at the paste that
-        # caused it. Verified on hardware.
-        await self._drop(connection.endpoint)
-        await self._dial(connection, servicer)
-        return connection
-
-    async def _drop(self, endpoint: str) -> None:
-        connection = self._connections.pop(endpoint, None)
-        task = self._tasks.pop(endpoint, None)
-        if connection is not None:
-            await connection.stop()
-        if task is not None:
-            task.cancel()
-
-    async def remove(self, endpoint: str) -> bool:
-        entries = [e for e in self.saved() if _endpoint_of(e) != endpoint]
-        self._save(entries)
-        self.credentials.forget_connection_secret(endpoint)
-        existed = endpoint in self._connections
-        await self._drop(endpoint)
-        return existed
-
-    async def start_all(self, servicer) -> None:
-        for entry in self.saved():
-            try:
-                await self._dial(self._connection_for(entry), servicer)
-            except InvalidConnectionString as exc:
-                logger.warning(
-                    "Ignoring a saved connection that no longer parses: %s", exc
-                )
-
-    async def _dial(self, connection: Connection, servicer) -> None:
+    async def _add(self, connection: Connection, servicer) -> Connection:
         from worker.inbound.connector import NodeConnection  # noqa: PLC0415
 
-        if connection.endpoint in self._connections:
+        original_entries = self.saved()
+        old_secret = self.credentials.connection_secret(connection.endpoint)
+        old_fingerprint = self.credentials.connection_fingerprint(connection.endpoint)
+        existing = self._connections.get(connection.endpoint)
+        existing_task = self._tasks.get(connection.endpoint)
+
+        # Pasting the already-running key is idempotent. Probing it would be a
+        # duplicate Attach and could disturb state deliberately retained by
+        # that same key.
+        if (
+            existing is not None
+            and old_secret == connection.secret
+            and old_fingerprint == connection.fingerprint
+        ):
+            if existing_task is not None and not existing_task.done():
+                return connection
+            # Terminal registration failures leave their diagnostic connector
+            # in the snapshot. Re-pasting after an upgrade/repair must really
+            # redial, not mistake that dead object for a healthy connection.
+            if self._connections.get(connection.endpoint) is existing:
+                self._connections.pop(connection.endpoint, None)
+            if self._tasks.get(connection.endpoint) is existing_task:
+                self._tasks.pop(connection.endpoint, None)
+            try:
+                await self._dial(connection, servicer, wait_until_ready=True)
+            except BaseException as operation:
+                try:
+                    await _finish_rollback(
+                        self._restore_failed_redial(
+                            connection.endpoint, existing, existing_task
+                        ),
+                        description="restore the previous inbound connector",
+                    )
+                except InboundConnectionRollbackError as rollback:
+                    raise rollback from operation
+                raise
+            return connection
+
+        if existing is not None:
+            # Authenticate and apply identity/version policy before touching
+            # the only working connector or its durable credential.
+            await NodeConnection(servicer, connection).probe()
+
+        entries = [
+            entry
+            for entry in original_entries
+            if _endpoint_of(entry) != connection.endpoint
+        ]
+        # Keyed by endpoint: re-pasting a rotated key for the same machine
+        # replaces it rather than leaving a dead entry that retries forever.
+        entries.append(connection.endpoint)
+        try:
+            if existing is not None:
+                # An offline connector can own work retained on the node. Its
+                # shutdown guard must run before replacement state is persisted.
+                await self._drop(connection.endpoint)
+            self.credentials.remember_connection_secret(
+                connection.endpoint, connection.secret, connection.fingerprint
+            )
+            self._save(entries)
+            await self._dial(
+                connection, servicer, wait_until_ready=existing is not None
+            )
+        except BaseException as operation:
+            try:
+                await _finish_rollback(
+                    self._rollback_add(
+                        connection,
+                        servicer,
+                        existing,
+                        original_entries,
+                        old_secret,
+                        old_fingerprint,
+                    ),
+                    description="restore the previous inbound connection",
+                )
+            except InboundConnectionRollbackError as rollback:
+                raise rollback from operation
+            raise
+        return connection
+
+    async def _restore_failed_redial(
+        self, endpoint: str, existing, existing_task: Optional[asyncio.Task]
+    ) -> None:
+        failure = None
+        candidate = self._connections.get(endpoint)
+        candidate_task = self._tasks.get(endpoint)
+        if candidate is not None and candidate is not existing:
+            try:
+                await self._close_candidate(endpoint, candidate, candidate_task)
+            except BaseException as exc:
+                failure = exc
+        if endpoint not in self._connections:
+            self._connections[endpoint] = existing
+        if endpoint not in self._tasks and existing_task is not None:
+            self._tasks[endpoint] = existing_task
+        if failure is not None:
+            raise InboundConnectionRollbackError(
+                "The previous GPU-machine connector could not be restored safely."
+            ) from failure
+
+    async def _close_candidate(self, endpoint: str, candidate, candidate_task) -> None:
+        try:
+            close = getattr(candidate, "close", None)
+            if callable(close):
+                await close()
+        finally:
+            if candidate_task is not None:
+                candidate_task.cancel()
+                await asyncio.gather(candidate_task, return_exceptions=True)
+            if self._connections.get(endpoint) is candidate:
+                self._connections.pop(endpoint, None)
+            if self._tasks.get(endpoint) is candidate_task:
+                self._tasks.pop(endpoint, None)
+
+    async def _rollback_add(
+        self,
+        connection: Connection,
+        servicer,
+        existing,
+        original_entries: list[str],
+        old_secret: str,
+        old_fingerprint: str,
+    ) -> None:
+        """Restore both live and durable generations after a failed replacement."""
+        endpoint = connection.endpoint
+        failures = []
+        candidate = self._connections.get(endpoint)
+        candidate_task = self._tasks.get(endpoint)
+        if candidate is not None and candidate is not existing:
+            try:
+                await self._close_candidate(endpoint, candidate, candidate_task)
+            except BaseException as exc:
+                failures.append(exc)
+        try:
+            if old_secret:
+                self.credentials.remember_connection_secret(
+                    endpoint, old_secret, old_fingerprint
+                )
+            else:
+                self.credentials.forget_connection_secret(endpoint)
+        except BaseException as exc:
+            failures.append(exc)
+        try:
+            self._save(original_entries)
+        except BaseException as exc:
+            failures.append(exc)
+        if (
+            not failures
+            and existing is not None
+            and old_secret
+            and endpoint not in self._connections
+        ):
+            try:
+                await self._dial(
+                    Connection(
+                        host=connection.host,
+                        port=connection.port,
+                        secret=old_secret,
+                        fingerprint=old_fingerprint,
+                    ),
+                    servicer,
+                )
+            except BaseException as exc:
+                failures.append(exc)
+        if failures:
+            raise InboundConnectionRollbackError(
+                "The previous GPU-machine connection could not be restored safely. "
+                "It remains stopped; fix its connection/settings storage, then "
+                "paste the original connection again."
+            ) from failures[0]
+
+    async def _drop(self, endpoint: str) -> None:
+        connection = self._connections.get(endpoint)
+        task = self._tasks.get(endpoint)
+        if connection is not None:
+            await connection.stop()
+            if self._connections.get(endpoint) is connection:
+                self._connections.pop(endpoint, None)
+        if self._tasks.get(endpoint) is task:
+            self._tasks.pop(endpoint, None)
+        if task is not None:
+            task.cancel()
+            await asyncio.gather(task, return_exceptions=True)
+
+    async def remove(self, endpoint: str) -> bool:
+        async with self._lifecycle_lock:
+            return await self._remove(endpoint)
+
+    async def _remove(self, endpoint: str) -> bool:
+        existed = endpoint in self._connections
+        original_entries = self.saved()
+        secret = self.credentials.connection_secret(endpoint)
+        fingerprint = self.credentials.connection_fingerprint(endpoint)
+        previous_connection = None
+        if secret and fingerprint:
+            try:
+                previous_connection = self._connection_for(endpoint)
+            except InvalidConnectionString:
+                pass
+        # A disconnected node deliberately retains work for reconnect. Do not
+        # erase the only connector/key capable of delivering terminal shutdown.
+        try:
+            await self._drop(endpoint)
+            entries = [e for e in original_entries if _endpoint_of(e) != endpoint]
+            # Remove the protected credential first. If that durable write
+            # fails, restore both durable generations and the live connector.
+            self.credentials.forget_connection_secret(endpoint)
+            self._save(entries)
+        except BaseException as operation:
+            try:
+                await _finish_rollback(
+                    self._rollback_remove(
+                        endpoint,
+                        original_entries,
+                        secret,
+                        fingerprint,
+                        previous_connection,
+                    ),
+                    description="restore removed inbound connection state",
+                )
+            except InboundConnectionRollbackError as rollback:
+                raise rollback from operation
+            raise
+        return existed
+
+    async def _rollback_remove(
+        self,
+        endpoint: str,
+        original_entries: list[str],
+        secret: str,
+        fingerprint: str,
+        previous_connection: Optional[Connection],
+    ) -> None:
+        failures = []
+        try:
+            if secret:
+                self.credentials.remember_connection_secret(
+                    endpoint, secret, fingerprint
+                )
+        except BaseException as exc:
+            failures.append(exc)
+        try:
+            self._save(original_entries)
+        except BaseException as exc:
+            failures.append(exc)
+        if (
+            not failures
+            and previous_connection is not None
+            and self._servicer is not None
+            and endpoint not in self._connections
+        ):
+            try:
+                await self._dial(previous_connection, self._servicer)
+            except BaseException as exc:
+                failures.append(exc)
+        if failures:
+            raise InboundConnectionRollbackError(
+                "The removed GPU-machine connection could not be restored safely. "
+                "It remains stopped; fix its connection/settings storage, then retry."
+            ) from failures[0]
+
+    async def start_all(self, servicer) -> None:
+        async with self._lifecycle_lock:
+            for entry in self.saved():
+                try:
+                    await self._dial(self._connection_for(entry), servicer)
+                except InvalidConnectionString as exc:
+                    logger.warning(
+                        "Ignoring a saved connection that no longer parses: %s", exc
+                    )
+
+    async def _dial(
+        self, connection: Connection, servicer, *, wait_until_ready: bool = False
+    ) -> None:
+        from worker.inbound.connector import NodeConnection  # noqa: PLC0415
+
+        self._servicer = servicer
+        existing = self._connections.get(connection.endpoint)
+        if existing is not None:
+            if wait_until_ready and getattr(existing, "_connection", None) != connection:
+                from worker.inbound.connector import (  # noqa: PLC0415
+                    InboundConnectionError,
+                )
+
+                raise InboundConnectionError(
+                    "A different connection to that GPU machine is already active."
+                )
             return
         node = NodeConnection(servicer, connection)
         self._connections[connection.endpoint] = node
-        self._tasks[connection.endpoint] = asyncio.create_task(
+        task = asyncio.create_task(
             node.run_forever(), name=f"inbound-node-{connection.endpoint}"
         )
+        task.add_done_callback(self._observe_connection_result)
+        self._tasks[connection.endpoint] = task
+        if wait_until_ready:
+            await node.wait_until_registered(task)
+
+    @staticmethod
+    def _observe_connection_result(task: asyncio.Task) -> None:
+        """Retrieve terminal dial errors; NodeConnection retains the UI detail."""
+        if task.cancelled():
+            return
+        try:
+            task.exception()
+        except asyncio.CancelledError:
+            pass
 
     async def stop(self) -> None:
+        async with self._lifecycle_lock:
+            await self._stop_all()
+
+    async def _stop_all(self) -> None:
         for connection in list(self._connections.values()):
-            await connection.stop()
-        for task in list(self._tasks.values()):
+            close = getattr(connection, "close", None)
+            if callable(close):
+                await close()
+            else:
+                await connection.stop()
+        tasks = list(self._tasks.values())
+        for task in tasks:
             task.cancel()
+        if tasks:
+            await asyncio.gather(*tasks, return_exceptions=True)
         self._connections.clear()
         self._tasks.clear()
 

backend/worker/pool.py

@@ -17,7 +17,7 @@ from dataclasses import dataclass, field
 from typing import Iterator, Optional
 
 from worker.breaker import BreakerRegistry
-from worker.capacity import WorkerCapacity, derive_concurrency
+from worker.capacity import WorkerCapacity, clamp_concurrency, derive_concurrency
 from worker.clock import resolve
 from worker.identity import Session
 from worker.registry import RemoteWorker
@@ -58,6 +58,9 @@ class ConnectedWorker:
     # The address this worker connected FROM, as the control plane saw it.
     address: str = ""
     draining: bool = False
+    # Registration handoff temporarily stops new assignments without
+    # conflating that transport state with a user-requested drain/shutdown.
+    registration_pending: bool = False
     # Attempt ids this worker claims to be running. Rebuilt on every reconnect
     # from its own report, never inferred.
     in_flight: set[str] = field(default_factory=set)
@@ -82,7 +85,11 @@ class ConnectedWorker:
         """
         if self.stale():
             return "offline"
-        if self.draining or self.capacity.available_slots <= 0:
+        if (
+            self.draining
+            or self.registration_pending
+            or self.capacity.available_slots <= 0
+        ):
             return "busy"
         return "ready"
 
@@ -175,7 +182,7 @@ class WorkerPool:
             epoch=epoch,
             capacity=WorkerCapacity(
                 worker_id=record.id,
-                max_concurrent_tasks=max(1, max_concurrent_tasks),
+                max_concurrent_tasks=clamp_concurrency(max_concurrent_tasks),
                 backend=backend,
             ),
             connected_at=stamp,
@@ -231,6 +238,10 @@ class WorkerPool:
     def disconnect(self, worker_id: str) -> Optional[ConnectedWorker]:
         return self._connected.pop(worker_id, None)
 
+    def restore_connection(self, worker: ConnectedWorker) -> None:
+        """Restore an exact live snapshot after replacement activation fails."""
+        self._connected[worker.worker_id] = worker
+
     def get(self, worker_id: str) -> Optional[ConnectedWorker]:
         return self._connected.get(worker_id)
 
@@ -300,10 +311,14 @@ class WorkerPool:
                 worker.capacity.slots[key] = ModelSlot(
                     engine=cap.get("engine", ""),
                     model_id=cap.get("model_id", ""),
-                    derived_concurrency=max(0, declared),
+                    derived_concurrency=clamp_concurrency(
+                        declared, allow_zero=True
+                    ),
                 )
             else:
-                slot.derived_concurrency = max(0, declared)
+                slot.derived_concurrency = clamp_concurrency(
+                    declared, allow_zero=True
+                )
 
     def stale_workers(self, *, now: Optional[float] = None) -> list[ConnectedWorker]:
         return [w for w in self if w.stale(now=now)]

backend/worker/protocol/worker_v1.proto

@@ -7,7 +7,9 @@
 // Rules of the road (goal_v2.md A5):
 //   * Additive-only within v1. Never renumber, never reuse a field number.
 //   * Version negotiation happens at Register; the server may refuse with
-//     UPGRADE_REQUIRED. Supported skew window is N-2.
+//     UPGRADE_REQUIRED. Semantic protocol v2 is intentionally incompatible
+//     with v1 because enrollment became a durable two-phase handshake; never
+//     infer a release-based skew window across that boundary.
 //   * The Control stream carries SMALL messages only. Artifacts (reference
 //     audio in, rendered audio/video out) move through UploadResult /
 //     DownloadArtifact. A large payload on the control stream head-of-line

backend/worker/registry.py

@@ -19,12 +19,15 @@ from __future__ import annotations
 
 import json
 import logging
+import threading
 import time
 import uuid
+from contextlib import contextmanager
 from dataclasses import dataclass, field
 from typing import Optional
 
 from core.db import db_conn
+from worker.capacity import clamp_concurrency
 from worker.clock import resolve
 from worker import identity
 
@@ -33,6 +36,14 @@ logger = logging.getLogger("omnivoice.worker")
 # Default scheduling preference. Higher wins; equal priorities fall through to
 # least-busy, which is the actual default behaviour for a homogeneous setup.
 _DEFAULT_PRIORITY = 50
+_AUTHORITY_LOCK = threading.RLock()
+
+
+@contextmanager
+def authority_guard():
+    """Serialize durable authority changes with live-session publication."""
+    with _AUTHORITY_LOCK:
+        yield
 
 
 @dataclass
@@ -180,6 +191,54 @@ def purge_expired_enrollments(*, now: Optional[float] = None) -> int:
 # ── Workers ────────────────────────────────────────────────────────────────
 
