feat(agent): let users choose model route

drizzle/0026_assistant_model_route.sql

@@ -0,0 +1,5 @@
+ALTER TABLE "users" ADD COLUMN "assistant_model_route" text;--> statement-breakpoint
+UPDATE "users"
+SET "assistant_model_route" = "assistant_model",
+    "assistant_model" = NULL
+WHERE "assistant_model" IN ('auto', 'uncensored');

drizzle/meta/_journal.json

@@ -183,6 +183,13 @@
       "when": 1783560000000,
       "tag": "0025_assistant_model",
       "breakpoints": true
+    },
+    {
+      "idx": 26,
+      "version": "7",
+      "when": 1783561000000,
+      "tag": "0026_assistant_model_route",
+      "breakpoints": true
     }
   ]
-}
+}

src/app/api/agent/chat/route.ts

@@ -32,7 +32,7 @@ export async function POST(request: Request) {
     return apiError(parsed.error.issues[0]?.message ?? "Validation error", 400);
   }
 
-  const modelList = await listLlmModels();
+  const modelList = await listLlmModels({ route: assistant.modelRoute });
   const modelResolution = resolveAssistantModel({
     requestedModel: parsed.data.model,
     savedModel: assistant.model,

src/app/api/agent/models/route.ts

@@ -1,7 +1,7 @@
 import { apiError, apiJson } from "@/lib/api-handler";
 import { resolveApiAuth } from "@/lib/api-auth";
 import { getAssistantPreferences } from "@/lib/assistant-preference";
-import { listLlmModels, resolveAssistantModel } from "@/lib/llm/models";
+import { isValidAssistantModelRoute, listLlmModels, resolveAssistantModel } from "@/lib/llm/models";
 
 export const dynamic = "force-dynamic";
 
@@ -21,10 +21,16 @@ export async function GET(request: Request) {
     return noStore(apiError("Assistant not enabled", 403));
   }
 
-  const modelList = await listLlmModels();
+  const requestedRoute = new URL(request.url).searchParams.get("route");
+  if (requestedRoute !== null && !isValidAssistantModelRoute(requestedRoute)) {
+    return noStore(apiError("Invalid assistant model route", 400));
+  }
+
+  const modelRoute = requestedRoute ?? assistant.modelRoute;
+  const modelList = await listLlmModels({ route: modelRoute });
   const resolved = resolveAssistantModel({
     requestedModel: null,
-    savedModel: assistant.model,
+    savedModel: requestedRoute === null ? assistant.model : null,
     fallbackModel: modelList.fallbackModel,
     models: modelList.models,
   });
@@ -34,6 +40,7 @@ export async function GET(request: Request) {
       models: modelList.models,
       selectedModel: resolved.model,
       fallbackModel: modelList.fallbackModel,
+      route: modelList.route,
       degraded: modelList.degraded,
     }),
   );

src/app/layout.tsx

@@ -104,6 +104,7 @@ export default async function RootLayout({ children }: { children: React.ReactNo
   let signedIn = false;
   let assistantEnabled = false;
   let assistantName = DEFAULT_ASSISTANT_NAME;
+  let assistantModelRoute: string | null = null;
   let assistantModel: string | null = null;
 
   const session = await auth();
@@ -118,6 +119,7 @@ export default async function RootLayout({ children }: { children: React.ReactNo
         themeNavStyle: users.themeNavStyle,
         assistantEnabled: users.assistantEnabled,
         assistantName: users.assistantName,
+        assistantModelRoute: users.assistantModelRoute,
         assistantModel: users.assistantModel,
       })
       .from(users)
@@ -131,6 +133,7 @@ export default async function RootLayout({ children }: { children: React.ReactNo
       navStyle = row.themeNavStyle as NavStyle;
       assistantEnabled = row.assistantEnabled;
       assistantName = row.assistantName?.trim() || DEFAULT_ASSISTANT_NAME;
+      assistantModelRoute = row.assistantModelRoute?.trim() || null;
       assistantModel = row.assistantModel?.trim() || null;
     }
     userDashboards = await db
@@ -189,6 +192,7 @@ export default async function RootLayout({ children }: { children: React.ReactNo
               configured={isLlmConfigured()}
               userId={session.user.id}
               assistantName={assistantName}
+              assistantModelRoute={assistantModelRoute}
               assistantModel={assistantModel}
             />
           ) : null}

src/app/settings/assistant-actions.ts

@@ -9,9 +9,15 @@ import {
   MAX_ASSISTANT_NAME_LENGTH,
   MAX_ASSISTANT_SYSTEM_PROMPT_LENGTH,
 } from "@/lib/assistant-config";
-import { isValidLlmModelId, listLlmModels } from "@/lib/llm/models";
+import {
+  isValidAssistantModelRoute,
+  isValidLlmModelId,
+  listLlmModels,
+  type AssistantModelRoute,
+} from "@/lib/llm/models";
 import { users } from "@/modules/_core/schema";
 import { getCurrentSession } from "@/lib/session";
+import { getAssistantPreferences } from "@/lib/assistant-preference";
 
 const assistantNameSchema = z
   .string()
@@ -51,6 +57,19 @@ export async function setAssistantSystemPrompt(prompt: string | null): Promise<v
   revalidateAssistantSurfaces();
 }
 
