diff --git a/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfiguration.svelte b/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfiguration.svelte
new file mode 100644
index 0000000000..23e0ab6451
--- /dev/null
+++ b/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfiguration.svelte
@@ -0,0 +1,141 @@
+<script lang="ts">
+	import { LOAD_DEFAULTS, type ModelOverride } from '../utils';
+	import ModelsManagerModelConfigurationHeader from './ModelsManagerModelConfigurationHeader.svelte';
+	import ModelsManagerModelConfigurationInference from './ModelsManagerModelConfigurationInference.svelte';
+	import ModelsManagerModelConfigurationInformation from './ModelsManagerModelConfigurationInformation.svelte';
+	import ModelsManagerModelConfigurationLoad from './ModelsManagerModelConfigurationLoad.svelte';
+	import * as Tabs from '$lib/components/ui/tabs';
+	import { SETTINGS_KEYS } from '$lib/constants';
+	import { ServerModelStatus } from '$lib/enums';
+	import { HuggingFaceService } from '$lib/services';
+	import { modelsStore, settingsStore } from '$lib/stores';
+	import type { HfModelDetailInfo } from '$lib/types/huggingface';
+	import type { ModelOption } from '$lib/types/models';
+
+	interface Props {
+		isCustomized: boolean;
+		onClose: () => void;
+		onSave: (override: ModelOverride) => void;
+		onToggleLoad: () => void;
+		onUseInNewChat: () => void;
+		option: ModelOption;
+		override?: ModelOverride;
+	}
+
+	let { isCustomized, onClose, onSave, onToggleLoad, onUseInNewChat, option, override }: Props =
+		$props();
+
+	let tab = $state('information');
+	// edits in this pane, cleared on reset; the stored override is the baseline
+	let edits = $state<ModelOverride | null>(null);
+	let draft = $derived(edits ?? override ?? {});
+
+	// the props cache is a plain Map, so the read has to name its version to stay reactive
+	let serverProps = $derived.by(() => {
+		void modelsStore.props.cacheVersion;
+
+		return modelsStore.props.getModelProps(option.model);
+	});
+	let status = $derived.by(() => modelsStore.getModelStatus(option.model));
+	let isOperationInProgress = $derived(modelsStore.status.isOperationInProgress(option.model));
+	let isLoaded = $derived(
+		(status === ServerModelStatus.LOADED || status === ServerModelStatus.SLEEPING) &&
+			!isOperationInProgress
+	);
+	let loadProgress = $derived(
+		isOperationInProgress ? modelsStore.status.getLoadProgress(option.model) : null
+	);
+
+	// The server only reports the full metadata once a model is loaded, which loads
+	// it. When discovery is on, the Hub fills those gaps instead.
+	let hubDetails = $state<HfModelDetailInfo | null>(null);
+	// the window the model can take, from the listing or the Hub; a loaded model's runtime
+	// context only bounds it when nothing else says otherwise
+	let contextMax = $derived(
+		option.contextLength ??
+			hubDetails?.gguf?.context_length ??
+			serverProps?.default_generation_settings?.n_ctx ??
+			LOAD_DEFAULTS.contextLength
+	);
+
+	$effect(() => {
+		const repo = option.model.split(':')[0] ?? option.model;
+
+		let cancelled = false;
+
+		hubDetails = null;
+
+		if (!settingsStore.config[SETTINGS_KEYS.ENABLE_DISCOVER_MODELS] || !repo.includes('/')) {
+			return;
+		}
+
+		void HuggingFaceService.getDetails(repo)
+			.then((details) => {
+				if (!cancelled) hubDetails = details;
+			})
+			.catch(() => {});
+
+		return () => {
+			cancelled = true;
+		};
+	});
+
+	function resetDraft(): void {
+		edits = null;
+	}
+</script>
+
+<div class="flex h-full min-h-0 flex-col">
+	<ModelsManagerModelConfigurationHeader
+		{isCustomized}
+		{isLoaded}
+		{onClose}
+		{onToggleLoad}
+		{onUseInNewChat}
+		{option}
+		{status}
+	/>
+
+	<Tabs.Root class="mt-3 min-h-0 flex-1 gap-0" onValueChange={(value) => (tab = value)} value={tab}>
+		<div class="pl-4">
+			<Tabs.List class="w-full">
+				<Tabs.Trigger value="information">Information</Tabs.Trigger>
+
+				<Tabs.Trigger value="load">Load</Tabs.Trigger>
+
+				<Tabs.Trigger value="inference">Inference</Tabs.Trigger>
+			</Tabs.List>
+		</div>
+
+		<div class="min-h-0 flex-1 overflow-y-auto py-4 pl-4">
+			<Tabs.Content value="information">
+				<ModelsManagerModelConfigurationInformation
+					draftSetting={override?.load?.speculativeDecoding ?? null}
+					hub={hubDetails}
+					{option}
+					{serverProps}
+				/>
+			</Tabs.Content>
+
+			<Tabs.Content value="load">
+				<ModelsManagerModelConfigurationLoad
+					{contextMax}
+					{draft}
+					{loadProgress}
+					onChange={(next) => (edits = next)}
+					onReset={resetDraft}
+					onSave={() => onSave(draft)}
+				/>
+			</Tabs.Content>
+
+			<Tabs.Content value="inference">
+				<ModelsManagerModelConfigurationInference
+					{draft}
+					onChange={(next) => (edits = next)}
+					onReset={resetDraft}
+					onSave={() => onSave(draft)}
+				/>
+			</Tabs.Content>
+		</div>
+	</Tabs.Root>
+</div>
diff --git a/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationHeader.svelte b/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationHeader.svelte
new file mode 100644
index 0000000000..3212720802
--- /dev/null