 
+def _new_worker(
+    *,
+    name: str,
+    public_key: bytes,
+    endpoint: str,
+    host: Optional[dict],
+    capabilities: Optional[list[dict]],
+    max_concurrent_tasks: int,
+    consent_granted: bool,
+    stamp: float,
+) -> RemoteWorker:
+    key_id = identity.key_id_for(public_key)
+    return RemoteWorker(
+        id=uuid.uuid4().hex[:12],
+        name=name or key_id,
+        key_id=key_id,
+        public_key=public_key,
+        endpoint=endpoint,
+        host=host or {},
+        capabilities=capabilities or [],
+        max_concurrent_tasks=clamp_concurrency(max_concurrent_tasks),
+        consent_granted_at=stamp if consent_granted else None,
+        created_at=stamp,
+    )
+
+
+def _insert_worker(conn, worker: RemoteWorker) -> None:
+    conn.execute(
+        "INSERT INTO remote_workers "
+        "(id, name, key_id, public_key, enabled, revoked, priority, endpoint, host_json, "
+        " capabilities_json, max_concurrent_tasks, session_epoch, consent_granted_at, created_at) "
+        "VALUES (?, ?, ?, ?, 1, 0, ?, ?, ?, ?, ?, 0, ?, ?)",
+        (
+            worker.id,
+            worker.name,
+            worker.key_id,
+            worker.public_key,
+            worker.priority,
+            worker.endpoint,
+            json.dumps(worker.host),
+            json.dumps(worker.capabilities),
+            worker.max_concurrent_tasks,
+            worker.consent_granted_at,
+            worker.created_at,
+        ),
+    )
+
+
 def enroll_worker(
     *,
     name: str,
@@ -198,49 +257,128 @@ def enroll_worker(
     your own desktop is not agreeing to use someone else's.
     """
     stamp = resolve(now)
-    key_id = identity.key_id_for(public_key)
-    existing = get_by_key_id(key_id)
-    if existing is not None:
-        return existing
-    worker = RemoteWorker(
-        id=uuid.uuid4().hex[:12],
-        name=name or key_id,
-        key_id=key_id,
+    worker = _new_worker(
+        name=name,
         public_key=public_key,
         endpoint=endpoint,
-        host=host or {},
-        capabilities=capabilities or [],
-        max_concurrent_tasks=max(1, int(max_concurrent_tasks)),
-        consent_granted_at=stamp if consent_granted else None,
-        created_at=stamp,
+        host=host,
+        capabilities=capabilities,
+        max_concurrent_tasks=max_concurrent_tasks,
+        consent_granted=consent_granted,
+        stamp=stamp,
     )
-    with db_conn() as conn:
-        conn.execute(
-            "INSERT INTO remote_workers "
-            "(id, name, key_id, public_key, enabled, revoked, priority, endpoint, host_json, "
-            " capabilities_json, max_concurrent_tasks, session_epoch, consent_granted_at, created_at) "
-            "VALUES (?, ?, ?, ?, 1, 0, ?, ?, ?, ?, ?, 0, ?, ?)",
-            (
-                worker.id,
-                worker.name,
-                worker.key_id,
-                worker.public_key,
-                worker.priority,
-                worker.endpoint,
-                json.dumps(worker.host),
-                json.dumps(worker.capabilities),
-                worker.max_concurrent_tasks,
-                worker.consent_granted_at,
-                worker.created_at,
-            ),
-        )
+    with _AUTHORITY_LOCK, db_conn() as conn:
+        row = conn.execute(
+            "SELECT * FROM remote_workers WHERE key_id = ?", (worker.key_id,)
+        ).fetchone()
+        if row is not None:
+            return _row_to_worker(row)
+        _insert_worker(conn, worker)
     logger.info("Enrolled remote worker %s (%s)", worker.name, worker.key_id)
     return worker
 
 
+def enroll_with_token(
+    token: identity.EnrollmentToken,
+    *,
+    name: str,
+    public_key: bytes,
+    endpoint: str = "",
+    host: Optional[dict] = None,
+    capabilities: Optional[list[dict]] = None,
+    max_concurrent_tasks: int = 1,
+    consent_granted: bool = True,
+    now: Optional[float] = None,
+) -> Optional[RemoteWorker]:
+    """Consume a token and bind its worker identity in one transaction."""
+    stamp = resolve(now)
+    worker = _new_worker(
+        name=name,
+        public_key=public_key,
+        endpoint=endpoint,
+        host=host,
+        capabilities=capabilities,
+        max_concurrent_tasks=max_concurrent_tasks,
+        consent_granted=consent_granted,
+        stamp=stamp,
+    )
+    with _AUTHORITY_LOCK, db_conn() as conn:
+        enrollment = conn.execute(
+            "SELECT secret_hash, expires_at, used_at FROM remote_worker_enrollments "
+            "WHERE token_id = ?",
+            (token.token_id,),
+        ).fetchone()
+        if (
+            enrollment is None
+            or enrollment["used_at"] is not None
+            or stamp > float(enrollment["expires_at"])
+            or not identity.constant_time_equals(
+                enrollment["secret_hash"], token.secret_hash
+            )
+        ):
+            return None
+        existing_row = conn.execute(
+            "SELECT * FROM remote_workers WHERE key_id = ?", (worker.key_id,)
+        ).fetchone()
+        if existing_row is not None and bool(existing_row["revoked"]):
+            return None
+        enrolled = _row_to_worker(existing_row) if existing_row is not None else worker
+        consumed = conn.execute(
+            "UPDATE remote_worker_enrollments SET used_at = ?, used_by_worker = ? "
+            "WHERE token_id = ? AND used_at IS NULL",
+            (stamp, enrolled.id, token.token_id),
+        )
+        if consumed.rowcount != 1:
+            return None
+        if existing_row is None:
+            _insert_worker(conn, worker)
+    if existing_row is None:
+        logger.info("Enrolled remote worker %s (%s)", worker.name, worker.key_id)
+    return enrolled
+
+
+def recover_enrollment_with_token(
+    token: identity.EnrollmentToken, *, public_key: bytes
+) -> Optional[RemoteWorker]:
+    """Resolve a spent token only to the exact key it originally enrolled.
+
+    This is only the durable lookup half of recovery. The transport must also
+    verify a fresh signature from this key before issuing another session; a
+    spent token and an observed public key are not proof of private-key
+    possession.
+    """
+    with _AUTHORITY_LOCK, db_conn() as conn:
+        enrollment = conn.execute(
+            "SELECT secret_hash, used_at, used_by_worker "
+            "FROM remote_worker_enrollments WHERE token_id = ?",
+            (token.token_id,),
+        ).fetchone()
+        if (
+            enrollment is None
+            or enrollment["used_at"] is None
+            or not enrollment["used_by_worker"]
+            or not identity.constant_time_equals(
+                enrollment["secret_hash"], token.secret_hash
+            )
+        ):
+            return None
+        row = conn.execute(
+            "SELECT * FROM remote_workers WHERE id = ?",
+            (enrollment["used_by_worker"],),
+        ).fetchone()
+    if row is None:
+        return None
+    worker = _row_to_worker(row)
+    if worker.revoked or worker.public_key != public_key:
+        return None
+    return worker
+
+
 def get(worker_id: str) -> Optional[RemoteWorker]:
-    with db_conn() as conn:
-        row = conn.execute("SELECT * FROM remote_workers WHERE id = ?", (worker_id,)).fetchone()
+    with _AUTHORITY_LOCK, db_conn() as conn:
+        row = conn.execute(
+            "SELECT * FROM remote_workers WHERE id = ?", (worker_id,)
+        ).fetchone()
     return _row_to_worker(row) if row else None
 
 
@@ -268,7 +406,7 @@ def begin_session(worker_id: str, *, now: Optional[float] = None) -> int:
     otherwise delivers two accepts for one assignment.
     """
     stamp = resolve(now)
-    with db_conn() as conn:
+    with _AUTHORITY_LOCK, db_conn() as conn:
         conn.execute(
             "UPDATE remote_workers SET session_epoch = session_epoch + 1, last_seen_at = ? WHERE id = ?",
             (stamp, worker_id),
@@ -292,29 +430,79 @@ def update_capabilities(
     capabilities: list[dict],
     host: Optional[dict] = None,
     max_concurrent_tasks: Optional[int] = None,
+    _conn=None,
 ) -> None:
-    sets = ["capabilities_json = ?"]
-    params: list = [json.dumps(capabilities)]
-    if host is not None:
-        sets.append("host_json = ?")
-        params.append(json.dumps(host))
-    if max_concurrent_tasks is not None:
-        sets.append("max_concurrent_tasks = ?")
-        params.append(max(1, int(max_concurrent_tasks)))
-    params.append(worker_id)
-    with db_conn() as conn:
-        conn.execute(f"UPDATE remote_workers SET {', '.join(sets)} WHERE id = ?", params)
+    params = (
+        json.dumps(capabilities),
+        json.dumps(host) if host is not None else None,
+        clamp_concurrency(max_concurrent_tasks)
+        if max_concurrent_tasks is not None
+        else None,
+        worker_id,
+    )
+    sql = """
+        UPDATE remote_workers
+        SET capabilities_json = ?,
+            host_json = COALESCE(?, host_json),
+            max_concurrent_tasks = COALESCE(?, max_concurrent_tasks)
+        WHERE id = ?
+    """
+    if _conn is not None:
+        # The caller owns the surrounding SQLite transaction. Acquiring the
+        # authority lock inside it can deadlock against revoke, which takes
+        # that lock before waiting for the same database write lock.
+        _conn.execute(sql, params)
+    else:
+        with _AUTHORITY_LOCK:
+            with db_conn() as conn:
+                conn.execute(sql, params)
+
+
+def update_policy(
+    worker_id: str,
+    *,
+    name: Optional[str] = None,
+    enabled: Optional[bool] = None,
+    priority: Optional[int] = None,
+) -> Optional[RemoteWorker]:
+    """Atomically update user-controlled policy and return the committed row."""
+    with _AUTHORITY_LOCK, db_conn() as conn:
+        row = conn.execute(
+            "SELECT * FROM remote_workers WHERE id = ?", (worker_id,)
+        ).fetchone()
+        if row is None:
+            return None
+        if name is not None or enabled is not None or priority is not None:
+            conn.execute(
+                """
+                UPDATE remote_workers
+                SET name = COALESCE(?, name),
+                    enabled = COALESCE(?, enabled),
+                    priority = COALESCE(?, priority)
+                WHERE id = ?
+                """,
+                (
+                    name,
+                    1 if enabled is True else 0 if enabled is False else None,
+                    max(0, min(100, int(priority))) if priority is not None else None,
+                    worker_id,
+                ),
+            )
+            row = conn.execute(
+                "SELECT * FROM remote_workers WHERE id = ?", (worker_id,)
+            ).fetchone()
+    return _row_to_worker(row) if row is not None else None
 
 
 def set_enabled(worker_id: str, enabled: bool) -> None:
-    with db_conn() as conn:
+    with _AUTHORITY_LOCK, db_conn() as conn:
         conn.execute(
             "UPDATE remote_workers SET enabled = ? WHERE id = ?", (1 if enabled else 0, worker_id)
         )
 
 
 def set_priority(worker_id: str, priority: int) -> None:
-    with db_conn() as conn:
+    with _AUTHORITY_LOCK, db_conn() as conn:
         conn.execute(
             "UPDATE remote_workers SET priority = ? WHERE id = ?",
             (max(0, min(100, int(priority))), worker_id),
@@ -322,7 +510,7 @@ def set_priority(worker_id: str, priority: int) -> None:
 
 
 def rename(worker_id: str, name: str) -> None:
-    with db_conn() as conn:
+    with _AUTHORITY_LOCK, db_conn() as conn:
         conn.execute("UPDATE remote_workers SET name = ? WHERE id = ?", (name, worker_id))
 
 
@@ -334,7 +522,7 @@ def revoke(worker_id: str, *, now: Optional[float] = None) -> bool:
     who could simply enroll again.
     """
     stamp = resolve(now)
-    with db_conn() as conn:
+    with _AUTHORITY_LOCK, db_conn() as conn:
         cur = conn.execute(
             "UPDATE remote_workers SET revoked = 1, revoked_at = ?, enabled = 0 WHERE id = ?",
             (stamp, worker_id),
@@ -345,15 +533,23 @@ def revoke(worker_id: str, *, now: Optional[float] = None) -> bool:
 
 
 def is_revoked(key_id: str) -> bool:
-    with db_conn() as conn:
+    with _AUTHORITY_LOCK, db_conn() as conn:
         row = conn.execute(
             "SELECT revoked FROM remote_workers WHERE key_id = ?", (key_id,)
         ).fetchone()
     return bool(row and row["revoked"])
 
 
+def is_enabled(worker_id: str) -> bool:
+    with _AUTHORITY_LOCK, db_conn() as conn:
+        row = conn.execute(
+            "SELECT enabled FROM remote_workers WHERE id = ?", (worker_id,)
+        ).fetchone()
+    return bool(row and row["enabled"])
+
+
 def grant_consent(worker_id: str, *, now: Optional[float] = None) -> None:
-    with db_conn() as conn:
+    with _AUTHORITY_LOCK, db_conn() as conn:
         conn.execute(
             "UPDATE remote_workers SET consent_granted_at = ? WHERE id = ?",
             (resolve(now), worker_id),
@@ -397,16 +593,20 @@ def authenticate(
 
 __all__ = [
     "RemoteWorker",
+    "authority_guard",
     "authenticate",
     "begin_session",
     "create_enrollment",
+    "enroll_with_token",
     "enroll_worker",
     "get",
     "get_by_key_id",
     "grant_consent",
     "is_revoked",
+    "is_enabled",
     "list_workers",
     "purge_expired_enrollments",
+    "recover_enrollment_with_token",
     "redeem_enrollment",
     "rename",
     "revoke",
@@ -414,4 +614,5 @@ __all__ = [
     "set_priority",
     "touch",
     "update_capabilities",
+    "update_policy",
 ]

backend/worker/scheduler.py

@@ -27,14 +27,20 @@ be minutes.
 from __future__ import annotations
 
 import asyncio
+import copy
 import enum
 import logging
 import uuid
-from dataclasses import dataclass
+from dataclasses import dataclass, fields
+from functools import partial
 from typing import Callable, Optional
 
 from worker import deadlines as deadline_policy
-from worker import task_store
+from worker import registry, task_store
+from worker.async_utils import (
+    to_thread_and_defer_cancellation,
+    to_thread_and_drain_on_cancel,
+)
 from worker.breaker import Attribution
 from worker.capacity import WorkerCapacity
 from worker.clock import resolve
@@ -100,6 +106,29 @@ class SchedulerStopped(RuntimeError):
     """
 
 
+def _adopt_task_state(target: Task, source: Task) -> None:
+    """Commit a reconciled copy while preserving public task/attempt identities."""
+    current_attempts = {attempt.attempt_id: attempt for attempt in target.attempts}
+    adopted_attempts = []
+    for source_attempt in source.attempts:
+        target_attempt = current_attempts.get(source_attempt.attempt_id)
+        if target_attempt is None:
+            target_attempt = copy.deepcopy(source_attempt)
+        else:
+            for item in fields(Attempt):
+                setattr(
+                    target_attempt,
+                    item.name,
+                    copy.deepcopy(getattr(source_attempt, item.name)),
+                )
+        adopted_attempts.append(target_attempt)
+
+    for item in fields(Task):
+        if item.name != "attempts":
+            setattr(target, item.name, copy.deepcopy(getattr(source, item.name)))
+    target.attempts[:] = adopted_attempts
+
+
 @dataclass(frozen=True)
 class Assignment:
     """One task bound to one worker, ready to send."""
@@ -110,6 +139,18 @@ class Assignment:
     deadlines: deadline_policy.Deadlines
 
 
+@dataclass(frozen=True)
+class _ReconciliationGeneration:
+    """Immutable task copies persisted before their live generation is adopted."""
+
+    worker_id: str
+    stamp: float
+    originals: tuple[Task, ...]
+    snapshots: tuple[Task, ...]
+    candidates: tuple[Task, ...]
+    zombies: tuple[str, ...] = ()
+
+
 class Scheduler:
     """Owns the queue, the selection pipeline, and the deadline sweeper."""
 
@@ -137,6 +178,10 @@ class Scheduler:
         # task_id → when its progress was last written through. Cleared with
         # the task, so it cannot outlive what it describes.
         self._progress_saved_at: dict[str, float] = {}
+        # Async producers serialize admission while durable input staging is
+        # off-loop. This keeps queue/idempotency decisions atomic without
+        # holding the event loop hostage to a multi-gigabyte copy and hash.
+        self._submission_lock = asyncio.Lock()
 
     # ── Persistence seam ──────────────────────────────────────────────────
 
@@ -281,12 +326,99 @@ class Scheduler:
         )
         if deadline_seconds:
             task.deadline_at = stamp + deadline_seconds
-        self._tasks[task.task_id] = task
         if self._persist:
-            task_store.create(task, now=stamp)
+            persisted = task_store.create(task, now=stamp)
+            if persisted.task_id != task.task_id:
+                self._tasks.setdefault(persisted.task_id, persisted)
+                return persisted
+            task = persisted
+        # Publish to the dispatcher only after durable admission succeeds. A
+        # failed DB/input-stage write must not send work the API rejected.
+        self._tasks[task.task_id] = task
         self._emit("queued", task)
         return task
 
+    async def submit_async(
+        self,
+        *,
+        operation: str,
+        engine: str,
+        model_id: str,
+        params: Optional[dict] = None,
+        priority: PriorityClass = PriorityClass.INTERACTIVE,
+        idempotency_key: Optional[str] = None,
+        max_attempts: int = 3,
+        deadline_seconds: Optional[float] = None,
+        pinned_worker_id: Optional[str] = None,
+        now: Optional[float] = None,
+    ) -> Task:
+        """Admit durably without hashing/copying task inputs on the app loop."""
+        async with self._submission_lock:
+            stamp = resolve(now)
+            if idempotency_key:
+                for existing in self._tasks.values():
+                    if existing.idempotency_key == idempotency_key:
+                        return existing
+                if self._persist:
+                    stored, cancelled = await to_thread_and_defer_cancellation(
+                        task_store.get_by_idempotency_key, idempotency_key
+                    )
+                    if stored is not None:
+                        self._tasks.setdefault(stored.task_id, stored)
+                        if cancelled:
+                            raise asyncio.CancelledError
+                        return stored
+                    if cancelled:
+                        raise asyncio.CancelledError
+
+            if self.queue_depth >= self.max_queue_depth:
+                raise QueueFull(
+                    f"The remote task queue is full ({self.max_queue_depth} waiting). "
+                    "Wait for current work to finish, or add another worker."
+                )
+
+            task = Task(
+                task_id=uuid.uuid4().hex[:16],
+                operation=operation,
+                engine=engine,
+                model_id=model_id,
+                params=params or {},
+                priority=priority,
+                idempotency_key=idempotency_key,
+                max_attempts=max_attempts,
+                created_at=stamp,
+                pinned_worker_id=pinned_worker_id,
+            )
+            if deadline_seconds:
+                task.deadline_at = stamp + deadline_seconds
+            if self._persist:
+                persisted, cancelled = await to_thread_and_defer_cancellation(
+                    partial(task_store.create, task, now=stamp)
+                )
+                if persisted.task_id != task.task_id:
+                    self._tasks.setdefault(persisted.task_id, persisted)
+                    if cancelled:
+                        raise asyncio.CancelledError
+                    return persisted
+                task = persisted
+                if cancelled:
+                    # The durable copy cannot be interrupted midway. Settle it
+                    # before publication so cancellation can never dispatch a
+                    # task whose caller did not finish submitting it.
+                    task.cancel(reason="submission was cancelled", now=resolve())
+                    await to_thread_and_drain_on_cancel(
+                        partial(task_store.save, task, now=resolve())
+                    )
+                    self._tasks[task.task_id] = task
+                    self._emit("cancelled", task)
+                    raise asyncio.CancelledError
+
+            # The dispatcher only learns the task after input bytes and its DB
+            # row are durable. A failed stage cannot escape as runnable work.
+            self._tasks[task.task_id] = task
+            self._emit("queued", task)
+            return task
+
     def adopt(self, task: Task) -> None:
         """Take ownership of a task loaded from disk after a restart."""
         self._tasks[task.task_id] = task
@@ -378,7 +510,11 @@ class Scheduler:
         for worker in self.pool:
             if task.pinned_worker_id and worker.worker_id != task.pinned_worker_id:
                 continue
-            if not worker.record.schedulable or worker.draining:
+            if (
+                not worker.record.schedulable
+                or worker.draining
+                or worker.registration_pending
+            ):
                 continue
             if worker.stale(now=stamp):
                 continue
@@ -469,6 +605,13 @@ class Scheduler:
         empty or every waiting task is blocked on capacity. A task with no
         capable worker at all is failed here rather than left to age out.
         """
+        # Selection and binding are one authority read. A revoke/disable route
+        # holds the same lock until its durable row and live pool record agree,
+        # so another thread cannot bind work in that publication window.
+        with registry.authority_guard():
+            return self._next_assignment(now=now)
+
+    def _next_assignment(self, *, now: Optional[float] = None) -> Optional[Assignment]:
         stamp = resolve(now)
         for task in self._queued_order():
             try:
@@ -700,9 +843,25 @@ class Scheduler:
         if epoch is not None and attempt.session_epoch != epoch:
             return False, task
 