+export async function setAssistantModelRoute(route: AssistantModelRoute): Promise<void> {
+  if (!isValidAssistantModelRoute(route)) {
+    throw new Error("Invalid assistant model route");
+  }
+
+  const { user } = await getCurrentSession();
+  await db
+    .update(users)
+    .set({ assistantModelRoute: route, assistantModel: null })
+    .where(eq(users.id, user.id));
+  revalidateAssistantSurfaces();
+}
+
 export async function setAssistantModel(model: string | null): Promise<void> {
   const { user } = await getCurrentSession();
   const normalized = model?.trim() || null;
@@ -59,7 +78,8 @@ export async function setAssistantModel(model: string | null): Promise<void> {
     throw new Error("Invalid assistant model");
   }
 
-  const available = await listLlmModels();
+  const assistant = await getAssistantPreferences(user.id);
+  const available = await listLlmModels({ route: assistant.modelRoute });
   const requested = normalized === available.fallbackModel ? null : normalized;
 
   if (requested !== null && !available.models.some((option) => option.id === requested)) {

src/lib/assistant-preference.ts

@@ -1,6 +1,7 @@
 import { eq } from "drizzle-orm";
 import { db } from "@/lib/db";
 import { DEFAULT_ASSISTANT_NAME } from "@/lib/assistant-config";
+import { isValidAssistantModelRoute, type AssistantModelRoute } from "@/lib/llm/models";
 import { users } from "@/modules/_core/schema";
 
 export {
@@ -14,6 +15,7 @@ export type AssistantPreferences = {
   enabled: boolean;
   name: string;
   systemPrompt: string | null;
+  modelRoute: AssistantModelRoute | null;
   model: string | null;
 };
 
@@ -23,6 +25,7 @@ export async function getAssistantPreferences(userId: string): Promise<Assistant
       assistantEnabled: users.assistantEnabled,
       assistantName: users.assistantName,
       assistantSystemPrompt: users.assistantSystemPrompt,
+      assistantModelRoute: users.assistantModelRoute,
       assistantModel: users.assistantModel,
     })
     .from(users)
@@ -33,6 +36,10 @@ export async function getAssistantPreferences(userId: string): Promise<Assistant
     enabled: row?.assistantEnabled ?? false,
     name: row?.assistantName?.trim() || DEFAULT_ASSISTANT_NAME,
     systemPrompt: row?.assistantSystemPrompt ?? null,
+    modelRoute:
+      row?.assistantModelRoute && isValidAssistantModelRoute(row.assistantModelRoute)
+        ? row.assistantModelRoute
+        : null,
     model: row?.assistantModel?.trim() || null,
   };
 }

src/lib/llm/models.ts

@@ -5,9 +5,14 @@ export type LlmModelOption = {
   label: string;
 };
 
+export const ASSISTANT_MODEL_ROUTES = ["auto", "uncensored"] as const;
+
+export type AssistantModelRoute = (typeof ASSISTANT_MODEL_ROUTES)[number];
+
 export type LlmModelsResult = {
   models: LlmModelOption[];
   fallbackModel: string;
+  route: AssistantModelRoute;
   degraded: boolean;
 };
 
@@ -28,6 +33,22 @@ export function isValidLlmModelId(value: string): boolean {
   );
 }
 
+export function isValidAssistantModelRoute(value: string): value is AssistantModelRoute {
+  return ASSISTANT_MODEL_ROUTES.some((route) => route === value);
+}
+
+function resolveModelRoute(
+  route: AssistantModelRoute | null | undefined,
+  config: LlmConfig,
+): { route: AssistantModelRoute; fallbackModel: string } {
+  if (route) {
+    return { route, fallbackModel: route };
+  }
+
+  const defaultRoute: AssistantModelRoute = config.model === "uncensored" ? "uncensored" : "auto";
+  return { route: defaultRoute, fallbackModel: config.model };
+}
+
 export function normalizeLlmModelsPayload(payload: unknown): LlmModelOption[] {
   const data =
     typeof payload === "object" && payload !== null && "data" in payload
@@ -51,14 +72,15 @@ export function normalizeLlmModelsPayload(payload: unknown): LlmModelOption[] {
 export async function listLlmModels(options?: {
   config?: LlmConfig;
   fetchImpl?: typeof fetch;
+  route?: AssistantModelRoute | null;
 }): Promise<LlmModelsResult> {
   const config = options?.config ?? getLlmConfig();
   const fetchImpl = options?.fetchImpl ?? fetch;
-  const fallbackModel = config.model;
+  const { route, fallbackModel } = resolveModelRoute(options?.route, config);
   const fallbackOption = { id: fallbackModel, label: fallbackModel };
 