+++ b/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationHeader.svelte
@@ -0,0 +1,172 @@
+<script lang="ts">
+	import { Eject, Loader2, Power, Server, SquarePen, X } from '@lucide/svelte';
+	import { Logo, ModelAvatar, ModelId } from '$lib/components/app';
+	import { BackendIcon } from '$lib/components/app/backends';
+	import { Badge } from '$lib/components/ui/badge';
+	import { Button } from '$lib/components/ui/button';
+	import { LOCAL_BACKEND_ID } from '$lib/constants';
+	import { ModelCapability, ServerModelStatus } from '$lib/enums';
+	import type { ModelOption } from '$lib/types/models';
+	import { getBackend } from '$lib/utils/api-base';
+	import { getBackendCapabilities } from '$lib/utils/backend';
+
+	interface Props {
+		isCustomized: boolean;
+		isLoaded: boolean;
+		onClose: () => void;
+		onToggleLoad: () => void;
+		onUseInNewChat: () => void;
+		option: ModelOption;
+		/** Load state reported by the server, null when it does not report one. */
+		status: ServerModelStatus | null;
+	}
+
+	let { isCustomized, isLoaded, onClose, onToggleLoad, onUseInNewChat, option, status }: Props =
+		$props();
+
+	let supportsToolUse = $derived(option.capabilities.includes(ModelCapability.TOOL_USE));
+	let supportsThinking = $derived(option.capabilities.includes(ModelCapability.REASONING));
+	let backend = $derived(getBackend(option.backendId));
+	let backendName = $derived(backend?.name ?? null);
+	let capabilities = $derived(getBackendCapabilities(backend));
+	// what the provider speaks decides which server endpoints the pane can read
+	let compatTitle = $derived(
+		capabilities.props
+			? 'llama.cpp server: reads /props and /slots'
+			: 'OpenAI-compatible: no /props, /slots, load or unload'
+	);
+
+	let isLoading = $derived(status === ServerModelStatus.LOADING);
+
+	let statusLabel = $derived.by(() => {
+		// a provider that cannot load or unload reports no load state, so its
+		// models are simply on offer
+		if (!capabilities.loadUnload) return 'Available';
+
+		if (status === ServerModelStatus.LOADING) return 'Loading';
+
+		if (status === ServerModelStatus.FAILED) return 'Failed to load';
+
+		if (status === ServerModelStatus.SLEEPING) return 'Sleeping';
+
+		if (isLoaded) return 'Loaded';
+
+		return 'Not loaded';
+	});
+
+	let statusDot = $derived.by(() => {
+		if (!capabilities.loadUnload) return 'bg-green-500';
+
+		if (status === ServerModelStatus.FAILED) return 'bg-red-500';
+
+		if (status === ServerModelStatus.LOADING) return 'bg-muted-foreground/50 animate-pulse';
+
+		if (status === ServerModelStatus.SLEEPING) return 'bg-orange-400';
+
+		if (isLoaded) return 'bg-green-500';
+
+		return 'bg-muted-foreground/50';
+	});
+</script>
+
+<header class="space-y-2.5 pt-3 pl-4">
+	<div class="flex items-start justify-between gap-3">
+		<div class="flex min-w-0 items-center gap-2">
+			<!-- same geometry as the discover details header: base org, quant org badge -->
+			<ModelAvatar
+				{option}
+				quantPositionClass="-bottom-1.5 -right-1.5"
+				quantSize="h-6 w-6"
+				showBaseModelAvatar
+				size="h-12 w-12"
+			/>
+
+			<div class="min-w-0">
+				<div class="flex items-center gap-2">
+					<ModelId
+						aliases={option.aliases}
+						class="min-w-0"
+						modalities={option.modalities}
+						modelId={option.model}
+						{supportsThinking}
+						{supportsToolUse}
+						tags={option.tags}
+						title={option.model}
+					/>
+
+					{#if isCustomized}
+						<Badge class="h-5 shrink-0 px-1.5 text-[10px]" variant="secondary">CUSTOMIZED</Badge>
+					{/if}
+				</div>
+
+				<!-- one line, one voice: state, protocol, owner -->
+				<p class="mt-1 flex items-center gap-x-2 text-xs text-muted-foreground">
+					{#if backend && backendName}
+						<span class="flex min-w-0 items-center gap-1" title={compatTitle}>
+							<BackendIcon {backend} class="size-4">
+								{#snippet fallback()}
+									{#if backend.id === LOCAL_BACKEND_ID}
+										<Logo class="shrink-0" style="--size: 0.75rem" />
+									{:else}
+										<Server class="size-4 shrink-0 opacity-70" />
+									{/if}
+								{/snippet}
+							</BackendIcon>
+
+							<!-- a long provider name shortens rather than growing the header -->
+							<span class="truncate">{backendName}</span>
+						</span>
+					{/if}
+
+					<span class="flex shrink-0 items-center gap-1">
+						<span class="h-2 w-2 shrink-0 rounded-full {statusDot}"></span>
+
+						{statusLabel}
+					</span>
+				</p>
+			</div>
+		</div>
+
+		<Button
+			aria-label="Close details"
+			class="h-7 w-7"
+			onclick={onClose}
+			size="icon"
+			variant="ghost"
+		>
+			<X class="h-4 w-4" />
+		</Button>
+	</div>
+
+	<div class="flex gap-2">
+		<Button class="flex-1 gap-1.5" onclick={onUseInNewChat} size="sm" variant="outline">
+			<SquarePen class="h-3.5 w-3.5" />
+
+			Start a new chat
+		</Button>
+
+		{#if capabilities.loadUnload}
+			<Button
+				class="flex-1 gap-1.5"
+				disabled={isLoading}
+				onclick={onToggleLoad}
+				size="sm"
+				variant="outline"
+			>
+				{#if isLoading}
+					<Loader2 class="h-3.5 w-3.5 animate-spin" />
+
+					Loading...