-        committed, attempt = task.commit_result(
-            attempt_id, result_ref=result_ref, session_epoch=epoch, now=stamp
-        )
+        if self._persist:
+            # Prepare and persist the terminal generation before publishing it
+            # to the live graph.  A failed database write must leave the
+            # redelivery path looking exactly as it did before this frame:
+            # still in flight, still consuming capacity, and not yet credited
+            # as a breaker success.
+            candidate = copy.deepcopy(task)
+            committed, _ = candidate.commit_result(
+                attempt_id, result_ref=result_ref, session_epoch=epoch, now=stamp
+            )
+            if committed:
+                task_store.commit_result(candidate, result_json=result, now=stamp)
+            else:
+                task_store.save(candidate, now=stamp)
+            _adopt_task_state(task, candidate)
+        else:
+            committed, attempt = task.commit_result(
+                attempt_id, result_ref=result_ref, session_epoch=epoch, now=stamp
+            )
         self._release_slot(task, attempt, now=stamp)
         worker = self.pool.get(attempt.worker_id)
         if worker is not None and committed:
@@ -711,10 +870,6 @@ class Scheduler:
                 WorkerCapacity.slot_key(task.engine, task.model_id),
                 now=stamp,
             )
-        if committed and self._persist:
-            task_store.commit_result(task, result_json=result, now=stamp)
-        elif self._persist:
-            self._save(task, now=stamp)
         self._emit("completed" if committed else "duplicate", task)
         return committed, task
 
@@ -779,28 +934,85 @@ class Scheduler:
         This is where the duplicate-execution bug would live if a disconnect
         were treated as a failure: the worker may be seconds from delivering.
         """
+        generation = self.prepare_disconnected(worker_id, now=now)
+        try:
+            # A disconnect is one generation, just like reconnect. If any
+            # write fails, neither the database nor the live task graph may
+            # contain a half-marked set of attempts.
+            self.persist_reconciliation(generation)
+            return self.apply_disconnected(generation)
+        finally:
+            # A failed persistence write must not leave a dead stream's pool
+            # entry schedulable.
+            self.pool.disconnect(worker_id)
+
+    def prepare_disconnected(
+        self,
+        worker_id: str,
+        *,
+        now: Optional[float] = None,
+        include_task_ids: Optional[set[str]] = None,
+    ) -> _ReconciliationGeneration:
+        """Build, but do not publish, one disconnected task generation."""
         stamp = resolve(now)
-        affected: list[Task] = []
-        for task in self._tasks.values():
+        included = include_task_ids or set()
+        originals = tuple(
+            task
+            for task in self._tasks.values()
+            if task.task_id in included
+            or (
+                task.active_attempt is not None
+                and task.active_attempt.worker_id == worker_id
+            )
+        )
+        snapshots = tuple(copy.deepcopy(task) for task in originals)
+        candidates = tuple(copy.deepcopy(task) for task in snapshots)
+        for task in candidates:
             attempt = task.active_attempt
             if attempt is None or attempt.worker_id != worker_id:
                 continue
             grace = deadline_policy.default_grace_seconds(task.operation)
-            task.mark_disconnected(attempt.attempt_id, grace_seconds=grace, now=stamp)
-            affected.append(task)
-            self._save(task, now=stamp)
-            self._emit("worker_lost", task)
-        self.pool.disconnect(worker_id)
-        return affected
+            task.mark_disconnected(
+                attempt.attempt_id, grace_seconds=grace, now=stamp
+            )
+        return _ReconciliationGeneration(
+            worker_id=worker_id,
+            stamp=stamp,
+            originals=originals,
+            snapshots=snapshots,
+            candidates=candidates,
+        )
 
     def on_reconnected(
-        self, worker_id: str, *, in_flight: set[str], now: Optional[float] = None
+        self,
+        worker_id: str,
+        *,
+        in_flight: set[str],
+        now: Optional[float] = None,
+        before_persist: Optional[Callable[[object], None]] = None,
     ) -> list[str]:
         """Reconcile against what the worker says it is running.
 
         Returns attempt ids the worker should cancel — work we have already
         written off, which it must stop burning a GPU on.
         """
+        generation = self.prepare_reconnected(
+            worker_id, in_flight=in_flight, now=now
+        )
+        self.persist_reconciliation(
+            generation, before_persist=before_persist
+        )
+        return self.apply_reconnected(generation)
+
+    def prepare_reconnected(
+        self,
+        worker_id: str,
+        *,
+        in_flight: set[str],
+        now: Optional[float] = None,
+        include_task_ids: Optional[set[str]] = None,
+    ) -> _ReconciliationGeneration:
+        """Build, but do not publish, one reconnect reconciliation."""
         stamp = resolve(now)
         known = {
             attempt.attempt_id: attempt
@@ -813,7 +1025,20 @@ class Scheduler:
             for attempt_id in in_flight
             if attempt_id not in known or known[attempt_id].state.terminal
         ]
-        for task in list(self._tasks.values()):
+        included = include_task_ids or set()
+        originals = tuple(
+            task
+            for task in self._tasks.values()
+            if task.task_id in included
+            or (
+                not task.state.terminal
+                and task.active_attempt is not None
+                and task.active_attempt.worker_id == worker_id
+            )
+        )
+        snapshots = tuple(copy.deepcopy(task) for task in originals)
+        candidates = tuple(copy.deepcopy(task) for task in snapshots)
+        for task in candidates:
             if task.state.terminal:
                 continue
             reconcile(
@@ -826,8 +1051,84 @@ class Scheduler:
                 resume_lease_seconds=self._budget_for(task).progress_lease_seconds,
                 now=stamp,
             )
-            self._save(task, now=stamp)
-        return zombies
+        return _ReconciliationGeneration(
+            worker_id=worker_id,
+            stamp=stamp,
+            originals=originals,
+            snapshots=snapshots,
+            candidates=candidates,
+            zombies=tuple(zombies),
+        )
+
+    def persist_reconciliation(
+        self,
+        generation: _ReconciliationGeneration,
+        *,
+        before_persist: Optional[Callable[[object], None]] = None,
+    ) -> None:
+        """Persist a prepared generation without touching the live task graph."""
+        # Nothing in the live graph changes until the complete generation is
+        # durable. A failed write therefore cannot queue duplicate execution
+        # while the old worker is still finishing the original attempt.
+        if self._persist or before_persist is not None:
+            task_store.save_many(
+                generation.candidates if self._persist else [],
+                now=generation.stamp,
+                before_save=before_persist,
+            )
+
+    def reconciliation_is_current(
+        self, generation: _ReconciliationGeneration
+    ) -> bool:
+        """Did the live graph remain unchanged while persistence was off-loop?"""
+        return all(
+            self._tasks.get(original.task_id) is original
+            and original == snapshot
+            for original, snapshot in zip(
+                generation.originals, generation.snapshots
+            )
+        )
+
+    def apply_reconnected(
+        self, generation: _ReconciliationGeneration
+    ) -> list[str]:
+        """Publish a durable reconnect generation on the scheduler's loop."""
+        if not self.reconciliation_is_current(generation):
+            raise RuntimeError("reconnect reconciliation generation changed")
+        changed: list[Task] = []
+        for original, candidate in zip(
+            generation.originals, generation.candidates
+        ):
+            if original != candidate:
+                changed.append(original)
+            _adopt_task_state(original, candidate)
+        for task in changed:
+            self._emit(
+                "failed"
+                if task.state.terminal
+                else "requeued"
+                if task.state is TaskState.QUEUED
+                else "resumed",
+                task,
+            )
+        return list(generation.zombies)
+
+    def apply_disconnected(
+        self, generation: _ReconciliationGeneration
+    ) -> list[Task]:
+        """Publish a durable disconnect generation on the scheduler's loop."""
+        if not self.reconciliation_is_current(generation):
+            raise RuntimeError("disconnect reconciliation generation changed")
+        changed: list[Task] = []
+        for original, candidate in zip(
+            generation.originals, generation.candidates
+        ):
+            if original != candidate:
+                changed.append(original)
+            _adopt_task_state(original, candidate)
+        for task in changed:
+            self._emit("worker_lost", task)
+        return list(generation.originals)
 
     def cancel(self, task_id: str, *, reason: str = "cancelled", now: Optional[float] = None) -> bool:
         task = self._tasks.get(task_id)

backend/worker/service.py

@@ -15,6 +15,7 @@ import asyncio
 import logging
 import os
 from typing import Optional
+from urllib.parse import urlsplit
 
 from worker.clock import resolve
 
@@ -24,10 +25,52 @@ logger = logging.getLogger("omnivoice.worker")
 _SWEEP_INTERVAL_SECONDS = 5.0
 # How often the dispatcher looks for queued work it can place.
 _DISPATCH_INTERVAL_SECONDS = 1.0
+# Finished rows and their artifacts remain inspectable for a week, then leave
+# in bounded batches. The loop keeps a long-lived control plane from growing
+# forever while the startup pass covers apps that are rarely left open.
+_ARTIFACT_GC_INTERVAL_SECONDS = 60 * 60.0
+_ARTIFACT_GC_BATCH_SIZE = 250
 
 DEFAULT_PORT = 7443
 
 
+class EndpointCertificateError(ValueError):
+    """An advertised endpoint is not usable with the live TLS certificate."""
+
+
+def _endpoint_host(endpoint: str) -> str:
+    """Extract the hostname from the gRPC ``host:port`` enrollment target."""
+    target = (endpoint or "").strip()
+    try:
+        parsed = urlsplit(f"//{target}")
+        host = parsed.hostname
+        port = parsed.port
+    except ValueError as exc:
+        raise EndpointCertificateError(
+            f"Worker endpoint must be host:port; got {endpoint!r}."
+        ) from exc
+    if (
+        not host
+        or port is None
+        or parsed.username is not None
+        or parsed.password is not None
+        or parsed.path
+        or parsed.query
+        or parsed.fragment
+    ):
+        raise EndpointCertificateError(
+            f"Worker endpoint must be host:port; got {endpoint!r}."
+        )
+    return host
+
+
+def _format_endpoint(host: str, port: int) -> str:
+    clean_host = (host or "").strip().strip("[]")
+    if ":" in clean_host:
+        clean_host = f"[{clean_host}]"
+    return f"{clean_host}:{port}"
+
+
 def remote_workers_enabled() -> bool:
     """Opt-in gate. Off unless the user turned it on.
 
@@ -93,6 +136,7 @@ class ControlPlane:
         self._server = None
         self._tasks: list[asyncio.Task] = []
         self._started = False
+        self._lifecycle_lock = asyncio.Lock()
         self.startup_error: Optional[str] = None
         # The port we actually bound, which is not necessarily the configured
         # one — an enrollment token carries this, so advertising the config
@@ -108,6 +152,10 @@ class ControlPlane:
         return self.credentials.fingerprint if self.credentials else ""
 
     async def start(self, *, port: Optional[int] = None) -> None:
+        async with self._lifecycle_lock:
+            await self._start(port=port)
+
+    async def _start(self, *, port: Optional[int] = None) -> None:
         if self._started:
             return
         # Imported here, not at module scope: a user who never enables remote
@@ -117,25 +165,39 @@ class ControlPlane:
         from worker.scheduler import Scheduler  # noqa: PLC0415
         from worker.transport.server import WorkerServicer, serve  # noqa: PLC0415
 
-        locations = paths()
-        os.makedirs(locations["root"], exist_ok=True)
-        self.credentials = tls.load_or_create(
-            locations["certificate"], locations["private_key"]
-        )
-        self.pool = WorkerPool()
-        self.scheduler = Scheduler(self.pool)
-        # Recover anything that was in flight when the app last quit. The
-        # workers holding those tasks may still be rendering.
-        self.scheduler.restore()
-
-        self.servicer = WorkerServicer(
-            self.scheduler,
-            self.pool,
-            artifact_dir=locations["artifacts"],
-            cert_fingerprint=self.credentials.fingerprint,
-        )
         self._port = port or control_port()
         try:
+            locations = paths()
+            os.makedirs(locations["root"], exist_ok=True)
+            wanted_hostnames = tls.default_hostnames()
+            advertised_host = _endpoint_host(self.default_endpoint())
+            if advertised_host not in wanted_hostnames:
+                wanted_hostnames.append(advertised_host)
+            self.credentials = tls.load_or_create(
+                locations["certificate"],
+                locations["private_key"],
+                hostnames=wanted_hostnames,
+            )
+            self.pool = WorkerPool()
+            self.scheduler = Scheduler(self.pool)
+            try:
+                await self._artifact_gc_once(
+                    locations["artifacts"], retry_upload_parts=False
+                )
+            except Exception:
+                # Retention is best-effort. A locked file or temporarily busy
+                # database must not make remote compute unavailable.
+                logger.exception("Remote worker startup artifact sweep failed")
+            # Recover anything that was in flight when the app last quit. The
+            # workers holding those tasks may still be rendering.
+            self.scheduler.restore()
+
+            self.servicer = WorkerServicer(
+                self.scheduler,
+                self.pool,
+                artifact_dir=locations["artifacts"],
+                cert_fingerprint=self.credentials.fingerprint,
+            )
             self._server = await serve(
                 self.servicer,
                 port=self._port,
@@ -148,12 +210,20 @@ class ControlPlane:
         self._tasks = [
             asyncio.create_task(self._sweep_loop(), name="worker-sweep"),
             asyncio.create_task(self._dispatch_loop(), name="worker-dispatch"),
+            asyncio.create_task(
+                self._artifact_gc_loop(locations["artifacts"]),
+                name="worker-artifact-gc",
+            ),
         ]
         self._started = True
         self.startup_error = None
         logger.info("Remote worker control plane started on port %d", self._port)
 
     async def stop(self) -> None:
+        async with self._lifecycle_lock:
+            await self._stop()
+
+    async def _stop(self) -> None:
         for task in self._tasks:
             task.cancel()
         if self._tasks:
@@ -199,6 +269,29 @@ class ControlPlane:
             except Exception:
                 logger.exception("Worker sweep failed")
 
+    async def _artifact_gc_once(
+        self, artifact_root: str, *, retry_upload_parts: bool = True
+    ) -> int:
+        """Purge one bounded retention batch without blocking the event loop."""
+        from worker import task_store  # noqa: PLC0415
+
+        removed = await asyncio.to_thread(
+            task_store.purge_finished,
+            root=artifact_root,
+            limit=_ARTIFACT_GC_BATCH_SIZE,
+        )
+        if retry_upload_parts and self.servicer is not None:
+            self.servicer.sweep_orphaned_upload_parts()
+        return removed
+
+    async def _artifact_gc_loop(self, artifact_root: str) -> None:
+        while True:
+            await asyncio.sleep(_ARTIFACT_GC_INTERVAL_SECONDS)
+            try:
+                await self._artifact_gc_once(artifact_root)
+            except Exception:
+                logger.exception("Remote worker artifact sweep failed")
+
     async def _dispatch_loop(self) -> None:
         """Place queued work on eligible workers.
 