   if (config.provider === "mock" || !config.baseUrl) {
-    return { models: [fallbackOption], fallbackModel, degraded: false };
+    return { models: [fallbackOption], fallbackModel, route, degraded: false };
   }
 
   try {
@@ -66,7 +88,7 @@ export async function listLlmModels(options?: {
     if (config.apiKey) headers.Authorization = `Bearer ${config.apiKey}`;
 
     const modelsUrl = new URL(`${config.baseUrl.replace(/\/$/, "")}/models`);
-    if (fallbackModel === "uncensored") {
+    if (route === "uncensored") {
       modelsUrl.searchParams.set("type", "uncensored");
     }
 
@@ -76,26 +98,27 @@ export async function listLlmModels(options?: {
     });
 
     if (!response.ok) {
-      return { models: [fallbackOption], fallbackModel, degraded: true };
+      return { models: [fallbackOption], fallbackModel, route, degraded: true };
     }
 
     const models = normalizeLlmModelsPayload(await response.json());
 
     if (models.length === 0) {
-      return { models: [fallbackOption], fallbackModel, degraded: true };
+      return { models: [fallbackOption], fallbackModel, route, degraded: true };
     }
 
     if (!models.some((model) => model.id === fallbackModel)) {
-      return { models: [fallbackOption], fallbackModel, degraded: false };
+      return { models: [fallbackOption], fallbackModel, route, degraded: false };
     }
 
     return {
       models,
       fallbackModel,
+      route,
       degraded: false,
     };
   } catch {
-    return { models: [fallbackOption], fallbackModel, degraded: true };
+    return { models: [fallbackOption], fallbackModel, route, degraded: true };
   }
 }
 

src/modules/_core/schema.ts

@@ -38,6 +38,7 @@ export const users = pgTable("users", {
   assistantEnabled: boolean("assistant_enabled").notNull().default(false),
   assistantName: text("assistant_name").notNull().default("Assistant"),
   assistantSystemPrompt: text("assistant_system_prompt"),
+  assistantModelRoute: text("assistant_model_route"),
   assistantModel: text("assistant_model"),
   defaultEventReminderOffsets: jsonb("default_event_reminder_offsets")
     .notNull()

src/modules/agent/components/assistant-bubble.tsx

@@ -8,10 +8,17 @@ type Props = {
   configured: boolean;
   userId: string;
   assistantName: string;
+  assistantModelRoute: string | null;
   assistantModel: string | null;
 };
 
-export function AssistantBubble({ configured, userId, assistantName, assistantModel }: Props) {
+export function AssistantBubble({
+  configured,
+  userId,
+  assistantName,
+  assistantModelRoute,
+  assistantModel,
+}: Props) {
   const [open, setOpen] = useState(false);
 
   return (
@@ -42,6 +49,7 @@ export function AssistantBubble({ configured, userId, assistantName, assistantMo
             configured={configured}
             userId={userId}
             assistantName={assistantName}
+            assistantModelRoute={assistantModelRoute}
             assistantModel={assistantModel}
           />
         </div>

src/modules/agent/components/assistant-panel.tsx

@@ -2,9 +2,10 @@
 
 import { useCallback, useEffect, useRef, useState, useTransition } from "react";
 import { ChevronDown, ImagePlus, Loader2, Mic, RefreshCw, Send, Square } from "lucide-react";
-import { setAssistantModel } from "@/app/settings/assistant-actions";
+import { setAssistantModel, setAssistantModelRoute } from "@/app/settings/assistant-actions";
 import { Button } from "@/components/ui/button";
 import { Input } from "@/components/ui/input";
+import type { AssistantModelRoute } from "@/lib/llm/models";
 import { consumeAgentChatStream } from "../assistant-chat-stream";
 import {
   clearAssistantChat,
@@ -19,6 +20,7 @@ type Props = {
   configured: boolean;
   userId: string;
   assistantName: string;
+  assistantModelRoute: string | null;
   assistantModel: string | null;
 };
 
@@ -35,6 +37,7 @@ type ModelsResponse = {
   models: LlmModelOption[];
   selectedModel: string;
   fallbackModel: string;
+  route: AssistantModelRoute;
   degraded: boolean;
 };
 
@@ -50,7 +53,17 @@ async function uploadAssistantImage(file: File): Promise<string> {
   return payload.url;
 }
 