+				{:else if isLoaded}
+					<Eject class="h-3.5 w-3.5" />
+
+					Unload model
+				{:else}
+					<Power class="h-3.5 w-3.5" />
+
+					Load model
+				{/if}
+			</Button>
+		{/if}
+	</div>
+</header>
diff --git a/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationInference.svelte b/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationInference.svelte
new file mode 100644
index 0000000000..2445b68cdc
--- /dev/null
+++ b/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationInference.svelte
@@ -0,0 +1,281 @@
+<script lang="ts">
+	import { type ModelOverride, SAMPLING_DEFAULTS } from '../utils';
+	import { Braces, CircleDot, SlidersHorizontal, X } from '@lucide/svelte';
+	import { CollapsibleSection } from '$lib/components/app';
+	import { Button } from '$lib/components/ui/button';
+	import { Checkbox } from '$lib/components/ui/checkbox';
+	import { Input } from '$lib/components/ui/input';
+	import { Slider } from '$lib/components/ui/slider';
+	import { Switch } from '$lib/components/ui/switch';
+	import { Textarea } from '$lib/components/ui/textarea';
+
+	interface Props {
+		draft: ModelOverride;
+		onChange: (draft: ModelOverride) => void;
+		onReset: () => void;
+		onSave: () => void;
+	}
+
+	let { draft, onChange, onReset, onSave }: Props = $props();
+
+	let stopDraft = $state('');
+	let stopStrings = $derived(draft.stopStrings ?? []);
+
+	function patchSampling(patch: Partial<NonNullable<ModelOverride['sampling']>>): void {
+		onChange({ ...draft, sampling: { ...draft.sampling, ...patch } });
+	}
+
+	function addStopString(): void {
+		const value = stopDraft.trim();
+
+		if (!value) return;
+
+		onChange({ ...draft, stopStrings: [...stopStrings, value] });
+		stopDraft = '';
+	}
+
+	function removeStopString(value: string): void {
+		onChange({ ...draft, stopStrings: stopStrings.filter((s) => s !== value) });
+	}
+
+	const rowClass = 'flex items-center gap-3 py-2';
+	const sectionTrigger = 'flex w-full cursor-pointer items-center gap-2 py-2 text-left';
+</script>
+
+{#snippet numberInput(value: number, onInput: (value: number) => void)}
+	<Input
+		class="h-8 w-24 text-right text-sm"
+		min="0"
+		oninput={(event) => onInput(Number(event.currentTarget.value) || 0)}
+		type="number"
+		{value}
+	/>
+{/snippet}
+
+{#snippet samplingRow(
+	label: string,
+	value: number | null | undefined,
+	enabled: boolean,
+	onToggle: (enabled: boolean) => void,
+	onValue: (value: number) => void,
+	max: number,
+	step: number
+)}
+	<div class="space-y-1.5 py-2">
+		<div class="flex items-center gap-2">
+			<span class="text-sm">{label}</span>
+
+			<Checkbox checked={enabled} onCheckedChange={(checked) => onToggle(checked === true)} />
+
+			<span class="ml-auto">
+				{@render numberInput(value ?? 0, onValue)}
+			</span>
+		</div>
+
+		<Slider
+			disabled={!enabled}
+			{max}
+			onValueChange={(next: number) => onValue(next)}
+			{step}
+			type="single"
+			value={value ?? 0}
+		/>
+	</div>
+{/snippet}
+
+<p class="pb-2 text-sm font-medium text-muted-foreground">System prompt</p>
+
+<Textarea
+	class="min-h-24 text-sm"
+	oninput={(event) => onChange({ ...draft, systemPrompt: event.currentTarget.value })}
+	placeholder="Example, &quot;Only answer in rhymes&quot;"
+	value={draft.systemPrompt ?? ''}
+/>
+
+<p class="pt-1 text-right text-xs text-muted-foreground">Token count: N/A</p>
+
+<div class="mt-4">
+	<CollapsibleSection triggerClass={sectionTrigger}>
+		{#snippet trigger()}
+			<SlidersHorizontal class="h-3.5 w-3.5 text-muted-foreground" />
+
+			<span class="text-sm font-medium">Sampling</span>
+		{/snippet}
+
+		<div class="pt-1">
+			{@render samplingRow(
+				'Temperature',
+				draft.sampling?.temperature,
+				draft.sampling?.temperature !== null && draft.sampling?.temperature !== undefined,
+				(enabled) => patchSampling({ temperature: enabled ? SAMPLING_DEFAULTS.temperature : null }),
+				(value) => patchSampling({ temperature: value }),
+				2,
+				0.05
+			)}
+
+			{@render samplingRow(
+				'Top K',
+				draft.sampling?.topK,
+				draft.sampling?.topK !== null && draft.sampling?.topK !== undefined,
+				(enabled) => patchSampling({ topK: enabled ? SAMPLING_DEFAULTS.topK : null }),