@@ -233,10 +326,21 @@ class ControlPlane:
 
     def create_enrollment(self, *, endpoint: str = "", label: str = "", ttl_seconds: int = 900):
         """Mint a join token carrying this control plane's fingerprint."""
+        from worker import tls  # noqa: PLC0415
         from worker import registry  # noqa: PLC0415
 
+        advertised_endpoint = endpoint.strip() or self.default_endpoint()
+        advertised_host = _endpoint_host(advertised_endpoint)
+        if self.credentials is not None and not tls.covers(
+            self.credentials, advertised_host
+        ):
+            raise EndpointCertificateError(
+                f"The running certificate does not cover {advertised_host!r}. "
+                "Set OMNIVOICE_WORKER_ENDPOINT_HOST to that hostname and restart "
+                "VoiceStudio before creating this enrollment."
+            )
         return registry.create_enrollment(
-            endpoint=endpoint or self.default_endpoint(),
+            endpoint=advertised_endpoint,
             cert_fingerprint=self.fingerprint,
             label=label,
             ttl_seconds=ttl_seconds,
@@ -262,7 +366,7 @@ class ControlPlane:
             or tls.primary_ip()
             or "127.0.0.1"
         )
-        return f"{host}:{self._port or control_port()}"
+        return _format_endpoint(host, self._port or control_port())
 
     def snapshot(self, *, now: Optional[float] = None) -> dict:
         """Everything the workers UI needs in one call."""
@@ -380,6 +484,7 @@ async def stop() -> None:
 __all__ = [
     "ControlPlane",
     "DEFAULT_PORT",
+    "EndpointCertificateError",
     "control_plane",
     "control_port",
     "paths",

backend/worker/task_store.py

@@ -17,6 +17,7 @@ flips the task to completed, so the ack can only follow a durable fact.
 """
 from __future__ import annotations
 
+import errno
 import hashlib
 import json
 import logging
@@ -25,7 +26,8 @@ import os
 import re
 import shutil
 import time
-from typing import Iterable, Iterator, Optional
+import uuid
+from typing import Callable, Iterable, Iterator, Optional
 
 from core.db import db_conn
 from core.path_security import UnsafePath, resolve_within, safe_filename
@@ -133,6 +135,7 @@ INPUTS_PARAM_KEY = "inputs"
 
 _HASH_CHUNK_BYTES = 1024 * 1024
 _SAFE_EXTENSION = re.compile(r"^\.[A-Za-z0-9]{1,8}$")
+_CONTENT_ARTIFACT = re.compile(r"^([0-9a-f]{64})(?:\.[A-Za-z0-9]{1,8})?$")
 
 
 class InputStagingError(RuntimeError):
@@ -143,6 +146,60 @@ class InputStagingError(RuntimeError):
     """
 
 
+def _fsync_parent_directory(directory: str) -> None:
+    """Persist directory entry changes where the platform supports it."""
+    directory_flag = getattr(os, "O_DIRECTORY", None)
+    if directory_flag is None:
+        return
+    unsupported = {
+        errno.EINVAL,
+        getattr(errno, "ENOTSUP", errno.EINVAL),
+        getattr(errno, "EOPNOTSUPP", errno.EINVAL),
+    }
+    try:
+        descriptor = os.open(directory, os.O_RDONLY | directory_flag)
+    except OSError as exc:
+        if exc.errno in unsupported:
+            return
+        raise
+    try:
+        os.fsync(descriptor)
+    except OSError as exc:
+        if exc.errno not in unsupported:
+            raise
+    finally:
+        os.close(descriptor)
+
+
+def _fsync_file(path: str) -> None:
+    with open(path, "r+b") as handle:
+        os.fsync(handle.fileno())
+
+
+def _durable_makedirs(directory: str) -> None:
+    """Create a directory hierarchy and persist each parent entry."""
+    target = os.path.abspath(directory)
+    missing: list[str] = []
+    current = target
+    while not os.path.isdir(current):
+        if os.path.exists(current):
+            raise NotADirectoryError(current)
+        missing.append(current)
+        parent = os.path.dirname(current)
+        if parent == current:
+            break
+        current = parent
+    for path in reversed(missing):
+        try:
+            os.mkdir(path)
+        except FileExistsError:
+            if not os.path.isdir(path):
+                raise
+        _fsync_parent_directory(os.path.dirname(path) or ".")
+    if not missing:
+        _fsync_parent_directory(os.path.dirname(target) or ".")
+
+
 def artifact_root(*, create_dir: bool = True) -> str:
     """The directory the control plane serves artifacts from.
 
@@ -154,7 +211,7 @@ def artifact_root(*, create_dir: bool = True) -> str:
 
     root = paths()["artifacts"]
     if create_dir:
-        os.makedirs(os.path.join(root, INPUTS_DIRNAME), exist_ok=True)
+        _durable_makedirs(os.path.join(root, INPUTS_DIRNAME))
     return root
 
 
@@ -183,7 +240,40 @@ def _digest(path: str) -> tuple[str, int]:
     return digest.hexdigest(), size
 
 
-def stage_input(source: str, *, root: Optional[str] = None, now: Optional[float] = None) -> dict:
+def _staged_entry_matches(path: str, entry: dict) -> bool:
+    """Verify staged bytes against both metadata and their content address."""
+    if not os.path.isfile(path):
+        return False
+    artifact_id = str(entry.get("artifact_id") or "")
+    portable_name = artifact_id.replace("\\", "/").rsplit("/", 1)[-1]
+    named = _CONTENT_ARTIFACT.fullmatch(portable_name)
+    if named is None:
+        return False
+    try:
+        actual_digest, actual_size = _digest(path)
+    except OSError:
+        return False
+    recorded_digest = str(entry.get("sha256") or "").strip().lower()
+    recorded_size = entry.get("size_bytes")
+    if recorded_digest and actual_digest != recorded_digest:
+        return False
+    if recorded_size is not None:
+        try:
+            if actual_size != int(recorded_size):
+                return False
+        except (TypeError, ValueError):
+            return False
+    if actual_digest != named.group(1):
+        return False
+    # Backfill metadata on a legacy row once its content address proves it.
+    entry["sha256"] = actual_digest
+    entry["size_bytes"] = actual_size
+    return True
+
+
+def stage_input(
+    source: str, *, root: Optional[str] = None, now: Optional[float] = None
+) -> dict:
     """Copy one input into the artifact store, keyed by its content hash.
 
     Returns the record that ends up on the task row. ``source`` is kept in it
@@ -195,27 +285,58 @@ def stage_input(source: str, *, root: Optional[str] = None, now: Optional[float]
     try:
         digest, size = _digest(source)
     except OSError as exc:
-        raise InputStagingError(f"Could not read the task input {source!r}: {exc}") from exc
+        raise InputStagingError(
+            f"Could not read the task input {source!r}: {exc}"
+        ) from exc
 
     artifact_id = os.path.join(INPUTS_DIRNAME, f"{digest}{_extension(source)}")
     try:
         destination = resolve_within(base, artifact_id)
     except UnsafePath as exc:  # pragma: no cover — the id is ours, hex only
-        raise InputStagingError(f"Refusing to stage {source!r} outside the artifact store") from exc
+        raise InputStagingError(
+            f"Refusing to stage {source!r} outside the artifact store"
+        ) from exc
 
+    partial = destination.with_name(
+        f".{destination.name}.{uuid.uuid4().hex}.part"
+    )
     try:
-        destination.parent.mkdir(parents=True, exist_ok=True)
-        # Same size at a content-addressed name means the same bytes: the only
-        # writer is the rename below, so a truncated file cannot exist here.
-        if not (destination.is_file() and destination.stat().st_size == size):
-            partial = destination.with_name(destination.name + ".part")
+        _durable_makedirs(str(destination.parent))
+        expected = {
+            "artifact_id": artifact_id,
+            "sha256": digest,
+            "size_bytes": size,
+        }
+        if not _staged_entry_matches(str(destination), expected):
             shutil.copyfile(source, partial)
+            copied_digest, copied_size = _digest(str(partial))
+            if copied_digest != digest or copied_size != size:
+                raise InputStagingError(
+                    f"The task input {source!r} changed while it was being staged."
+                )
+            _fsync_file(str(partial))
             os.replace(partial, destination)
+            _fsync_parent_directory(str(destination.parent))
         # Freshness, not decoration: the purge dates an unreferenced input by
         # its mtime, so re-using a staged voice has to renew it.
         os.utime(destination, (stamp, stamp))
+        _fsync_file(str(destination))
+        _fsync_parent_directory(str(destination.parent))
     except OSError as exc:
-        raise InputStagingError(f"Could not stage the task input {source!r}: {exc}") from exc
+        raise InputStagingError(
+            f"Could not stage the task input {source!r}: {exc}"
+        ) from exc
+    finally:
+        try:
+            os.remove(partial)
+        except FileNotFoundError:
+            pass
+        except OSError:
+            logger.debug(
+                "Could not remove the staged-input partial %s",
+                partial,
+                exc_info=True,
+            )
 
     filename = os.path.basename(str(source)) or f"{digest}{_extension(source)}"
     return {
@@ -254,8 +375,13 @@ def ensure_staged(
     """
     params = task.params if isinstance(task.params, dict) else {}
     recorded = params.get(INPUTS_PARAM_KEY)
-    entries: list[dict] = [e for e in recorded if isinstance(e, dict)] if isinstance(recorded, list) else []
-    if root:
+    entries: list[dict] = (
+        [e for e in recorded if isinstance(e, dict)]
+        if isinstance(recorded, list)
+        else []
+    )
+    base = root or artifact_root()
+    if entries:
         # A task may have been staged when it was submitted under the default
         # store, then dispatched by a servicer configured with another store.
         # Recorded metadata is not proof that this servicer can serve it.
@@ -263,7 +389,10 @@ def ensure_staged(
         for entry in entries:
             artifact_id = str(entry.get("artifact_id") or "")
             try:
-                available = bool(artifact_id and resolve_within(root, artifact_id).is_file())
+                path = resolve_within(base, artifact_id)
+                available = bool(
+                    artifact_id and _staged_entry_matches(str(path), entry)
+                )
             except UnsafePath:
                 available = False
             if available:
@@ -271,7 +400,7 @@ def ensure_staged(
                 continue
             source = str(entry.get("source") or "")
             if source and os.path.isfile(source):
-                replacement = stage_input(source, root=root, now=now)
+                replacement = stage_input(source, root=base, now=now)
                 replacement.update(key=entry.get("key"), index=entry.get("index"))
                 refreshed.append(replacement)
             else:
@@ -289,7 +418,7 @@ def ensure_staged(
         # id here. Only what exists on this disk is an input.
         if not os.path.isfile(value):
             continue
-        entry = stage_input(value, root=root, now=now)
+        entry = stage_input(value, root=base, now=now)
         entry["key"] = key
         entry["index"] = index
         entries.append(entry)
@@ -345,6 +474,63 @@ def _referenced_artifacts(conn) -> set[str]:
     return referenced
 
 
+def _purge_result_directories(
+    task_ids: Iterable[str], *, root: Optional[str] = None
+) -> tuple[list[str], int]:
+    """Delete result directories before their owning rows become unreachable."""
+    task_ids = list(task_ids)
+    try:
+        base = root or artifact_root(create_dir=False)
+    except Exception:  # pragma: no cover — no data dir at all
+        logger.debug("No artifact root to purge", exc_info=True)
+        return [], 0
+    if not os.path.isdir(base):
+        return task_ids, 0
+
+    cleaned: list[str] = []
+    removed = 0
+    for task_id in task_ids:
+        try:
+            path = resolve_within(base, safe_filename(task_id))
+        except UnsafePath:
+            continue
+        if not os.path.exists(path):
+            try:
+                _fsync_parent_directory(base)
+            except OSError:
+                logger.debug(
+                    "Could not persist task artifact cleanup at %s",
+                    base,
+                    exc_info=True,
+                )
+                continue
+            cleaned.append(task_id)
+            continue
+        if not os.path.isdir(path):
+            logger.warning("Refusing to purge non-directory task artifact %s", path)
+            continue
+        try:
+            shutil.rmtree(path)
+        except OSError:
+            logger.debug("Could not purge task artifacts at %s", path, exc_info=True)
+            continue
+        try:
+            _fsync_parent_directory(base)
+        except OSError:
+            # The bytes are gone from this process's view, but the directory
+            # deletion is not a crash-durable fact yet. Keep the DB row as the
+            # retry index until a later sweep can establish that barrier.
+            logger.debug(
+                "Could not persist task artifact cleanup at %s",
+                base,
+                exc_info=True,
+            )
+            continue
+        cleaned.append(task_id)
+        removed += 1
+    return cleaned, removed
+
+
 def purge_artifacts(
     task_ids: Iterable[str], referenced: set[str], *, cutoff: float, root: Optional[str] = None
 ) -> int:
@@ -356,7 +542,7 @@ def purge_artifacts(
     rows were judged by. Nothing here raises — a purge that fails is a disk
     that stays fuller than we wanted, not a failed request.
     """
-    removed = 0
+    _cleaned, removed = _purge_result_directories(task_ids, root=root)
     try:
         base = root or artifact_root(create_dir=False)
     except Exception:  # pragma: no cover — no data dir at all
@@ -365,15 +551,6 @@ def purge_artifacts(
     if not os.path.isdir(base):
         return 0
 
-    for task_id in task_ids:
-        try:
-            path = resolve_within(base, safe_filename(task_id))
-        except UnsafePath:
-            continue
-        if os.path.isdir(path):
-            shutil.rmtree(path, ignore_errors=True)
-            removed += 1
-
     inputs_dir = os.path.join(base, INPUTS_DIRNAME)
     try:
         names = os.listdir(inputs_dir)
@@ -484,6 +661,25 @@ def _upsert_attempts(conn, task: Task) -> None:
         )
 
 
+def _save_with_conn(conn, task: Task, *, stamp: float) -> None:
+    conn.execute(
+        "UPDATE remote_tasks SET state=?, excluded_json=?, error_json=?, result_ref=?, "
+        "updated_at=?, deadline_at=?, finished_at=?, pinned_worker_id=? WHERE id=?",
+        (
+            task.state.value,
+            json.dumps(sorted(task.excluded_workers)),
+            _dump_error(task.error),
+            task.result_ref,
+            stamp,
+            task.deadline_at,
+            task.finished_at,
+            task.pinned_worker_id,
+            task.task_id,
+        ),
+    )
+    _upsert_attempts(conn, task)
+
+
 def save(task: Task, *, now: Optional[float] = None) -> None:
     """Write the whole task + attempt graph.
 
@@ -492,22 +688,22 @@ def save(task: Task, *, now: Optional[float] = None) -> None:
     """
     stamp = resolve(now)
     with db_conn() as conn:
-        conn.execute(
-            "UPDATE remote_tasks SET state=?, excluded_json=?, error_json=?, result_ref=?, "
-            "updated_at=?, deadline_at=?, finished_at=?, pinned_worker_id=? WHERE id=?",
-            (
-                task.state.value,
-                json.dumps(sorted(task.excluded_workers)),
-                _dump_error(task.error),
-                task.result_ref,
-                stamp,
-                task.deadline_at,
-                task.finished_at,
-                task.pinned_worker_id,
-                task.task_id,
-            ),
-        )
-        _upsert_attempts(conn, task)
+        _save_with_conn(conn, task, stamp=stamp)
+
+
+def save_many(
+    tasks: Iterable[Task],
+    *,
+    now: Optional[float] = None,
+    before_save: Optional[Callable[[object], None]] = None,
+) -> None:
+    """Persist one reconciliation generation atomically."""
+    stamp = resolve(now)
+    with db_conn() as conn:
+        if before_save is not None:
+            before_save(conn)
+        for task in tasks:
+            _save_with_conn(conn, task, stamp=stamp)
 
 
 def commit_result(
@@ -584,23 +780,30 @@ def load_unfinished() -> list[Task]:
     still be rendering, and reconciliation decides each one's fate once the
     workers reconnect.
     """
-    live = ", ".join(f"'{s.value}'" for s in TaskState if not s.terminal)
+    live = json.dumps([s.value for s in TaskState if not s.terminal])
     with db_conn() as conn:
         rows = conn.execute(
-            f"SELECT * FROM remote_tasks WHERE state IN ({live}) ORDER BY priority ASC, created_at ASC"
+            """
+            SELECT * FROM remote_tasks
+            WHERE state IN (SELECT value FROM json_each(?))
+            ORDER BY priority ASC, created_at ASC
+            """,
+            (live,),
         ).fetchall()
         return [_row_to_task(r, _attempts_for(conn, r["id"])) for r in rows]
 
 
 def list_tasks(*, states: Optional[Iterable[TaskState]] = None, limit: int = 100) -> list[Task]:
-    sql = "SELECT * FROM remote_tasks"
-    params: list = []
     if states:
-        placeholders = ", ".join("?" for _ in states)
-        sql += f" WHERE state IN ({placeholders})"
-        params.extend(s.value for s in states)
-    sql += " ORDER BY created_at DESC LIMIT ?"
-    params.append(limit)
+        sql = """
+            SELECT * FROM remote_tasks
+            WHERE state IN (SELECT value FROM json_each(?))
+            ORDER BY created_at DESC LIMIT ?
+        """
+        params = (json.dumps([s.value for s in states]), limit)
+    else:
+        sql = "SELECT * FROM remote_tasks ORDER BY created_at DESC LIMIT ?"
+        params = (limit,)
     with db_conn() as conn:
         rows = conn.execute(sql, params).fetchall()
         return [_row_to_task(r, _attempts_for(conn, r["id"])) for r in rows]
@@ -611,6 +814,7 @@ def purge_finished(
     older_than_seconds: float = 7 * 24 * 3600,
     now: Optional[float] = None,
     root: Optional[str] = None,
+    limit: Optional[int] = None,
 ) -> int:
     """Drop old finished tasks — rows *and* the bytes they own.
 
