upstream-archive/VoiceStudio
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

docs(mcp): the first generation also downloads the model inside the budget

Browse Source
This commit is contained in:
Palash Debnath
2026-09-10 16:16:01 -07:00
parent bd9be744eb
commit a35c7babdf
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/mcp.md
+1 -1
View File
@@ -27,7 +27,7 @@ or another tool by path:
| Variable | Values | Effect |
|---|---|---|
| `OMNIVOICE_MCP_OUTPUT_MODE` | `resources` (default) · `files` · `both` | `resources` returns `wav_base64` inline (the original contract). `files` returns `audio_url` (the render served at `/audio/<audio_id>.wav`, which the backend keeps anyway) and, when a base path is set, `output_path` — the WAV written into that directory. `both` returns everything. |
| `OMNIVOICE_MCP_TIMEOUT_S` | seconds (default: follows the backend) | How long a tool waits on the backend. Unset, each tool waits as long as the backend itself would, plus 30 s, and never less than 120 s. `transcribe` follows `OMNIVOICE_ASR_TRANSCRIBE_TIMEOUT_S` (300 s by default, queue time included). `generate_speech` follows the backend's queue wait (`OMNIVOICE_GPU_QUEUE_TIMEOUT_S`, 1,800 s by default) plus the larger of `OMNIVOICE_GENERATE_TIMEOUT_S` and `OMNIVOICE_CPU_GENERATE_TIMEOUT_S`, plus 1 s per 40 characters past 1,200. The agent then gets the backend's own timeout error instead of an empty one. Set this to use one fixed wait for every tool. |
| `OMNIVOICE_MCP_TIMEOUT_S` | seconds (default: follows the backend) | How long a tool waits on the backend. Unset, each tool waits as long as the backend itself would, plus 30 s, and never less than 120 s. `transcribe` follows `OMNIVOICE_ASR_TRANSCRIBE_TIMEOUT_S` (300 s by default, queue time included). `generate_speech` follows the backend's queue wait (`OMNIVOICE_GPU_QUEUE_TIMEOUT_S`, 1,800 s by default) plus the larger of `OMNIVOICE_GENERATE_TIMEOUT_S` and `OMNIVOICE_CPU_GENERATE_TIMEOUT_S`, plus 1 s per 40 characters past 1,200. The agent then gets the backend's own timeout error instead of an empty one. Set this to use one fixed wait for every tool. On a machine that hasn't downloaded the OmniVoice model yet, the first `generate_speech` also downloads it (about 2.3 GB) inside the backend's budget: install the model first (first-run setup, or Model Catalogue) or raise `OMNIVOICE_GENERATE_TIMEOUT_S` before the first call. |
| `OMNIVOICE_MCP_BASE_PATH` | a directory | The **security boundary** for file-shaped traffic. `transcribe(audio_path=…)` and `clone_voice(ref_audio_path=…)` read only from inside it (relative paths resolve against it, absolute paths must already lie within it, symlinks are resolved before the check), and files mode writes only into it. With no base path configured, path arguments are refused with a reason. |
Input files are opened through confined, no-follow descriptors after path