+				(value) => patchSampling({ topK: value }),
+				200,
+				1
+			)}
+
+			{@render samplingRow(
+				'Top P',
+				draft.sampling?.topP,
+				draft.sampling?.topP !== null && draft.sampling?.topP !== undefined,
+				(enabled) => patchSampling({ topP: enabled ? SAMPLING_DEFAULTS.topP : null }),
+				(value) => patchSampling({ topP: value }),
+				1,
+				0.01
+			)}
+
+			{@render samplingRow(
+				'Min P',
+				draft.sampling?.minP,
+				draft.sampling?.minP !== null && draft.sampling?.minP !== undefined,
+				(enabled) => patchSampling({ minP: enabled ? SAMPLING_DEFAULTS.minP : null }),
+				(value) => patchSampling({ minP: value }),
+				1,
+				0.01
+			)}
+
+			{@render samplingRow(
+				'Repeat Penalty',
+				draft.sampling?.repeatPenalty,
+				draft.sampling?.repeatPenalty !== null && draft.sampling?.repeatPenalty !== undefined,
+				(enabled) =>
+					patchSampling({ repeatPenalty: enabled ? SAMPLING_DEFAULTS.repeatPenalty : null }),
+				(value) => patchSampling({ repeatPenalty: value }),
+				2,
+				0.01
+			)}
+		</div>
+	</CollapsibleSection>
+
+	<CollapsibleSection triggerClass={sectionTrigger}>
+		{#snippet trigger()}
+			<span class="text-sm font-medium">Stop strings</span>
+		{/snippet}
+
+		<div class="space-y-2 pt-1">
+			{#if stopStrings.length > 0}
+				<div class="flex flex-wrap gap-1.5">
+					{#each stopStrings as value (value)}
+						<button
+							class="inline-flex cursor-pointer items-center gap-1 rounded-full border border-border px-2 py-0.5 text-xs"
+							onclick={() => removeStopString(value)}
+							type="button"
+						>
+							{value}
+
+							<X class="h-3 w-3" />
+						</button>
+					{/each}
+				</div>
+			{/if}
+
+			<Input
+				bind:value={stopDraft}
+				class="h-9 text-sm"
+				onkeydown={(event) => {
+					if (event.key === 'Enter') {
+						event.preventDefault();
+						addStopString();
+					}
+				}}
+				placeholder="Enter a string and press Enter"
+			/>
+		</div>
+	</CollapsibleSection>
+
+	<CollapsibleSection triggerClass={sectionTrigger}>
+		{#snippet trigger()}
+			<Braces class="h-3.5 w-3.5 text-muted-foreground" />
+
+			<span class="text-sm font-medium">Structured output</span>
+		{/snippet}
+
+		<div class="space-y-2 pt-1">
+			<div class={rowClass}>
+				<span class="text-sm">Enabled</span>
+
+				<Switch
+					checked={draft.structuredOutput?.enabled ?? false}
+					class="ml-auto"
+					onCheckedChange={(checked) =>
+						onChange({
+							...draft,
+							structuredOutput: {
+								enabled: checked === true,
+								schema: draft.structuredOutput?.schema ?? ''
+							}
+						})}
+				/>
+			</div>
+
+			<Textarea
+				class="min-h-20 font-mono text-xs"
+				disabled={!draft.structuredOutput?.enabled}
+				oninput={(event) =>
+					onChange({
+						...draft,
+						structuredOutput: {
+							enabled: draft.structuredOutput?.enabled ?? false,
+							schema: event.currentTarget.value
+						}
+					})}
+				placeholder={'{ }'}
+				value={draft.structuredOutput?.schema ?? ''}
+			/>
+		</div>
+	</CollapsibleSection>
+
+	<CollapsibleSection triggerClass={sectionTrigger}>
+		{#snippet trigger()}
+			<CircleDot class="h-3.5 w-3.5 text-muted-foreground" />
+
+			<span class="text-sm font-medium">Reasoning</span>
+		{/snippet}
+
+		<div class="space-y-2 pt-1">
+			<div class={rowClass}>
+				<span class="text-sm">Enable Thinking</span>
+
+				<Switch
+					checked={draft.reasoning?.enabled ?? false}
+					class="ml-auto"
+					onCheckedChange={(checked) =>
+						onChange({
+							...draft,
+							reasoning: {
+								budget: draft.reasoning?.budget ?? 'Unrestricted',
+								enabled: checked === true
+							}
+						})}
+				/>
+			</div>
+
+			<p class="text-xs text-muted-foreground">
+				Controls whether the model will think before replying
+			</p>
+
+			<div class={rowClass}>
+				<span class="text-sm">Reasoning Budget</span>
+
+				<span class="ml-auto text-sm text-muted-foreground">
+					{draft.reasoning?.budget ?? 'Unrestricted'}
+				</span>
+			</div>
+		</div>
+	</CollapsibleSection>
+</div>
+
+<div class="flex justify-end gap-2 pt-4">
+	<Button onclick={onReset} variant="ghost">Reset</Button>
+
+	<Button onclick={onSave}>Save</Button>
+</div>