@@ -619,31 +823,66 @@ def purge_finished(
     audio. Neither was ever deleted, so the feature grew the user's disk for
     as long as they used it.
     """
+    if limit is not None and limit <= 0:
+        return 0
     cutoff = resolve(now) - older_than_seconds
-    terminal = ", ".join(f"'{s.value}'" for s in TaskState if s.terminal)
+    terminal = json.dumps([s.value for s in TaskState if s.terminal])
     with db_conn() as conn:
         doomed = [
             row["id"]
             for row in conn.execute(
-                f"SELECT id FROM remote_tasks WHERE state IN ({terminal}) AND finished_at < ?",
-                (cutoff,),
+                """
+                SELECT id FROM remote_tasks
+                WHERE state IN (SELECT value FROM json_each(?))
+                  AND finished_at < ?
+                ORDER BY finished_at ASC, id ASC
+                LIMIT ?
+                """,
+                (terminal, cutoff, -1 if limit is None else int(limit)),
             ).fetchall()
         ]
-        conn.execute(
-            f"DELETE FROM remote_task_attempts WHERE task_id IN "
-            f"(SELECT id FROM remote_tasks WHERE state IN ({terminal}) AND finished_at < ?)",
-            (cutoff,),
-        )
-        cur = conn.execute(
-            f"DELETE FROM remote_tasks WHERE state IN ({terminal}) AND finished_at < ?", (cutoff,)
-        )
-        removed = cur.rowcount
+    # Results are attempt-scoped. Delete them before their task rows, so a
+    # crash or transient Windows lock cannot erase the only index from which a
+    # future sweep could find those bytes.
+    cleaned, _artifacts_removed = _purge_result_directories(doomed, root=root)
+    with db_conn() as conn:
+        eligible: list[str] = []
+        if cleaned:
+            eligible = [
+                row["id"]
+                for row in conn.execute(
+                    """
+                    SELECT id FROM remote_tasks
+                    WHERE id IN (SELECT value FROM json_each(?))
+                      AND state IN (SELECT value FROM json_each(?))
+                      AND finished_at < ?
+                    """,
+                    (json.dumps(cleaned), terminal, cutoff),
+                ).fetchall()
+            ]
+        removed = 0
+        if eligible:
+            conn.execute(
+                """
+                DELETE FROM remote_task_attempts
+                WHERE task_id IN (SELECT value FROM json_each(?))
+                """,
+                (json.dumps(eligible),),
+            )
+            cur = conn.execute(
+                """
+                DELETE FROM remote_tasks
+                WHERE id IN (SELECT value FROM json_each(?))
+                """,
+                (json.dumps(eligible),),
+            )
+            removed = cur.rowcount
         # Read the survivors inside the same transaction that deleted the
         # rows: an input is only unreferenced relative to what is left.
         referenced = _referenced_artifacts(conn)
-    # Filesystem work outside the transaction — a slow rmtree must not hold
-    # SQLite's write lock against the dispatch loop.
-    purge_artifacts(doomed, referenced, cutoff=cutoff, root=root)
+    # Shared content-addressed inputs remain discoverable without their old
+    # task row, so they can be swept after the transaction.
+    purge_artifacts((), referenced, cutoff=cutoff, root=root)
     return removed
 
 

backend/worker/tls.py

@@ -20,15 +20,21 @@ There is deliberately no way to disable verification.
 from __future__ import annotations
 
 import datetime as _dt
+import errno
 import ipaddress
 import logging
 import os
 import socket
 import ssl
+import tempfile
+import threading
+from collections.abc import Iterator
+from contextlib import contextmanager
 from dataclasses import dataclass
 from typing import Optional
 
 from cryptography import x509
+from cryptography.exceptions import InvalidSignature, UnsupportedAlgorithm
 from cryptography.hazmat.primitives import hashes, serialization
 from cryptography.hazmat.primitives.asymmetric import ec
 from cryptography.x509.oid import NameOID
@@ -39,6 +45,8 @@ logger = logging.getLogger("omnivoice.worker")
 
 _CERT_VALID_DAYS = 825  # the CA/Browser Forum maximum; long enough to be quiet
 _RENEW_WITHIN_DAYS = 30
+_PENDING_PAIR_MAGIC = b"omnivoice-tls-pair-v1\n"
+_PROCESS_CREDENTIAL_LOCK = threading.Lock()
 
 
 def unverified_client_context() -> ssl.SSLContext:
@@ -72,7 +80,7 @@ def _san_entries(hostnames: list[str]) -> list[x509.GeneralName]:
     """
     entries: list[x509.GeneralName] = []
     for host in hostnames:
-        host = (host or "").strip()
+        host = _normalise_host(host)
         if not host:
             continue
         try:
@@ -118,9 +126,23 @@ def covers(credentials: "ServerCredentials", host: str) -> bool:
         ).value
     except Exception:
         return False
-    names = set(san.get_values_for_type(x509.DNSName))
-    names |= {str(ip) for ip in san.get_values_for_type(x509.IPAddress)}
-    return host in names
+    names = {_normalise_host(name) for name in san.get_values_for_type(x509.DNSName)}
+    names |= {
+        _normalise_host(str(ip)) for ip in san.get_values_for_type(x509.IPAddress)
+    }
+    return _normalise_host(host) in names
+
+
+def _normalise_host(host: str) -> str:
+    """Canonicalise a SAN identity for comparison, not for DNS resolution."""
+    candidate = (host or "").strip().strip("[]")
+    try:
+        return str(ipaddress.ip_address(candidate))
+    except ValueError:
+        # DNS names are case-insensitive, and a trailing root dot does not
+        # name a different host. cryptography intentionally does no matching
+        # for us because the certificate is its own trust root.
+        return candidate.rstrip(".").lower()
 
 
 def default_hostnames() -> list[str]:
@@ -206,25 +228,50 @@ def load_or_create(
     desktop control plane presents as "my workers all went offline" with
     nothing in the UI explaining why.
     """
-    existing = _load(cert_path, key_path)
-    wanted = hostnames or default_hostnames()
-    # Every explicitly requested identity must be present. This matters for an
-    # inbound listener bound to a user-entered LAN address: keeping a stable
-    # certificate that does not name that address makes mandatory hostname
-    # verification fail even though its fingerprint is correct.
-    missing = [
-        host for host in wanted if existing is not None and not covers(existing, host)
-    ]
-    if existing is not None and not _expiring_soon(existing) and not missing:
-        return existing
-    if existing is not None:
-        reason = (
-            "expiring" if _expiring_soon(existing) else "missing a requested hostname"
-        )
-        logger.info("Control-plane certificate is %s — regenerating.", reason)
-    credentials = generate_self_signed(hostnames=wanted)
-    _save(cert_path, key_path, credentials)
-    return credentials
+    with _credential_lock(cert_path):
+        existing = _load(cert_path, key_path)
+        pending_present, pending = _load_pending_pair(cert_path, key_path)
+        if existing is not None and pending_present:
+            # A valid final pair is either the old generation (the transaction
+            # never began replacing it) or the fully committed new one. Both
+            # are safer than rotating a pin merely because cleanup was cut
+            # short. The lock makes it safe to remove the stale journal here.
+            _remove_pending_pair(cert_path, key_path)
+        elif existing is None and pending is not None:
+            logger.warning(
+                "Recovering an interrupted control-plane TLS credential update."
+            )
+            _install_pair(cert_path, key_path, pending)
+            _remove_pending_pair(cert_path, key_path)
+            existing = pending
+
+        wanted = hostnames or default_hostnames()
+        # Every explicitly requested identity must be present. This matters for
+        # a listener bound to a user-entered address: keeping a stable
+        # certificate that does not name it makes mandatory hostname
+        # verification fail even though its fingerprint is correct.
+        missing = [
+            host
+            for host in wanted
+            if existing is not None and not covers(existing, host)
+        ]
+        if existing is not None and not _expiring_soon(existing) and not missing:
+            return existing
+        if existing is not None:
+            reason = (
+                "expiring"
+                if _expiring_soon(existing)
+                else "missing a requested hostname"
+            )
+            logger.info("Control-plane certificate is %s — regenerating.", reason)
+        elif pending_present and pending is None:
+            logger.warning(
+                "Ignoring a corrupt interrupted TLS credential update and "
+                "generating a new pair."
+            )
+        credentials = generate_self_signed(hostnames=wanted)
+        _save(cert_path, key_path, credentials)
+        return credentials
 
 
 def _load(cert_path: str, key_path: str) -> Optional[ServerCredentials]:
@@ -233,11 +280,40 @@ def _load(cert_path: str, key_path: str) -> Optional[ServerCredentials]:
             cert_pem = fh.read()
         with open(key_path, "rb") as fh:
             key_pem = fh.read()
-    except (FileNotFoundError, PermissionError):
+    except FileNotFoundError:
         return None
+    return _credentials_from_pem(cert_pem, key_pem)
+
+
+def _credentials_from_pem(
+    cert_pem: bytes, key_pem: bytes
+) -> Optional[ServerCredentials]:
+    """Parse a credential pair only when the private key belongs to the cert."""
     try:
         certificate = x509.load_pem_x509_certificate(cert_pem)
-    except ValueError:
+        private_key = serialization.load_pem_private_key(key_pem, password=None)
+        certificate_public_key = certificate.public_key().public_bytes(
+            serialization.Encoding.DER,
+            serialization.PublicFormat.SubjectPublicKeyInfo,
+        )
+        private_public_key = private_key.public_key().public_bytes(
+            serialization.Encoding.DER,
+            serialization.PublicFormat.SubjectPublicKeyInfo,
+        )
+        certificate.public_key().verify(
+            certificate.signature,
+            certificate.tbs_certificate_bytes,
+            ec.ECDSA(certificate.signature_hash_algorithm),
+        )
+    except (
+        AttributeError,
+        InvalidSignature,
+        TypeError,
+        UnsupportedAlgorithm,
+        ValueError,
+    ):
+        return None
+    if certificate_public_key != private_public_key:
         return None
     return ServerCredentials(
         certificate_pem=cert_pem,
@@ -258,19 +334,205 @@ def _expiring_soon(
 
 
 def _save(cert_path: str, key_path: str, credentials: ServerCredentials) -> None:
-    os.makedirs(os.path.dirname(os.path.abspath(cert_path)), exist_ok=True)
-    with open(cert_path, "wb") as fh:
-        fh.write(credentials.certificate_pem)
-    # The private key gets the same 0600 treatment as worker keys.
-    fd = os.open(key_path, os.O_WRONLY | os.O_CREAT | os.O_TRUNC, 0o600)
-    try:
-        os.write(fd, credentials.private_key_pem)
-    finally:
-        os.close(fd)
+    """Durably replace a pair, leaving enough state to finish after a crash.
+
+    No filesystem atomically renames two independent paths. A mode-0600
+    journal is therefore made durable first; the loader uses it only when the
+    final paths are torn or corrupt. Each final path is itself an atomic
+    sibling rename, and the journal is removed only after both directory
+    entries and their contents are durable.
+    """
+    pending_path = _pending_pair_path(cert_path, key_path)
+    _atomic_write(pending_path, _encode_pending_pair(credentials))
+    _install_pair(cert_path, key_path, credentials)
+    _remove_pending_pair(cert_path, key_path)
+
+
+def _install_pair(
+    cert_path: str, key_path: str, credentials: ServerCredentials
+) -> None:
+    # Key first, certificate second: the public certificate is the commit
+    # marker. A crash between them is detected by _load and recovered from the
+    # already-durable pending pair.
+    _atomic_write(key_path, credentials.private_key_pem)
     try:
         os.chmod(key_path, 0o600)
     except OSError:
+        # Windows and some network filesystems do not honour POSIX modes; the
+        # key remains in the app's per-user data directory there.
         pass
+    _atomic_write(cert_path, credentials.certificate_pem)
+    installed = _load(cert_path, key_path)
+    if installed is None or installed.fingerprint != credentials.fingerprint:
+        raise OSError("TLS credential pair failed verification after persistence")
+
+
+def _atomic_write(path: str, payload: bytes) -> None:
+    """Fsync and atomically replace one mode-restricted credential file."""
+    directory = os.path.dirname(os.path.abspath(path))
+    _durable_makedirs(directory)
+    fd, temporary = tempfile.mkstemp(
+        prefix=f".{os.path.basename(path)}.", suffix=".tmp", dir=directory
+    )
+    try:
+        with os.fdopen(fd, "wb") as fh:
+            fh.write(payload)
+            fh.flush()
+            os.fsync(fh.fileno())
+        os.replace(temporary, path)
+        _fsync_parent_directory(directory)
+    except Exception:
+        try:
+            os.unlink(temporary)
+        except FileNotFoundError:
+            pass
+        raise
+
+
+def _pending_pair_path(cert_path: str, key_path: str) -> str:
+    cert_name = os.path.basename(cert_path)
+    key_name = os.path.basename(key_path)
+    directory = os.path.dirname(os.path.abspath(cert_path))
+    return os.path.join(directory, f".{cert_name}.{key_name}.pending")
+
+
+def _encode_pending_pair(credentials: ServerCredentials) -> bytes:
+    certificate = credentials.certificate_pem
+    return (
+        _PENDING_PAIR_MAGIC
+        + str(len(certificate)).encode("ascii")
+        + b"\n"
+        + certificate
+        + credentials.private_key_pem
+    )
+
+
+def _load_pending_pair(
+    cert_path: str, key_path: str
+) -> tuple[bool, Optional[ServerCredentials]]:
+    path = _pending_pair_path(cert_path, key_path)
+    try:
+        with open(path, "rb") as fh:
+            raw = fh.read()
+    except FileNotFoundError:
+        return False, None
+    if not raw.startswith(_PENDING_PAIR_MAGIC):
+        return True, None
+    size_line, separator, payload = raw[len(_PENDING_PAIR_MAGIC) :].partition(b"\n")
+    if not separator:
+        return True, None
+    try:
+        certificate_size = int(size_line)
+    except ValueError:
+        return True, None
+    if certificate_size <= 0 or certificate_size >= len(payload):
+        return True, None
+    return True, _credentials_from_pem(
+        payload[:certificate_size], payload[certificate_size:]
+    )
+
+
+def _remove_pending_pair(cert_path: str, key_path: str) -> None:
+    path = _pending_pair_path(cert_path, key_path)
+    try:
+        os.unlink(path)
+    except FileNotFoundError:
+        return
+    _fsync_parent_directory(os.path.dirname(path) or ".")
+
+
+def _fsync_parent_directory(directory: str) -> None:
+    """Make a preceding directory-entry change durable where supported."""
+    directory_flag = getattr(os, "O_DIRECTORY", None)
+    if directory_flag is None:
+        return
+    unsupported = {
+        errno.EINVAL,
+        getattr(errno, "ENOTSUP", errno.EINVAL),
+        getattr(errno, "EOPNOTSUPP", errno.EINVAL),
+    }
+    try:
+        descriptor = os.open(directory, os.O_RDONLY | directory_flag)
+    except OSError as exc:
+        if exc.errno in unsupported:
+            return
+        raise
+    try:
+        os.fsync(descriptor)
+    except OSError as exc:
+        if exc.errno not in unsupported:
+            raise
+    finally:
+        os.close(descriptor)
+
+
+def _durable_makedirs(directory: str) -> None:
+    """Persist every newly-created credential-directory entry."""
+    target = os.path.abspath(directory)
+    missing: list[str] = []
+    current = target
+    while not os.path.isdir(current):
+        if os.path.exists(current):
+            if os.path.isdir(current):
+                break
+            raise NotADirectoryError(current)
+        missing.append(current)
+        parent = os.path.dirname(current)
+        if parent == current:
+            break
+        current = parent
+
+    for path in reversed(missing):
+        try:
+            os.mkdir(path)
+        except FileExistsError:
+            if not os.path.isdir(path):
+                raise
+        _fsync_parent_directory(os.path.dirname(path) or ".")
+
+    if not missing:
+        # Retry the exact barrier that may have failed after a prior mkdir.
+        _fsync_parent_directory(os.path.dirname(target) or ".")
+
+
+@contextmanager
+def _credential_lock(cert_path: str) -> Iterator[None]:
+    """Serialise first-start/renewal across threads and app processes."""
+    directory = os.path.dirname(os.path.abspath(cert_path))
+    _durable_makedirs(directory)
+    lock_path = os.path.join(directory, f".{os.path.basename(cert_path)}.lock")
+    with _PROCESS_CREDENTIAL_LOCK:
+        descriptor = os.open(lock_path, os.O_RDWR | os.O_CREAT, 0o600)
+        is_windows = os.name == "nt"
+        acquired = False
+        try:
+            if is_windows:
+                import msvcrt  # noqa: PLC0415
+
+                if os.fstat(descriptor).st_size == 0:
+                    os.write(descriptor, b"\0")
+                    os.fsync(descriptor)
+                os.lseek(descriptor, 0, os.SEEK_SET)
+                msvcrt.locking(descriptor, msvcrt.LK_LOCK, 1)
+            else:
+                import fcntl  # noqa: PLC0415
+
+                fcntl.flock(descriptor, fcntl.LOCK_EX)
+            acquired = True
+            yield
+        finally:
+            try:
+                if acquired and is_windows:
+                    import msvcrt  # noqa: PLC0415
+
+                    os.lseek(descriptor, 0, os.SEEK_SET)
+                    msvcrt.locking(descriptor, msvcrt.LK_UNLCK, 1)
+                elif acquired:
+                    import fcntl  # noqa: PLC0415
+
+                    fcntl.flock(descriptor, fcntl.LOCK_UN)
+            finally:
+                os.close(descriptor)
 
 
 def pin_matches(certificate_der: bytes, expected_fingerprint: str) -> bool:

backend/worker/transport/client.py

@@ -48,8 +48,10 @@ from typing import Awaitable, Callable, Optional, Protocol
 
 import grpc
 
+from worker.async_utils import drain_task, to_thread_and_drain_on_cancel
 from worker import errors as worker_errors
 from worker import identity
+from worker.capacity import clamp_concurrency
 from worker.errors import ErrorClass, WorkerError
 from worker.identity import EnrollmentToken, WorkerKeypair
 from worker.protocol.gen import worker_v1_pb2 as pb
@@ -103,6 +105,8 @@ class ArtifactTransport(Protocol):
 
     async def stage_in(self, ref: pb.ArtifactRef, destination: str) -> None: ...
 
+    def result_acked(self, artifacts: list[pb.ArtifactRef]) -> None: ...
+
 # Used when an assignment carries no lease (an older control plane, or a test).
 # Mirrors deadlines.py's _HEARTBEAT_GRACE_S * 4.
 _DEFAULT_PROGRESS_LEASE_SECONDS = 120.0
@@ -115,6 +119,29 @@ _MIN_KEEPALIVE_INTERVAL_SECONDS = 0.05
 _EXECUTOR_KWARGS = frozenset({"on_progress", "on_model_loading", "fetch_input"})
 
 
+def _write_all(handle, payload: bytes) -> None:
+    """Write a complete chunk, including through short-writing file wrappers."""
+    remaining = memoryview(payload)
+    while remaining:
+        written = handle.write(remaining)
+        if written is None or written <= 0:
+            raise OSError("input destination made no write progress")
+        remaining = remaining[written:]
+
+
+def _close_and_remove(handle, destination: str) -> None:
+    """Finish file cleanup as one blocking operation after cancellation."""
+    if handle is not None:
+        with contextlib.suppress(OSError):
+            handle.close()
+    with contextlib.suppress(OSError):
+        os.remove(destination)
+
+
+class TerminalRegistrationError(RuntimeError):
+    """A registration failure that reconnecting cannot repair."""
+
+
 def keepalive_interval(lease_seconds: float) -> float:
     """How often a running task must renew its progress lease.
 
@@ -267,28 +294,44 @@ class WorkerClient:
         cancel: Optional[Callable[[str], Awaitable[None]]] = None,
         capability_probe: Optional[Callable[[], list[dict]]] = None,
         on_registered: Optional[Callable[[str], None]] = None,
+        on_activated: Optional[Callable[[str], None]] = None,
         artifacts: Optional["ArtifactTransport"] = None,
+        drain_active_work: Optional[Callable[[], Awaitable[None]]] = None,
     ) -> None:
         self.config = config
+        self.config.max_concurrent_tasks = clamp_concurrency(
+            self.config.max_concurrent_tasks
+        )
         self._execute = execute
         self._cancel = cancel
         self._capability_probe = capability_probe
+        # Discovery imports engine adapters and inspects model storage. Share
+        # one off-loop probe across reconnect/task/prewarm/idle refresh races;
+        # a cancelled waiter drains it before returning so no detached probe
+        # can mutate global engine state after authority is gone.
+        self._capability_probe_task: Optional[asyncio.Task] = None
         # Outbound mode moves artifacts with RPCs this side initiates
         # (UploadResult / DownloadArtifact), which is only possible because
         # this side dialled. In inbound mode the node cannot call the panel at
         # all, so both directions are driven from the panel and this hook
         # swaps in the staging that makes that work. None means outbound.
         self._artifacts = artifacts
+        self._drain_active_work = drain_active_work
         # Lets the agent persist the server-assigned id. Without it a restarted
         # worker signs its challenge with an empty worker_id, the signature
         # never matches, and reconnecting needs a fresh enrollment token —
         # which would make key-based identity pointless.
         self._on_registered = on_registered
+        # Register only reserves a provisional server generation. Readiness is
+        # published separately, after ConfigUpdate proves Control activated it.
+        self._on_activated = on_activated
+        self._activation_confirmed = False
         self._reporter_kwargs = _accepted_reporter_kwargs(execute)
         self._outbox = _Outbox()
         self._pending: dict[str, PendingResult] = {}
         self._running: dict[str, asyncio.Task] = {}
         self._keepalives: dict[str, asyncio.Task] = {}
+        self._maintenance: set[asyncio.Task] = set()
         self._epoch = 0
         self._session_token = ""
         # Negotiated by ConfigUpdate; None means "use the executor's own
@@ -298,6 +341,12 @@ class WorkerClient:
         # same channel rather than the control stream.
         self._stub = None
         self._stop = asyncio.Event()
+        # Drain is a graceful reconnect, not terminal shutdown. This event
+        # half-closes only the current stream after every active result is ACKed
+        # while ``_stop`` remains reserved for cancelling the agent itself.
+        self._reconnect_requested = asyncio.Event()
+        self._draining = False
+        self._accepting_assignments = True
 
     # ── Connection ────────────────────────────────────────────────────────
 
@@ -318,6 +367,8 @@ class WorkerClient:
 
     async def run_forever(self) -> None:
         """Connect, serve, and reconnect until stopped."""
+        if not self._stop.is_set():
+            self._accepting_assignments = True
         attempt = 0
         while not self._stop.is_set():
             try:
@@ -325,6 +376,13 @@ class WorkerClient:
                 attempt = 0
             except asyncio.CancelledError:
                 raise
+            except TerminalRegistrationError:
+                # The control plane has made a durable decision that this
+                # identity may not reconnect. Work deliberately survives an
+                # ordinary network drop, but must not survive revocation and
+                # keep using the GPU with no authority able to cancel it.
+                await self._cancel_active_work()
+                raise
             except Exception as exc:
                 attempt += 1
                 delay = backoff_delay(attempt)
@@ -338,6 +396,52 @@ class WorkerClient:
 
     async def stop(self) -> None:
         self._stop.set()
+        self._reconnect_requested.set()
+        await self._cancel_active_work()
+
+    async def _cancel_active_work(self) -> None:
+        """Cancel retained tasks without permanently disabling reconnect."""
+        # Close admission before taking any snapshots. Attach/Control may still
+        # have a frame ready while revocation drains an uninterruptible GPU
+        # call; accepting that frame here lets it escape the snapshot entirely.
+        self._accepting_assignments = False
+        maintenance = list(self._maintenance)
+        for task in maintenance:
+            task.cancel()
+        running = list(self._running.items())
+        keepalives = list(self._keepalives.values())
+        for key, task in running:
+            self._stop_keepalive(key)
+            task.cancel()
+        for keepalive in keepalives:
+            keepalive.cancel()
+        cancel_callbacks = []
+        if self._cancel is not None:
+            cancel_callbacks = [
+                asyncio.create_task(self._cancel(key.split("/")[0]))
+                for key, _task in running
+            ]
+        # Cancel every maintenance, task, and keepalive wrapper before awaiting
+        # any uninterruptible one.  A prewarm stuck in a model-load thread must
+        # not delay revocation of active user renders.
+        draining = [
+            *maintenance,
+            *(task for _key, task in running),
+            *keepalives,
+            *cancel_callbacks,
+        ]
+        if draining:
+            await asyncio.gather(
+                *draining, return_exceptions=True
+            )
+        self._maintenance.clear()
+        for key, task in running:
+            if self._running.get(key) is task:
+                self._running.pop(key, None)
+        if self._drain_active_work is not None:
+            await self._drain_active_work()
+        self._keepalives.clear()
+        self._pending.clear()
 
     async def _connect_once(self) -> None:
         async with self._channel() as channel:
@@ -359,6 +463,8 @@ class WorkerClient:
             try:
                 async for message in stream:
                     await self._on_server_message(message)
+                    if self._stop.is_set():
+                        break
             finally:
                 heartbeat.cancel()
                 # The channel closes with this block, so a stub kept past it
@@ -375,7 +481,27 @@ class WorkerClient:
     # before; inbound calls them from the Attach handler. Neither mode gets its
     # own copy of registration, zombie reconciliation or redelivery.
 
-    def build_register_request(self) -> pb.RegisterRequest:
+    async def _probe_capabilities(self) -> list[dict]:
+        if self._capability_probe is None:
+            return list(self.config.capabilities or [])
+        task = self._capability_probe_task
+        if task is None or task.done():
+            task = asyncio.create_task(
+                to_thread_and_drain_on_cancel(self._capability_probe),
+                name="worker-capability-probe",
+            )
+            self._capability_probe_task = task
+        try:
+            capabilities = await asyncio.shield(task)
+        except asyncio.CancelledError:
+            await drain_task(task)
+            raise
+        finally:
+            if task.done() and self._capability_probe_task is task:
+                self._capability_probe_task = None
+        return list(capabilities or [])
+
+    async def build_register_request(self) -> pb.RegisterRequest:
         """This worker's self-description. Identical in both modes."""
         challenge = identity.new_challenge()
         nonce = identity.new_challenge()
@@ -387,9 +513,8 @@ class WorkerClient:
                 nonce=nonce,
             )
         )
-        capabilities = (
-            self._capability_probe() if self._capability_probe else self.config.capabilities
-        )
+        capabilities = await self._probe_capabilities()
+        self.config.capabilities = capabilities
         return pb.RegisterRequest(
             envelope=pb.Envelope(sequence=self._epoch),
             protocol_version_min=PROTOCOL_VERSION,
@@ -403,7 +528,9 @@ class WorkerClient:
             key_id=self.config.keypair.key_id,
             host=codec.host_to_pb(self.config.host or describe_host()),
             capabilities=[codec.capability_to_pb(c) for c in capabilities],
-            max_concurrent_tasks=self.config.max_concurrent_tasks,
+            max_concurrent_tasks=clamp_concurrency(
+                self.config.max_concurrent_tasks
+            ),
             in_flight=[
                 codec.task_ref(t.split("/")[0], t.split("/")[1], self._epoch)
                 for t in self._running
@@ -415,27 +542,80 @@ class WorkerClient:
     async def accept_registration(self, response: pb.RegisterResponse) -> None:
         """Adopt the control plane's answer and recover in-flight state."""
         if response.error.code:
-            raise RuntimeError(f"{response.error.code}: {response.error.message}")
+            raise TerminalRegistrationError(
+                f"{response.error.code}: {response.error.message}"
+            )
+        # An enrollment token is already spent when this response arrives.
+        # Commit the reconnect identity before adopting the live session; if
+        # local durable state cannot be written, retrying the spent token can
+        # never repair the worker and must reach the caller immediately.
+        if self._on_registered is not None:
+            try:
+                await to_thread_and_drain_on_cancel(
+                    self._on_registered, response.worker_id
+                )
+            except Exception as exc:
+                raise TerminalRegistrationError(
+                    "LOCAL_STATE: accepted enrollment could not be persisted"
+                ) from exc
 
+        self._draining = False
+        self._reconnect_requested.clear()
+        if not self._stop.is_set():
+            self._accepting_assignments = True
+        self._activation_confirmed = False
         self._epoch = response.session_epoch
         self._session_token = response.session_token
         self.config.worker_id = response.worker_id
         # The token is spent; every later connection proves key possession.
         self.config.enrollment_token = ""
-        if self._on_registered is not None:
-            try:
-                self._on_registered(response.worker_id)
-            except Exception:
-                logger.warning("Could not persist the worker id", exc_info=True)
 
         authoritative = {ref.attempt_id for ref in response.authoritative_in_flight}
         await self._cancel_zombies(authoritative)
         await self._redeliver_pending()
 
+    def confirm_activation(self) -> None:
+        """Publish readiness once Control proves the provisional session live."""
+        if self._activation_confirmed:
+            return
+        if self._on_activated is not None:
+            try:
+                self._on_activated(self.config.worker_id)
+            except Exception as exc:
+                raise TerminalRegistrationError(
+                    "LOCAL_STATE: activated enrollment could not be published"
+                ) from exc
+        self._activation_confirmed = True
+
     async def next_outbound(self) -> pb.WorkerMessage:
         """The next frame this worker wants to send."""
         return await self._outbox.get()
 
+    def prepare_inbound_session(self) -> None:
+        """Start a fresh stream while retaining running and pending work.
+
+        An inbound listener creates the protocol owner once per panel key, not
+        once per transport generation. Frames queued for the dead stream are
+        stale, but `_running` and `_pending` are precisely the state the next
+        Register must reconcile and redeliver.
+        """
+        self._outbox = _Outbox()
+        self._session_token = ""
+        self._stub = None
+        self._draining = False
+        self._reconnect_requested.clear()
+        if not self._stop.is_set():
+            self._accepting_assignments = True
+
+    @property
+    def reconnect_requested(self) -> bool:
+        return self._reconnect_requested.is_set()
+
+    @property
+    def outbound_pending(self) -> bool:
+        """Whether a terminal/control frame still needs transport delivery."""
+        return not self._outbox.empty()
+
     def start_heartbeat(self, response: pb.RegisterResponse) -> asyncio.Task:
         """Begin the heartbeat this session's liveness depends on.
 
@@ -458,14 +638,27 @@ class WorkerClient:
         await self._on_server_message(message)
 
     async def _register(self, stub) -> pb.RegisterResponse:
-        return await stub.Register(self.build_register_request())
+        return await stub.Register(await self.build_register_request())
 
     # ── Outbound ──────────────────────────────────────────────────────────
 
     async def _outbound(self):
-        while True:
-            message = await self._outbox.get()
-            yield message
+        while not self._reconnect_requested.is_set():
+            message = asyncio.create_task(self._outbox.get())
+            reconnect = asyncio.create_task(self._reconnect_requested.wait())
+            try:
+                done, _pending = await asyncio.wait(
+                    {message, reconnect}, return_when=asyncio.FIRST_COMPLETED
+                )
+                if message not in done:
+                    return
+                yield message.result()
+                self._maybe_finish_drain()
+            finally:
+                for task in (message, reconnect):
+                    if not task.done():
+                        task.cancel()
+                await asyncio.gather(message, reconnect, return_exceptions=True)
 
     async def _send(self, message: pb.WorkerMessage, *, bulk: bool = False) -> None:
         """Enqueue a frame. ``bulk`` is for anything that can be large.
@@ -479,17 +672,19 @@ class WorkerClient:
     async def _heartbeat_loop(self, interval: float) -> None:
         while True:
             await asyncio.sleep(interval)
-            await self._send(
-                pb.WorkerMessage(
-                    heartbeat=pb.Heartbeat(
-                        active_tasks=len(self._running),
-                        available_slots=max(
-                            0, self.config.max_concurrent_tasks - len(self._running)
-                        ),
-                        resident_models=self._resident_models(),
-                    )
-                )
+            await self._send(self.heartbeat_message())
+
+    def heartbeat_message(self) -> pb.WorkerMessage:
+        """Build the worker's current liveness/capacity frame."""
+        return pb.WorkerMessage(
+            heartbeat=pb.Heartbeat(
+                active_tasks=len(self._running),
+                available_slots=max(
+                    0, self.config.max_concurrent_tasks - len(self._running)
+                ),
+                resident_models=self._resident_models(),
             )
+        )
 
     def _resident_models(self) -> list[str]:
         return [
@@ -503,7 +698,7 @@ class WorkerClient:
         if self._capability_probe is None:
             return
         try:
-            capabilities = self._capability_probe()
+            capabilities = await self._probe_capabilities()
             self.config.capabilities = capabilities
             await self._send(pb.WorkerMessage(capabilities=pb.CapabilityUpdate(
                 capabilities=[codec.capability_to_pb(c) for c in capabilities]
@@ -531,9 +726,14 @@ class WorkerClient:
         # keepalive for an attempt the server has disowned is exactly the
         # frame that resurrects a cancelled task.
         self._stop_keepalive(key)
-        task = self._running.pop(key, None)
+        task = self._running.get(key)
         if task is not None:
             task.cancel()
+            # CancelAck releases server capacity. Do not send it while an
+            # uninterruptible synthesis/load thread still owns the GPU, or the
+            # replacement assignment can overlap and corrupt or OOM the worker.
+            await asyncio.gather(task, return_exceptions=True)
+            self._running.pop(key, None)
         if self._cancel is not None:
             await self._cancel(key.split("/")[0])
 
@@ -550,24 +750,76 @@ class WorkerClient:
             )
         elif kind == "result_ack":
             # Only now is it safe to forget the result.
-            self._pending.pop(self._key(message.result_ack.ref), None)
+            pending = self._pending.pop(self._key(message.result_ack.ref), None)
+            if pending is not None and self._artifacts is not None:
+                result_acked_async = getattr(
+                    self._artifacts, "result_acked_async", None
+                )
+                if callable(result_acked_async):
+                    await result_acked_async(pending.artifacts)
+                else:
+                    result_acked = getattr(self._artifacts, "result_acked", None)
+                    if callable(result_acked):
+                        result_acked(pending.artifacts)
+            self._maybe_finish_drain()
         elif kind == "config":
             if message.config.max_concurrent_tasks:
-                self.config.max_concurrent_tasks = message.config.max_concurrent_tasks
+                self.config.max_concurrent_tasks = clamp_concurrency(
+                    message.config.max_concurrent_tasks
+                )
             if message.config.inline_result_threshold_bytes:
                 # Negotiated, so the two sides cannot drift: the control plane
                 # is the one that knows how much it is willing to take on the
                 # control stream, and it may lower this at any time.
                 self._inline_threshold = int(message.config.inline_result_threshold_bytes)
+            self.confirm_activation()
         elif kind == "ping":
             # Answer immediately; the server times the round trip.
             await self._send(pb.WorkerMessage(pong=pb.Pong(nonce=message.ping.nonce)))
         elif kind == "drain":
-            self._stop.set()
+            self._accepting_assignments = False
+            self._draining = True
+            if message.drain.reconnect_to:
+                self.config.endpoint = message.drain.reconnect_to
+            self._maybe_finish_drain()
         elif kind == "shutdown":
+            await self._cancel_active_work()
             self._stop.set()
+            if self._artifacts is not None:
+                await self._send(
+                    pb.WorkerMessage(
+                        goodbye=pb.WorkerGoodbye(
+                            reason="The control-plane connection was removed."
+                        )
+                    )
+                )
+            # Publish the terminal acknowledgement before asking the inbound
+            # Attach loop to leave.  Reversing these lets that loop observe
+            # reconnect_requested and close the stream with Goodbye still
+            # queued, so the control plane cannot prove remote work drained.
+            self._reconnect_requested.set()
         elif kind == "prewarm":
-            asyncio.create_task(self._on_prewarm(message.prewarm), name="worker-prewarm")
+            if not self._accepting_assignments:
+                return
+            task = asyncio.create_task(
+                self._on_prewarm(message.prewarm), name="worker-prewarm"
+            )
+            self._maintenance.add(task)
+            task.add_done_callback(self._maintenance_finished)
+
+    def _maintenance_finished(self, task: asyncio.Task) -> None:
+        self._maintenance.discard(task)
+        self._maybe_finish_drain()
+
+    def _maybe_finish_drain(self) -> None:
+        if (
+            self._draining
+            and not self._running
+            and not self._pending
+            and not self._maintenance
+            and self._outbox.empty()
+        ):
+            self._reconnect_requested.set()
 
     async def _on_prewarm(self, request: pb.PrewarmRequest) -> None:
         """Load/download a catalog model, then report the resulting capability."""
@@ -591,15 +843,20 @@ class WorkerClient:
                 await self._install_catalog_repo(repo_ids[0])
             from worker.executor import TaskExecutor  # noqa: PLC0415
 