diff --git a/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationInformation.svelte b/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationInformation.svelte
new file mode 100644
index 0000000000..c283acbb93
--- /dev/null
+++ b/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationInformation.svelte
@@ -0,0 +1,245 @@
+<script lang="ts">
+	import { modelDraftsFor, modelQuantLabel, modelSizeLabel, resolveModelSize } from '../utils';
+	import {
+		ActionIconCopyToClipboard,
+		BadgesModality,
+		CollapsibleSection
+	} from '$lib/components/app';
+	import { Badge } from '$lib/components/ui/badge';
+	import { HuggingFaceService } from '$lib/services';
+	import { modelsStore } from '$lib/stores';
+	import type { ApiLlamaCppServerProps } from '$lib/types/api';
+	import type { HfModelDetailInfo } from '$lib/types/huggingface';
+	import type { ModelOption } from '$lib/types/models';
+	import { getBackend } from '$lib/utils/api-base';
+	import { getBackendCapabilities } from '$lib/utils/backend';
+	import { formatFileSize, formatNumber, formatParameters } from '$lib/utils/formatters';
+
+	interface Props {
+		/** Draft the load settings name, when one is set. */
+		draftSetting?: string | null;
+		/** Hugging Face details, filled in when discovery is on and the server has none. */
+		hub?: HfModelDetailInfo | null;
+		option: ModelOption;
+		serverProps: ApiLlamaCppServerProps | null | undefined;
+	}
+
+	let { draftSetting = null, hub = null, option, serverProps }: Props = $props();
+
+	let quant = $derived(modelQuantLabel(option));
+	let size = $derived(modelSizeLabel(option));
+	// a GGUF carries the count in its metadata, a transformers repo in its SafeTensors index
+	let hubParameters = $derived(HuggingFaceService.parameterCount(hub));
+	let resolvedSize = $state<string | null>(null);
+
+	// the router rarely reports a size, the repo tree does
+	$effect(() => {
+		const target = option;
+
+		let cancelled = false;
+
+		resolvedSize = null;
+
+		void resolveModelSize(target)
+			.then((value) => {
+				if (!cancelled) resolvedSize = value;
+			})
+			.catch(() => {});
+
+		return () => {
+			cancelled = true;
+		};
+	});
+
+	// the window a model can take, from the listing or from the Hub
+	let contextLabel = $derived(
+		option.contextLength
+			? `${formatParameters(option.contextLength)} tokens`
+			: hub?.gguf?.context_length
+				? `${formatNumber(hub.gguf.context_length)} tokens`
+				: null
+	);
+	// the draft a load would speculate with, and any other sidecar on disk
+	let drafts = $derived(modelDraftsFor(option, draftSetting));
+	let activeDraft = $derived(drafts.find((draft) => draft.active) ?? null);
+	let idleDrafts = $derived(drafts.filter((draft) => !draft.active));
+	let meta = $derived(option.meta);
+	// a plain OpenAI-compatible endpoint has no /props or /slots to read
+	let reportsServerInfo = $derived(getBackendCapabilities(getBackend(option.backendId)).props);
+	// the server reports these once the model is loaded; the Hub knows them anyway
+	let gguf = $derived(hub?.gguf ?? null);
+	let modalities = $derived.by(() => {
+		void modelsStore.props.cacheVersion;
+
+		return modelsStore.props.getModelModalitiesArray(option.id);
+	});
+	let rows = $derived([
+		...(reportsServerInfo
+			? [
+					{
+						isCopyable: true,
+						isMono: true,
+						label: 'File Path',
+						value: serverProps?.model_path ?? null
+					}
+				]
+			: []),
+		{
+			label: 'Context Size',
+			// the server reports the context it runs with once the model is loaded; until then
+			// the listing, or the Hub, still says what the model can take
+			value: serverProps
+				? `${formatNumber(serverProps.default_generation_settings?.n_ctx ?? 0)} tokens`
+				: (contextLabel ?? null)
+		},
+		{
+			label: 'Model Size',
+			value: meta?.size ? formatFileSize(meta.size) : (resolvedSize ?? size)
+		},
+		{
+			label: 'Parameters',
+			value: meta?.n_params
+				? formatParameters(meta.n_params)
+				: (option.parsedId?.params ?? (hubParameters ? formatParameters(hubParameters) : null))
+		},
+		{ label: 'Embedding Size', value: meta?.n_embd ? formatNumber(meta.n_embd) : null },
+		{
+			label: 'Vocabulary Size',
+			value: meta?.n_vocab ? `${formatNumber(meta.n_vocab)} tokens` : null
+		},
+		{ isBadge: true, label: 'Quantization', value: quant },
+		{
+			isBadge: true,
+			label: 'Architecture',
+			value: (option.meta?.architecture as string) ?? gguf?.architecture ?? null
+		},
+		...(activeDraft
+			? [
+					{ isBadge: true, label: 'Draft sidecar', value: activeDraft.kind },
+					{
+						isCopyable: true,
+						label: 'Draft model',
+						value: activeDraft.model ?? option.model.split(':')[0]
+					},
+					{ isBadge: true, label: 'Draft quant', value: activeDraft.quant }
+				]
+			: []),
+		{
+			isBadge: true,
+			label: 'Other sidecars',
+			value: idleDrafts.length > 0 ? idleDrafts.map((draft) => draft.kind).join(', ') : null
+		},
+		...(reportsServerInfo
+			? [
+					{
+						label: 'Parallel Slots',
+						value: serverProps?.total_slots != null ? String(serverProps.total_slots) : null
+					},
+					{ isMono: true, label: 'Build Info', value: serverProps?.build_info ?? null }
+				]
+			: [])
+	] satisfies Array<{
+		isBadge?: boolean;
+		isCopyable?: boolean;
+		isMono?: boolean;
+		label: string;
+		value: string | null;
+	}>);
+
+	// a provider that reports little gets a short list: rows without a value say nothing,
+	// and a llama-compat server fills the rest in from /props once the model is loaded
+	let visibleRows = $derived(rows.filter((row) => row.value !== null));
+
+	const sectionTrigger = 'flex w-full cursor-pointer items-center gap-2 py-2 text-left';
+</script>
+
+<div class="space-y-0">
+	<div class="flex items-center gap-3 border-b border-border/30 py-2.5">
+		<span class="text-sm text-muted-foreground">Model</span>
+
+		<span class="ml-auto flex min-w-0 flex-1 items-center gap-2">
+			<span class="min-w-0 flex-1 overflow-x-auto font-mono text-xs whitespace-nowrap">
+				{option.model}
+			</span>
+
+			<ActionIconCopyToClipboard
+				ariaLabel="Copy model name to clipboard"
+				canCopy
+				text={option.model}
+			/>
+		</span>
+	</div>
+
+	{#each visibleRows as row (row.label)}
+		<div class="flex items-center gap-3 border-b border-border/30 py-2.5 last:border-b-0">
+			<span class="text-sm text-muted-foreground">{row.label}</span>
+
+			<span class="ml-auto flex min-w-0 items-center gap-2 {row.isCopyable ? 'flex-1' : ''}">
+				{#if row.value === null}
+					<span class="text-muted-foreground">—</span>
+				{:else}
+					<span
+						class={[
+							'min-w-0',
+							// a copyable value scrolls instead of clipping, the icon stays put
+							row.isCopyable ? 'flex-1 overflow-x-auto whitespace-nowrap' : 'truncate',
+							row.isBadge ? '' : 'text-sm',
+							row.isMono ? 'font-mono text-xs' : ''
+						]}
+					>
+						{#if row.isBadge}
+							<Badge class="h-5 px-1.5 text-[10px]" variant="secondary">{row.value}</Badge>
+						{:else}
+							{row.value}
+						{/if}
+					</span>
+
+					{#if row.isCopyable}
+						<ActionIconCopyToClipboard
+							ariaLabel="Copy value to clipboard"
+							canCopy
+							text={row.value}
+						/>
+					{/if}
+				{/if}
+			</span>
+		</div>
+	{/each}
+
+	{#if visibleRows.length === 0}
+		<p class="py-2.5 text-sm text-muted-foreground">This provider reports no model metadata.</p>
+	{/if}
+
+	{#if modalities.length > 0}
+		<div class="flex items-center gap-3 border-b border-border/30 py-2.5 last:border-b-0">
+			<span class="text-sm text-muted-foreground">Modalities</span>
+
+			<span class="ml-auto flex flex-wrap gap-1"><BadgesModality {modalities} /></span>
+		</div>
+	{/if}
+</div>
+
+{#if !reportsServerInfo}
+	<p class="pt-2 text-xs text-muted-foreground">
+		OpenAI-compatible provider: there is no /props or /slots, so only what the listing and the
+		model's repo report is shown.
+	</p>
+{:else if !serverProps}
+	<p class="pt-2 text-xs text-muted-foreground">
+		/props values - file path, slots, build info - appear once the model is loaded. Reading this
+		page never loads a model.
+	</p>
+{/if}
+
+<div class="mt-4">
+	<CollapsibleSection triggerClass={sectionTrigger}>
+		{#snippet trigger()}
+			<span class="text-sm font-medium">Chat Template</span>
+		{/snippet}
+
+		<pre
+			class="mt-1 max-h-48 overflow-auto rounded-md bg-muted/50 p-2 text-xs whitespace-pre-wrap">{serverProps?.chat_template ??