-            await asyncio.to_thread(TaskExecutor._load_backend, engine)
+            await to_thread_and_drain_on_cancel(TaskExecutor._load_backend, engine)
+        except asyncio.CancelledError:
+            raise
         except Exception:
             logger.warning("Prewarm failed for %s", engine or request.model_id, exc_info=True)
-        finally:
-            await self.refresh_capabilities()
+        await self.refresh_capabilities()
 
     async def _install_catalog_repo(self, repo_id: str) -> None:
         """Run the existing setup installer and pipe its hf_progress upstream."""
-        from api.routers.setup.download import InstallModelRequest, install_model  # noqa: PLC0415
+        from api.routers.setup.download import (  # noqa: PLC0415
+            InstallModelRequest,
+            cancel_install_and_wait,
+            install_model,
+        )
         from utils import download_aggregator, hf_progress  # noqa: PLC0415
 
         hf_progress.install()
@@ -627,12 +884,20 @@ class WorkerClient:
                 loop.call_soon_threadsafe(_finish)
 
         listener_id = hf_progress.register_listener(listener)
+        started_install = False
+        install_completed = False
         try:
-            await install_model(InstallModelRequest(repo_id=repo_id, target="local"))
+            response = await install_model(
+                InstallModelRequest(repo_id=repo_id, target="local")
+            )
+            started_install = response.get("status") == "install_started"
             event = await asyncio.wait_for(terminal, timeout=_FALLBACK_MODEL_LOAD_SECONDS)
             if event.get("phase") != "install_done":
                 raise RuntimeError(event.get("error") or "model install did not complete")
+            install_completed = True
         finally:
+            if started_install and not install_completed:
+                await cancel_install_and_wait(repo_id)
             hf_progress.unregister_listener(listener_id)
 
     @staticmethod
@@ -641,6 +906,20 @@ class WorkerClient:
 
     async def _on_assignment(self, assignment: pb.TaskAssignment) -> None:
         key = self._key(assignment.ref)
+        if not self._accepting_assignments or self._stop.is_set():
+            await self._send(
+                pb.WorkerMessage(
+                    rejected=pb.TaskRejected(
+                        ref=assignment.ref,
+                        error=pb.Error(
+                            error_class=pb.ERROR_CLASS_TRANSIENT,
+                            code="WORKER_STOPPING",
+                            message="The worker is relinquishing this control plane.",
+                        ),
+                    )
+                )
+            )
+            return
         if len(self._running) >= self.config.max_concurrent_tasks:
             # Declining because we are full is normal and penalty-free; the
             # scheduler's view of our capacity is only ever advisory.
@@ -672,9 +951,11 @@ class WorkerClient:
             # release the slot too, or the reserved task keeps running work
             # the scheduler never saw accepted — and double-executes after
             # reassignment. The stream-death case lands here as well.
-            task = self._running.pop(key, None)
+            task = self._running.get(key)
             if task is not None:
                 task.cancel()
+                await asyncio.gather(task, return_exceptions=True)
+                self._running.pop(key, None)
             raise
 
     async def _run(self, assignment: pb.TaskAssignment) -> None:
@@ -745,14 +1026,32 @@ class WorkerClient:
             failure: WorkerError = (
                 exc.error if isinstance(exc, TaskFailure) else worker_errors.from_exception(exc)
             )
+            if (
+                failure.error_class is ErrorClass.TIMEOUT
+                and self._drain_active_work is not None
+            ):
+                # A timeout only ends the coroutine's wait; Python cannot
+                # cancel the GPU thread underneath it. Do not send FAILED or
+                # release this admission slot until the executor proves that
+                # work relinquished the device, otherwise a capacity-1 worker
+                # can accept a replacement on top of the timed-out render.
+                await self._drain_active_work()
             await self._fail(assignment.ref, failure)
         finally:
             # Also covers the abnormal exits — cancellation, a crash between
             # the two _stop_keepalive calls above — so the timer can never
             # outlive the task that owns it.
             self._stop_keepalive(key)
-            self._running.pop(key, None)
-            await self.refresh_capabilities()
+            # Discovery may inspect the just-used backend. Keep the admission
+            # reservation until that off-loop probe has drained, otherwise a
+            # heartbeat can advertise the slot while this generation still
+            # owns task-finalization work. The inner finally also releases it
+            # when shutdown cancels the refresh waiter.
+            try:
+                await self.refresh_capabilities()
+            finally:
+                self._running.pop(key, None)
+                self._maybe_finish_drain()
 
     async def _fail(self, ref: pb.TaskRef, error: WorkerError) -> None:
         await self._send(
@@ -855,8 +1154,12 @@ class WorkerClient:
         await self._report_upload(ref, 0.0)
 
         offset = 0
+        metadata = ((SESSION_METADATA_KEY, self._session_token),)
         for _ in range(_MAX_UPLOAD_RESUMES):
-            ack = await stub.UploadResult(self._result_chunks(ref, artifact, payload, offset))
+            ack = await stub.UploadResult(
+                self._result_chunks(ref, artifact, payload, offset),
+                metadata=metadata,
+            )
             if ack.committed:
                 break
             resumed = int(ack.bytes_received)
@@ -967,25 +1270,29 @@ class WorkerClient:
         request.session_token = self._session_token
         offset = 0
         complete = False
+        handle = None
         try:
-            with open(destination, "wb") as handle:
-                async for chunk in self._stub.DownloadArtifact(request):
-                    if int(chunk.offset) != offset:
-                        raise RuntimeError(
-                            f"input offset {chunk.offset} did not match {offset} bytes received"
-                        )
-                    handle.write(chunk.data)
-                    offset += len(chunk.data)
-                    if chunk.last:
-                        complete = True
-                        break
+            handle = await to_thread_and_drain_on_cancel(open, destination, "wb")
+            async for chunk in self._stub.DownloadArtifact(request):
+                if int(chunk.offset) != offset:
+                    raise RuntimeError(
+                        f"input offset {chunk.offset} did not match {offset} bytes received"
+                    )
+                await to_thread_and_drain_on_cancel(_write_all, handle, chunk.data)
+                offset += len(chunk.data)
+                if chunk.last:
+                    complete = True
+                    break
+            await to_thread_and_drain_on_cancel(handle.close)
+            handle = None
+        except asyncio.CancelledError:
+            await to_thread_and_drain_on_cancel(_close_and_remove, handle, destination)
+            raise
         except Exception:
-            with contextlib.suppress(OSError):
-                os.remove(destination)
+            await to_thread_and_drain_on_cancel(_close_and_remove, handle, destination)
             raise
         if not complete:
-            with contextlib.suppress(OSError):
-                os.remove(destination)
+            await to_thread_and_drain_on_cancel(_close_and_remove, None, destination)
             raise RuntimeError("input download ended before its final chunk")
 
     def _executor_kwargs(self, assignment: pb.TaskAssignment) -> dict[str, Callable]:
@@ -1120,6 +1427,7 @@ __all__ = [
     "MAX_MESSAGE_BYTES",
     "UPLOAD_STAGE",
     "PendingResult",
+    "TerminalRegistrationError",
     "WorkerClient",
     "WorkerConfig",
     "backoff_delay",

backend/worker/transport/codec.py

@@ -10,7 +10,7 @@ import logging
 import os
 from typing import Optional
 
-from worker.capacity import derive_concurrency
+from worker.capacity import clamp_concurrency, derive_concurrency
 from worker.deadlines import Deadlines
 from worker.errors import ErrorClass, WorkerError
 from worker.lifecycle import Attempt, PriorityClass, Task
@@ -225,7 +225,7 @@ def capability_to_pb(cap: dict) -> pb.ModelCapability:
         resident=bool(cap.get("resident")),
         min_memory_bytes=int(cap.get("min_memory_bytes") or 0),
         precision=str(cap.get("precision") or ""),
-        derived_concurrency=max(0, declared),
+        derived_concurrency=clamp_concurrency(declared, allow_zero=True),
         cpu_fallback=bool(cap.get("cpu_fallback")),
         repo_ids=list(cap.get("repo_ids") or []),
         display_name=str(cap.get("display_name") or ""),
@@ -243,7 +243,9 @@ def capability_from_pb(message: pb.ModelCapability) -> dict:
         "resident": message.resident,
         "min_memory_bytes": message.min_memory_bytes,
         "precision": message.precision,
-        "derived_concurrency": message.derived_concurrency,
+        "derived_concurrency": clamp_concurrency(
+            message.derived_concurrency, allow_zero=True
+        ),
         "cpu_fallback": message.cpu_fallback,
         "repo_ids": list(message.repo_ids),
         "display_name": message.display_name,

bin/README.md

@@ -7,7 +7,8 @@ VoiceStudio installer. Today:
 |------|------------|---------|
 | `omnivoice-tts-darwin-arm64` | `ServeurpersoCom/omnivoice.cpp` @ pinned SHA | GGUF inference runtime — Apple Silicon |
 | `omnivoice-tts-darwin-x86_64` | same | Intel Mac |
-| `omnivoice-tts-linux-x86_64` | same | Linux |
+| `omnivoice-tts-linux-x86_64` | same | Linux (x86_64) |
+| `omnivoice-tts-linux-aarch64` | same | Linux ARM64 — Apple Silicon under Asahi; built with GGML Vulkan where the toolchain supports it, so the Honeykrisp driver can accelerate generation |
 | `omnivoice-tts-windows-x86_64.exe` | same | Windows |
 | `checksums.sha256` | computed by `scripts/build-omnivoice-tts.sh` | SHA-256 manifest — verified by `VoiceStudioGGUFBackend.is_available()` |
 

bun.lock

@@ -15,7 +15,7 @@
     },
     "frontend": {
       "name": "omnivoice-studio",
-      "version": "0.5.0",
+      "version": "0.5.2",
       "dependencies": {
         "@fontsource-variable/inter": "^5.3.0",
         "@fontsource-variable/source-serif-4": "^5.3.0",

deploy/docker-compose.yml

@@ -5,9 +5,11 @@
 #    docker compose -f deploy/docker-compose.yml --profile cpu up    # CPU mode
 #    docker compose -f deploy/docker-compose.yml --profile gpu up    # NVIDIA GPU
 #    docker compose -f deploy/docker-compose.yml --profile rocm up   # AMD GPU (ROCm)
+#    docker compose -f deploy/docker-compose.yml --profile worker-gpu up   # NVIDIA worker
+#    docker compose -f deploy/docker-compose.yml --profile worker-rocm up  # AMD worker
 #
-#  All services bind to port 3900, so they MUST be opt-in via profiles —
-#  otherwise `compose up` would race them and one would fail to bind.
+#  The Studio services publish the UI on loopback and the authenticated TLS
+#  worker control plane on 7443. Worker-only services publish no host port.
 #
 #  First run downloads ~4 GB of models. Progress is shown in logs.
 #  Open http://localhost:3900 once the health check passes.
@@ -15,9 +17,9 @@
 #  SECURITY: The port is bound to 127.0.0.1 by default — only this
 #  machine can reach the API. To expose VoiceStudio on your LAN (or
 #  through a reverse proxy / tunnel), change the port mapping to
-#  "0.0.0.0:3900:3900" or "3900:3900". VoiceStudio itself ships no
-#  authentication — if you expose it, put it behind a reverse proxy
-#  with auth (Caddy basic_auth, nginx + htpasswd, Tailscale, etc.).
+#  "0.0.0.0:3900:3900" or "3900:3900". Export a long random
+#  OMNIVOICE_API_KEY before starting a Studio profile; use HTTPS or an
+#  encrypted private overlay whenever traffic leaves a fully trusted LAN.
 # ──────────────────────────────────────────────────────────────
 
 services:
@@ -33,6 +35,7 @@ services:
     profiles: ["cpu"]
     ports:
       - "127.0.0.1:3900:3900"
+      - "${OMNIVOICE_WORKER_PUBLISH_HOST:-127.0.0.1}:${OMNIVOICE_WORKER_PORT:-7443}:${OMNIVOICE_WORKER_PORT:-7443}"
     volumes:
       - omnivoice-data:/app/omnivoice_data
     environment:
@@ -53,6 +56,11 @@ services:
       # discoverable. If you front the container with your own auth proxy on
       # loopback, set this to 0 to re-enable the strict gate.
       - OMNIVOICE_SERVER_MODE=1
+      # Required for server-mode settings, diagnostics, and other admin
+      # mutations because Docker NAT cannot prove a browser is loopback.
+      - OMNIVOICE_API_KEY=${OMNIVOICE_API_KEY:?export a long random OMNIVOICE_API_KEY before starting a Studio profile}
+      - OMNIVOICE_WORKER_PORT=${OMNIVOICE_WORKER_PORT:-7443}
+      - OMNIVOICE_WORKER_ENDPOINT_HOST=${OMNIVOICE_WORKER_ENDPOINT_HOST:-}
     healthcheck:
       test: ["CMD", "curl", "-sf", "http://localhost:3900/health"]
       interval: 30s
@@ -71,6 +79,7 @@ services:
     profiles: ["gpu"]
     ports:
       - "127.0.0.1:3900:3900"
+      - "${OMNIVOICE_WORKER_PUBLISH_HOST:-127.0.0.1}:${OMNIVOICE_WORKER_PORT:-7443}:${OMNIVOICE_WORKER_PORT:-7443}"
     volumes:
       - omnivoice-data:/app/omnivoice_data
     environment:
@@ -86,6 +95,9 @@ services:
       # See the CPU service above — relaxes the loopback origin gate for the
       # headless Docker deployment (issue #261). Set to 0 to re-enable it.
       - OMNIVOICE_SERVER_MODE=1
+      - OMNIVOICE_API_KEY=${OMNIVOICE_API_KEY:?export a long random OMNIVOICE_API_KEY before starting a Studio profile}
+      - OMNIVOICE_WORKER_PORT=${OMNIVOICE_WORKER_PORT:-7443}
+      - OMNIVOICE_WORKER_ENDPOINT_HOST=${OMNIVOICE_WORKER_ENDPOINT_HOST:-}
     healthcheck:
       test: ["CMD", "curl", "-sf", "http://localhost:3900/health"]
       interval: 30s
@@ -121,6 +133,7 @@ services:
     profiles: ["rocm"]
     ports:
       - "127.0.0.1:3900:3900"
+      - "${OMNIVOICE_WORKER_PUBLISH_HOST:-127.0.0.1}:${OMNIVOICE_WORKER_PORT:-7443}:${OMNIVOICE_WORKER_PORT:-7443}"
     devices:
       - /dev/kfd
       - /dev/dri
@@ -136,6 +149,9 @@ services:
       # loopback origin gate for the headless Docker deployment.
       - OMNIVOICE_BIND_HOST=0.0.0.0
       - OMNIVOICE_SERVER_MODE=1
+      - OMNIVOICE_API_KEY=${OMNIVOICE_API_KEY:?export a long random OMNIVOICE_API_KEY before starting a Studio profile}
+      - OMNIVOICE_WORKER_PORT=${OMNIVOICE_WORKER_PORT:-7443}
+      - OMNIVOICE_WORKER_ENDPOINT_HOST=${OMNIVOICE_WORKER_ENDPOINT_HOST:-}
       # RDNA3 consumer cards (RX 7900 XTX/XT and friends, gfx1100): if the
       # GPU is not detected, uncomment the override below. The backend
       # auto-sets it for known consumer GFX IDs, so try without it first.
@@ -148,5 +164,79 @@ services:
       start_period: 180s
     restart: unless-stopped
 
+  # ── Worker-only NVIDIA GPU mode
+  # Generate a join code on the control plane, then start with:
+  #   OMNIVOICE_WORKER_TOKEN='ovw_…' docker compose \
+  #     -f deploy/docker-compose.yml --profile worker-gpu up -d
+  # No port is published: uvicorn only hosts the application lifespan that
+  # owns the outbound worker agent. The browser UI is not needed or exposed.
+  omnivoice-worker-gpu:
+    image: ghcr.io/debpalash/omnivoice-studio:latest
+    build:
+      context: ..
+      dockerfile: deploy/Dockerfile
+    container_name: omnivoice-worker-gpu
+    profiles: ["worker-gpu"]
+    entrypoint: ["python3", "-m", "uvicorn"]
+    command: ["backend.main:app", "--host", "127.0.0.1", "--port", "3900"]
+    volumes:
+      - omnivoice-worker-data:/app/omnivoice_data
+    environment:
+      - HF_HOME=/app/omnivoice_data/huggingface
+      - HF_TOKEN=${HF_TOKEN:-}
+      - OMNIVOICE_DATA_DIR=/app/omnivoice_data
+      - PYTHONPATH=/app/backend
+      - PYTHONUNBUFFERED=1
+      - OMNIVOICE_SERVER_MODE=0
+      - OMNIVOICE_WORKER_MODE=1
+      - OMNIVOICE_WORKER_TOKEN=${OMNIVOICE_WORKER_TOKEN:-}
+      - OMNIVOICE_WORKER_ENDPOINT=${OMNIVOICE_WORKER_ENDPOINT:-}
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]
+    healthcheck:
+      test: ["CMD", "curl", "-fsS", "http://127.0.0.1:3900/workers/agent/readiness"]
+      interval: 10s
+      timeout: 5s
+      retries: 6
+      start_period: 180s
+    restart: unless-stopped
+
+  # ── Worker-only AMD GPU (ROCm) mode
+  omnivoice-worker-rocm:
+    image: ghcr.io/debpalash/omnivoice-studio:rocm
+    container_name: omnivoice-worker-rocm
+    profiles: ["worker-rocm"]
+    entrypoint: ["python3", "-m", "uvicorn"]
+    command: ["backend.main:app", "--host", "127.0.0.1", "--port", "3900"]
+    devices:
+      - /dev/kfd
+      - /dev/dri
+    volumes:
+      - omnivoice-worker-rocm-data:/app/omnivoice_data
+    environment:
+      - HF_HOME=/app/omnivoice_data/huggingface
+      - HF_TOKEN=${HF_TOKEN:-}
+      - OMNIVOICE_DATA_DIR=/app/omnivoice_data
+      - PYTHONPATH=/app/backend
+      - PYTHONUNBUFFERED=1
+      - OMNIVOICE_SERVER_MODE=0
+      - OMNIVOICE_WORKER_MODE=1
+      - OMNIVOICE_WORKER_TOKEN=${OMNIVOICE_WORKER_TOKEN:-}
+      - OMNIVOICE_WORKER_ENDPOINT=${OMNIVOICE_WORKER_ENDPOINT:-}
+    healthcheck:
+      test: ["CMD", "curl", "-fsS", "http://127.0.0.1:3900/workers/agent/readiness"]
+      interval: 10s
+      timeout: 5s
+      retries: 6
+      start_period: 180s
+    restart: unless-stopped
+
 volumes:
   omnivoice-data:
+  omnivoice-worker-data:
+  omnivoice-worker-rocm-data:

deploy/dockerhub-overview.md

@@ -43,21 +43,30 @@ the entire pipeline runs on CPU, just slower. Pull size: ~5 GB compressed
 ## Quick start (CPU)
 
 ```bash
+export OMNIVOICE_API_KEY="$(python3 -c 'import secrets; print(secrets.token_urlsafe(32))')"
+
 docker run -d --name omnivoice \
   -p 127.0.0.1:3900:3900 \
+  -e OMNIVOICE_API_KEY="$OMNIVOICE_API_KEY" \
   -v omnivoice-data:/app/omnivoice_data \
   -v ~/.cache/huggingface:/root/.cache/huggingface \
   palashdeb/omnivoice-studio:latest
 ```
 
 Open <http://localhost:3900>. The first run downloads a few GB of model weights —
-follow `docker logs -f omnivoice` to watch progress.
+follow `docker logs -f omnivoice` to watch progress. When the UI asks for an
+API key, paste the generated value; settings and diagnostic actions require
+this administrator session because Docker NAT hides the browser's true
+loopback origin.
 