+				gguf?.chat_template ??
+				'Not reported by the server.'}</pre>
+	</CollapsibleSection>
+</div>
diff --git a/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationLoad.svelte b/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationLoad.svelte
new file mode 100644
index 0000000000..b6f4ebf844
--- /dev/null
+++ b/tools/ui/src/lib/components/app/models/ModelsManager/ModelsManagerModelConfiguration/ModelsManagerModelConfigurationLoad.svelte
@@ -0,0 +1,195 @@
+<script lang="ts">
+	import {
+		LOAD_DEFAULTS,
+		type ModelLoadProgress,
+		type ModelOverride,
+		SPECULATIVE_OPTIONS
+	} from '../utils';
+	import { Info, ListOrdered } from '@lucide/svelte';
+	import { CollapsibleSection } from '$lib/components/app';
+	import { Badge } from '$lib/components/ui/badge';
+	import { Button } from '$lib/components/ui/button';
+	import { Input } from '$lib/components/ui/input';
+	import { Slider } from '$lib/components/ui/slider';
+	import { Switch } from '$lib/components/ui/switch';
+	import { formatParameters } from '$lib/utils/formatters';
+
+	interface Props {
+		contextMax: number;
+		draft: ModelOverride;
+		loadProgress: ModelLoadProgress | null;
+		onChange: (draft: ModelOverride) => void;
+		onReset: () => void;
+		onSave: () => void;
+	}
+
+	let { contextMax, draft, loadProgress, onChange, onReset, onSave }: Props = $props();
+
+	let load = $derived({ ...LOAD_DEFAULTS, ...draft.load });
+
+	function patchLoad(patch: Partial<typeof load>): void {
+		onChange({ ...draft, load: { ...draft.load, ...patch } });
+	}
+
+	const rowClass = 'flex items-center gap-3 py-2';
+	const sectionTrigger = 'flex w-full cursor-pointer items-center gap-2 py-2 text-left';
+</script>
+
+{#snippet numberInput(value: number, onInput: (value: number) => void)}
+	<Input
+		class="h-8 w-24 text-right text-sm"
+		min="0"
+		oninput={(event) => onInput(Number(event.currentTarget.value) || 0)}
+		type="number"
+		{value}
+	/>
+{/snippet}
+
+<p class="pb-2 text-sm font-medium text-muted-foreground">Context and offload</p>
+
+<div class="space-y-4">
+	<div class="space-y-1.5">
+		<div class="flex items-center gap-2">
+			<span class="text-sm">Context Length</span>
+
+			<span class="ml-auto">
+				{@render numberInput(load.contextLength, (value) => patchLoad({ contextLength: value }))}
+			</span>
+		</div>
+
+		<p class="text-xs text-muted-foreground">
+			Model supports up to
+			<Badge class="h-5 px-1.5 text-[10px]" variant="secondary">
+				{formatParameters(contextMax)}
+			</Badge>
+			tokens
+		</p>
+
+		<Slider
+			max={contextMax}
+			onValueChange={(next: number) => patchLoad({ contextLength: next })}
+			step={512}
+			type="single"
+			value={load.contextLength}
+		/>
+	</div>
+
+	<div class="space-y-1.5">
+		<div class="flex items-center gap-2">
+			<span class="text-sm">
+				GPU Offload <span class="text-muted-foreground">(layers)</span>
+			</span>
+
+			<span class="ml-auto">
+				{@render numberInput(load.gpuOffload, (value) => patchLoad({ gpuOffload: value }))}
+			</span>
+		</div>
+
+		<Slider
+			max={200}
+			onValueChange={(next: number) => patchLoad({ gpuOffload: next })}
+			step={1}
+			type="single"
+			value={load.gpuOffload}
+		/>
+	</div>
+</div>
+
+<div class="mt-4">
+	<CollapsibleSection triggerClass={sectionTrigger}>
+		{#snippet trigger()}
+			<ListOrdered class="h-3.5 w-3.5 text-muted-foreground" />
+
+			<span class="text-sm font-medium">Advanced load params</span>
+		{/snippet}
+
+		<div class="pt-1">
+			<div class={rowClass}>
+				<span class="text-sm">CPU Thread Pool Size</span>
+
+				<span class="ml-auto">
+					{@render numberInput(load.cpuThreads, (value) => patchLoad({ cpuThreads: value }))}
+				</span>
+			</div>
+
+			<div class={rowClass}>
+				<span class="text-sm">Evaluation Batch Size</span>
+
+				<span class="ml-auto">
+					{@render numberInput(load.batchSize, (value) => patchLoad({ batchSize: value }))}
+				</span>
+			</div>
+
+			<div class={rowClass}>
+				<span class="text-sm">Physical Batch Size</span>
+
+				<span class="ml-auto">
+					{@render numberInput(load.ubatchSize, (value) => patchLoad({ ubatchSize: value }))}
+				</span>
+			</div>
+
+			<div class={rowClass}>
+				<span class="text-sm">Flash Attention</span>
+
+				<Switch
+					checked={load.flashAttention}
+					class="ml-auto"
+					onCheckedChange={(checked) => patchLoad({ flashAttention: checked === true })}
+				/>
+			</div>
+
+			<div class={rowClass}>
+				<span class="text-sm">Keep Model in Memory</span>
+
+				<Switch
+					checked={load.keepInMemory}
+					class="ml-auto"
+					onCheckedChange={(checked) => patchLoad({ keepInMemory: checked === true })}
+				/>
+			</div>
+
+			<div class={rowClass}>
+				<span class="text-sm">Try mmap()</span>
+
+				<Switch
+					checked={load.useMmap}
+					class="ml-auto"
+					onCheckedChange={(checked) => patchLoad({ useMmap: checked === true })}
+				/>
+			</div>
+
+			<div class={rowClass}>
+				<span class="text-sm">Speculative Decoding</span>
+
+				<select
+					class="ml-auto h-8 cursor-pointer rounded-md border border-input bg-transparent px-2 text-sm capitalize"
+					onchange={(event) => patchLoad({ speculativeDecoding: event.currentTarget.value })}
+					value={load.speculativeDecoding}
+				>
+					{#each SPECULATIVE_OPTIONS as value (value)}
+						<option {value}>{value.replace('-', ' ')}</option>
+					{/each}
+				</select>
+			</div>
+		</div>
+	</CollapsibleSection>
+</div>
+
+<div class="mt-4 flex items-start gap-2 rounded-lg bg-muted/40 p-3 text-xs text-muted-foreground">
+	<Info class="mt-0.5 h-3.5 w-3.5 shrink-0" />
+
+	<p>Changes apply on next load - eject and re-load to pick them up.</p>
+</div>
+
+{#if loadProgress}
+	<p class="pt-2 text-xs text-muted-foreground">
+		Loading: {loadProgress.current}
+		{Math.round(loadProgress.value * 100)}%
+	</p>
+{/if}
+
+<div class="flex justify-end gap-2 pt-4">
+	<Button onclick={onReset} variant="ghost">Reset</Button>
+
+	<Button onclick={onSave}>Save</Button>
+</div>
diff --git a/tools/ui/src/lib/components/ui/slider/index.ts b/tools/ui/src/lib/components/ui/slider/index.ts
new file mode 100644
index 0000000000..f1524fc4f4
--- /dev/null
+++ b/tools/ui/src/lib/components/ui/slider/index.ts
@@ -0,0 +1,7 @@
+import Root from './slider.svelte';
+
+export {
+	Root,
+	//
+	Root as Slider
+};
diff --git a/tools/ui/src/lib/components/ui/slider/slider.svelte b/tools/ui/src/lib/components/ui/slider/slider.svelte
new file mode 100644
index 0000000000..7b2a978c60
--- /dev/null
+++ b/tools/ui/src/lib/components/ui/slider/slider.svelte
@@ -0,0 +1,53 @@
+<script lang="ts">
+	import { cn, type WithoutChildrenOrChild } from '$lib/components/ui/utils';
+	import { Slider as SliderPrimitive } from 'bits-ui';
+
+	let {
+		class: className,
+		orientation = 'horizontal',
+		ref = $bindable(null),
+		value = $bindable(),
+		...restProps
+	}: WithoutChildrenOrChild<SliderPrimitive.RootProps> = $props();
+</script>
+
+<!--
+Discriminated Unions + Destructing (required for bindable) do not
+get along, so we shut typescript up by casting `value` to `never`.
+-->
+<SliderPrimitive.Root
+	bind:ref
+	bind:value={value as never}
+	class={cn(
+		'data-[orientation=vertical]:min-h-40 relative flex w-full touch-none items-center select-none data-disabled:opacity-50 data-[orientation=vertical]:h-full data-[orientation=vertical]:w-auto data-[orientation=vertical]:flex-col',
+		className
+	)}
+	data-slot="slider"
+	{orientation}
+	{...restProps}
+>
+	{#snippet children({ thumbItems })}
+		<span
+			class={cn(
+				'bg-muted rounded-full data-[orientation=horizontal]:h-1 data-[orientation=horizontal]:w-full data-[orientation=vertical]:h-full data-[orientation=vertical]:w-1 relative grow overflow-hidden data-[orientation=horizontal]:w-full data-[orientation=vertical]:h-full'
+			)}
+			data-orientation={orientation}
+			data-slot="slider-track"
+		>
+			<SliderPrimitive.Range
+				class={cn(
+					'bg-primary absolute select-none data-[orientation=horizontal]:h-full data-[orientation=vertical]:w-full'
+				)}
+				data-slot="slider-range"
+			/>
+		</span>
+
+		{#each thumbItems as thumb (thumb.index)}
+			<SliderPrimitive.Thumb
+				class="border-ring ring-ring/50 relative size-3 rounded-full border bg-white transition-[color,box-shadow] after:absolute after:-inset-2 hover:ring-3 focus-visible:ring-3 focus-visible:outline-hidden active:ring-3 block shrink-0 select-none disabled:pointer-events-none disabled:opacity-50"
+				data-slot="slider-thumb"
+				index={thumb.index}
+			/>
+		{/each}
+	{/snippet}
+</SliderPrimitive.Root>