 ## Quick start (NVIDIA GPU)
 
 ```bash
+export OMNIVOICE_API_KEY="$(python3 -c 'import secrets; print(secrets.token_urlsafe(32))')"
+
 docker run -d --name omnivoice --gpus all \
   -p 127.0.0.1:3900:3900 \
+  -e OMNIVOICE_API_KEY="$OMNIVOICE_API_KEY" \
   -v omnivoice-data:/app/omnivoice_data \
   -v ~/.cache/huggingface:/root/.cache/huggingface \
   palashdeb/omnivoice-studio:latest
@@ -74,9 +83,12 @@ CUDA-only and runs on CPU on AMD hardware). No toolkit needed — pass the GPU
 through as device nodes; the host only needs the `amdgpu` kernel driver:
 
 ```bash
+export OMNIVOICE_API_KEY="$(python3 -c 'import secrets; print(secrets.token_urlsafe(32))')"
+
 docker run -d --name omnivoice \
   --device /dev/kfd --device /dev/dri \
   -p 127.0.0.1:3900:3900 \
+  -e OMNIVOICE_API_KEY="$OMNIVOICE_API_KEY" \
   -v omnivoice-data:/app/omnivoice_data \
   -v ~/.cache/huggingface:/root/.cache/huggingface \
   palashdeb/omnivoice-studio:rocm
@@ -87,8 +99,9 @@ Podman users: same two `--device` flags (Quadlet: `AddDevice=/dev/kfd` +
 `-e HSA_OVERRIDE_GFX_VERSION=11.0.0` if the GPU isn't detected — details in
 the [Docker install guide](https://github.com/debpalash/VoiceStudio/blob/main/docs/install/docker.md).
 
-There's also a Compose file in the repo with `cpu` / `gpu` / `rocm` profiles
-— see the [Docker install guide](https://github.com/debpalash/VoiceStudio/blob/main/docs/install/docker.md).
+There's also a Compose file in the repo with `cpu` / `gpu` / `rocm` profiles,
+plus `worker-gpu` / `worker-rocm` profiles that lend a headless GPU without
+publishing the web UI — see the [Docker install guide](https://github.com/debpalash/VoiceStudio/blob/main/docs/install/docker.md).
 
 ---
 
@@ -98,12 +111,12 @@ There's also a Compose file in the repo with `cpu` / `gpu` / `rocm` profiles
 |-----|--------------|
 | `:latest` | **Rolling preview** — latest commit on `main`, at or ahead of the last release. This is the preview channel; pin `:stable` for production. |
 | `:stable` | Most recent versioned release (updated on every `v*` git tag) |
-| `:0.5.0` | Exact release version |
+| `:0.5.2` | Exact release version |
 | `:0.5` | Latest patch within the `0.5` minor |
 | `:main` | Alias of the same rolling `main` build as `:latest` |
 | `:sha-xxxxxxx` | A specific commit (produced by manual workflow dispatch) |
 | `:rocm` | **AMD GPU (ROCm) build** of the rolling preview — the ROCm analogue of `:latest` |
-| `:stable-rocm`, `:0.5.0-rocm`, `:0.5-rocm`, `:sha-xxxxxxx-rocm` | ROCm builds of the corresponding tags above |
+| `:stable-rocm`, `:0.5.2-rocm`, `:0.5-rocm`, `:sha-xxxxxxx-rocm` | ROCm builds of the corresponding tags above |
 
 Preview builds always come from `main` and never version-sort below `:stable`,
 so upgrades flow naturally. The same images and tags

docs/RELEASING.md

@@ -96,7 +96,7 @@ bug to fix immediately, not backlog.
 | Channel | Source | Produced by | How to verify |
 |---|---|---|---|
 | GitHub Release: installers + signed `latest.json` (**Stable** updater channel) | the `vX.Y.Z` tag | `release.yml` on tag push | Release page has dmg (arm+intel), msi/exe, AppImage/deb, `latest.json`; body = the CHANGELOG section (not the auto-generated fallback), followed by per-platform checksums and a **Contributors** avatar strip (owner + every PR author for the tag — the `contributors-strip` job) |
-| **Preview** updater channel (rolling `preview` prerelease) | **`main` only** | `release.yml` nightly cron / manual dispatch | preview `latest.json` stamps `X.Y.Z-N` and semver-sorts above stable |
+| **Preview** updater channel (rolling `preview` prerelease) | **`main` only** | `release.yml` nightly cron / manual dispatch | preview `latest.json` uses main's version when it is ahead; otherwise it advances the stable patch, then appends `-N` so it semver-sorts above stable |
 | GHCR CUDA image: `:X.Y.Z`, `:X.Y`, `:stable` | the tag | `docker.yml` on tag push | `docker manifest inspect ghcr.io/debpalash/omnivoice-studio:X.Y.Z` |
 | GHCR ROCm image: `:X.Y.Z-rocm`, `:X.Y-rocm`, `:stable-rocm` | the tag | `docker.yml` on tag push | same, with `-rocm` suffix |
 | Docker Hub mirror of **all** the above tags | the tag | `docker.yml` (gated on `DOCKERHUB_*` secrets) | tag list at hub.docker.com/r/palashdeb/omnivoice-studio/tags |

docs/ROADMAP.md

@@ -18,7 +18,7 @@ Phase 5 · Productisation          ░░░░░░░░░░  0 / 5
 
 Design track        ▓▓▓▓▓▓▓▓▓░  ongoing · 14 primitives + ~67 migrated inline styles · DubTab/Header/Sidebar/CloneDesignTab drained
 Performance track   ▓▓▓░░░░░░░  underway · profiling, preload, isolated engines + cache-remix I/O
-Feature-magic track ░░░░░░░░░░  not started
+Feature-magic track ▓▓░░░░░░░░  underway · project-level casting board shipped
 Quality track       ▓▓░░░░░░░░  12 smoke tests, 10 error messages rewritten
 ```
 
@@ -201,19 +201,19 @@ None on the critical path to world-class. All are answers to real demand.
 | Kill per-segment disk round-trip | 🟡 | Long-video assembly stays disk-backed to bound RAM. Unchanged same-rate natural segments now skip the redundant decode → scratch encode → decode cycle; fresh segments still persist once and reload for assembly. |
 | Cold start ≤1.5 s to first audible sample | 🟡 | Installed models preload in the background and `scripts/bench_pipeline.py` measures cold/warm synthesis; target is not yet verified. |
 | Speculative regeneration on hover | ⏳ | — |
-| Crash-sandbox engines (subprocess isolation) | 🟡 | Killable sidecar engines and opt-in `omnivoice-subprocess` are live; the default in-process engine can still take down the server on a native crash. |
+| Crash-sandbox engines (subprocess isolation) | 🟡 | Killable sidecar engines and opt-in `omnivoice-subprocess` are live; the default OmniVoice engine now routes through the crash sandbox on MPS, while CUDA/ROCm/CPU retain the lower-overhead in-process path. |
 | Interaction budgets (<50 ms UI, <200 ms preview, <4 s first seg) | 🟡 | `/ws/tts` reports real TTFA, total generation time and RTF; frontend responsiveness instrumentation exists, but no cross-surface budget gate yet. |
 | Dedicated dev-week per quarter | ⏳ | Cadence not yet booked. |
 
-### ✨ Feature-magic track _(⏳ not started)_
+### ✨ Feature-magic track _(🟡 underway)_
 
 | Feature | Status | Phase gate |
 |------|:---:|------|
-| Project-level casting view (drag voices to speakers) | ⏳ | After Phase 3 |
+| Project-level casting view (drag voices to speakers) | ✅ | Shipped (#1767): the dub CAST strip expands into a casting board — drag voice chips onto speaker rows, keyboard listbox included, same fields as the dropdowns. |
 | Voice memory across projects | ⏳ | After Phase 4 |
 | Context-aware pipeline (video frames → pipeline decisions) | ⏳ | After Phase 4 |
 | On-device learning from corrections (user edits → LoRA) | ⏳ | Research only; possibly Phase 5+ |
-| Real-time dub preview (stream TTS as you edit) | ⏳ | After Phase 4.1 |
+| Real-time dub preview (stream TTS as you edit) | ✅ | Shipped 2026-09-02 (#1769) — opt-in "Live preview" toggle on the dub segment table streams the edited line over `/ws/tts` with its CAST voice; export path unchanged. |
 
 ### 🧪 Quality track _(🟡 underway)_
 

docs/adr/SPIKE-01-gguf-research.md

@@ -77,7 +77,7 @@ Both spike URLs are **real, live, and the intended artifacts**. The "VoiceStudio
 
 # GGUF variant — adds a bundled binary, not a Python dep
 # Built once per platform in CI and stored alongside the Tauri installer:
-#   bin/omnivoice-tts-{darwin-arm64,darwin-x86_64,windows-x86_64,linux-x86_64}
+#   bin/omnivoice-tts-{darwin-arm64,darwin-x86_64,windows-x86_64,linux-x86_64,linux-aarch64}
 # Quants pulled at first use via huggingface_hub
 ```
 
@@ -149,7 +149,8 @@ User opens app / first run           │  Settings → Engines → Default
 ┌─────────────────────────────────────────────────────────────────────────┐
 │ SubprocessBackend (Phase 2 primitive)                                   │
 │                                                                         │
-│  spawn:  bin/omnivoice-tts-{darwin-arm64|x86_64|win|linux}              │
+│  spawn:  bin/omnivoice-tts-{darwin-arm64|x86_64|win|linux|              │
+│                              linux-aarch64}                             │
 │          --model    $HF_HUB_CACHE/.../omnivoice-base-{Q}.gguf           │
 │          --codec    $HF_HUB_CACHE/.../omnivoice-tokenizer-{Q}.gguf      │
 │          --lang     <user lang>                                         │
@@ -229,7 +230,8 @@ bin/                          # bundled in installer (per platform)
 ├── omnivoice-tts-darwin-arm64
 ├── omnivoice-tts-darwin-x86_64
 ├── omnivoice-tts-windows-x86_64.exe
-└── omnivoice-tts-linux-x86_64
+├── omnivoice-tts-linux-x86_64
+└── omnivoice-tts-linux-aarch64
 
 .planning/decisions/
 ├── SPIKE-01-gguf.md          # ADR (this research → human review → ratified)

docs/agentic-voice.md

@@ -1,10 +1,14 @@
 # Agentic voice: VoiceStudio as a TTS/STT provider
 
-VoiceStudio exposes an **OpenAI-compatible API**, so any agent framework that
+VoiceStudio exposes a **local speech platform**—OpenAI-compatible batch audio,
+a versioned transcription WebSocket, native dictation control, and MCP—so any agent framework that
 speaks to OpenAI's audio endpoints can use your local VoiceStudio for speech —
 in your own cloned voice, with nothing leaving your machine. You bring the
 agent runtime; VoiceStudio is the voice.
 
+For dictating directly into Claude Code, Codex, Pi, Antigravity CLI, Herdr, or
+another focused prompt, use the [Rust control sidecar](speech-platform.md).
+
 This is "agentic v1": VoiceStudio is a provider, not the orchestrator. You wire
 your own agent (a support line, a desk assistant, a Discord persona) and point
 its TTS/STT at VoiceStudio.
@@ -16,13 +20,17 @@ its TTS/STT at VoiceStudio.
 
 ## The endpoints
 
-VoiceStudio serves these on `http://localhost:3900/v1` (or your
-[remote backend URL](remote-gpu.md)):
+VoiceStudio's service root is `http://localhost:3900` (or your
+[remote backend URL](remote-gpu.md)). OpenAI-compatible clients use
+`http://localhost:3900/v1` as their base URL, while discovery stays at the
+service root: `http://localhost:3900/.well-known/voicestudio-speech`.
 
 | OpenAI route | VoiceStudio support |
 |---|---|
 | `POST /v1/audio/speech` | TTS. `model` = engine id, `voice` = a voice-profile id (your clone) or preset, `response_format` incl. `pcm` and `wav`, `speed`. Default output is 24 kHz. |
 | `POST /v1/audio/transcriptions` | STT (Whisper-family). |
+| `WS /v1/audio/transcriptions/stream` | Live partial/final STT from PCM or WebM. |
+| `GET /.well-known/voicestudio-speech` | Machine-readable transport discovery. |
 | `GET /v1/audio/voices` | list available voices (VoiceStudio extension). |
 
 A contract test (`tests/test_agentic_provider_contract.py`) pins this request
@@ -74,10 +82,11 @@ single local agent, pipecat is lighter.
 
 ## Remote backend
 
-Running VoiceStudio on a [remote GPU box](remote-gpu.md)? Use that backend's URL
-as `base_url` and pass its `OMNIVOICE_API_KEY` as the `api_key` — the same
-bearer the rest of the app uses. Keep it on your tailnet, not the open
-internet.
+Running VoiceStudio on a [remote GPU box](remote-gpu.md)? Append `/v1` to that
+backend's service-root URL for the OpenAI client's `base_url`, and pass its
+`OMNIVOICE_API_KEY` as the `api_key` — the same bearer the rest of the app uses.
+Keep the unmodified service root for `/.well-known/voicestudio-speech`
+discovery, and keep the backend on your tailnet, not the open internet.
 
 ## Use your own voice responsibly
 

docs/api-auth.md

@@ -1,5 +1,8 @@
 # Authenticating the local API
 
+For an application backend consuming a pinned, private VoiceStudio deployment,
+start with the [production private API pattern](production-private-api.md).
+
 VoiceStudio's backend is **loopback-only and unauthenticated by default** — a
 script running on the same machine as `http://localhost:3900` needs no key, no
 PIN, no header. Everything on this page only matters once you reach the backend
@@ -201,7 +204,8 @@ ws://gpu-box:3900/ws/transcribe?api_key=<key>
 That URL form is retained for non-browser compatibility only. The first-party
 UI never constructs it. A bearer administrator session first calls
 `POST /api/auth/ws-ticket` and puts only the returned `ws_ticket` in the URL.
-Tickets are scoped to `/ws/transcribe` or `/ws/events`, expire after 30 seconds,
+Tickets are scoped to one of `/ws/transcribe`, `/ws/events` or `/ws/tts` (the
+live dub preview stream), expire after 30 seconds,
 return the same bounded `expires_in`/`expires_at` pair, and are consumed
 atomically at most once. Same-origin UI WebSockets use the
 HttpOnly session cookie and must pass exact `Origin` validation; `null`, missing,
@@ -335,7 +339,8 @@ drives exact-Origin checks and the session cookie's `Secure` attribute.
 For a public path prefix such as `/studio`, either strip that prefix before
 forwarding or configure the ASGI `root_path` to the same value. WebSocket ticket
 validation removes only that trusted, configured prefix; it never accepts an
-arbitrary path merely because it ends in `/ws/events` or `/ws/transcribe`.
+arbitrary path merely because it ends in `/ws/events`, `/ws/transcribe` or
+`/ws/tts`.
 
 ## Status codes
 

docs/dubbing/export.md

@@ -4,3 +4,20 @@ Video exports can contain both the source audio and one or more dubbed tracks.
 VoiceStudio marks the selected dubbed language as the default so ordinary video
 players and messaging apps play the dub immediately. Choose **Original** in the
 Default Track control when the source audio should play first instead.
+
+## Hardsub captions
+
+Turning on **Burn subtitles into picture (hardsub)** re-encodes the video with
+captions rendered into the frames. Two caption styles are available:
+
+- **Line** (default) — static line subtitles, unchanged from previous releases.
+- **Karaoke (word highlight)** — each word fills with the highlight colour in
+  sequence across the line. Word timings recorded at transcription time drive
+  the sweep when they still spell the burned text; otherwise (older jobs, or
+  translated tracks) the timing is spread evenly across each line. Karaoke is
+  not available together with the dual-layout (translation + original) style —
+  switching dual on falls back to the line burn.
+
+Smart Fit exports burn captions after the video retime, so both styles follow
+the fitted timeline. The karaoke script can also be downloaded on its own from
+`GET /dub/ass/{job_id}` as an `.ass` sidecar.