-export function AssistantPanel({ configured, userId, assistantName, assistantModel }: Props) {
+function normalizeAssistantModelRoute(value: string | null): AssistantModelRoute {
+  return value === "uncensored" ? "uncensored" : "auto";
+}
+
+export function AssistantPanel({
+  configured,
+  userId,
+  assistantName,
+  assistantModelRoute,
+  assistantModel,
+}: Props) {
   const [messages, setMessages] = useState<AssistantChatMessage[]>(() => loadAssistantChat(userId));
   const [input, setInput] = useState("");
   const [pendingImage, setPendingImage] = useState<PendingImage | null>(null);
@@ -59,6 +72,9 @@ export function AssistantPanel({ configured, userId, assistantName, assistantMod
   const [isPending, setIsPending] = useState(false);
   const [activityLabel, setActivityLabel] = useState<string | null>(null);
   const [modelOptions, setModelOptions] = useState<LlmModelOption[]>([]);
+  const [selectedRoute, setSelectedRoute] = useState<AssistantModelRoute>(() =>
+    normalizeAssistantModelRoute(assistantModelRoute),
+  );
   const [selectedModel, setSelectedModel] = useState(assistantModel ?? "");
   const [fallbackModel, setFallbackModel] = useState("");
   const [modelsDegraded, setModelsDegraded] = useState(false);
@@ -87,31 +103,40 @@ export function AssistantPanel({ configured, userId, assistantName, assistantMod
     };
   }, []);
 
-  const loadModels = useCallback(async (options?: { refresh?: boolean; signal?: AbortSignal }) => {
-    if (options?.signal?.aborted) return;
-
-    setModelsLoading(true);
-    try {
-      const response = await fetch(`/api/agent/models${options?.refresh ? "?refresh=1" : ""}`, {
-        cache: "no-store",
-        signal: options?.signal,
-      });
-      if (!response.ok) throw new Error("Model discovery unavailable");
-      const payload = (await response.json()) as ModelsResponse;
+  const loadModels = useCallback(
+    async (options?: { refresh?: boolean; route?: AssistantModelRoute; signal?: AbortSignal }) => {
       if (options?.signal?.aborted) return;
-      setModelOptions(payload.models);
-      setSelectedModel(payload.selectedModel);
-      setFallbackModel(payload.fallbackModel);
-      setModelsDegraded(payload.degraded);
-      setError(null);
-    } catch (err) {
-      if (err instanceof Error && err.name === "AbortError") return;
-      setModelsDegraded(true);
-      setError("Model discovery unavailable");
-    } finally {
-      if (!options?.signal?.aborted) setModelsLoading(false);
-    }
-  }, []);
+
+      setModelsLoading(true);
+      try {
+        const params = new URLSearchParams();
+        if (options?.refresh) params.set("refresh", "1");
+        if (options?.route) params.set("route", options.route);
+        const query = params.size > 0 ? `?${params.toString()}` : "";
+
+        const response = await fetch(`/api/agent/models${query}`, {
+          cache: "no-store",
+          signal: options?.signal,
+        });
+        if (!response.ok) throw new Error("Model discovery unavailable");
+        const payload = (await response.json()) as ModelsResponse;
+        if (options?.signal?.aborted) return;
+        setModelOptions(payload.models);
+        setSelectedRoute(payload.route);
+        setSelectedModel(payload.selectedModel);
+        setFallbackModel(payload.fallbackModel);
+        setModelsDegraded(payload.degraded);
+        setError(null);
+      } catch (err) {
+        if (err instanceof Error && err.name === "AbortError") return;
+        setModelsDegraded(true);
+        setError("Model discovery unavailable");
+      } finally {
+        if (!options?.signal?.aborted) setModelsLoading(false);
+      }
+    },
+    [],
+  );
 
   useEffect(() => {
     const controller = new AbortController();
@@ -175,6 +200,26 @@ export function AssistantPanel({ configured, userId, assistantName, assistantMod
     });
   }
 
+  function changeModelRoute(nextRoute: string) {
+    const route = normalizeAssistantModelRoute(nextRoute);
+
+    setSelectedRoute(route);
+    setSelectedModel(route);
+    setFallbackModel(route);
+    setModelOptions([]);
+    setModelsDegraded(false);
+    setError(null);
+
+    startTransition(async () => {
+      try {
+        await setAssistantModelRoute(route);
+        await loadModels({ route, refresh: true });
+      } catch (err) {
+        setError(err instanceof Error ? err.message : "Could not save assistant route");
+      }
+    });
+  }
+
   async function sendMessage() {
     const text = input.trim();
     const hasImage = pendingImage !== null;
@@ -242,6 +287,7 @@ export function AssistantPanel({ configured, userId, assistantName, assistantMod
   const inputDisabled = isPending || voiceState === "transcribing" || uploadingImage;
   const canSend =
     !isPending &&
+    !savingModel &&
     voiceState === "idle" &&
     !uploadingImage &&
     (input.trim().length > 0 || pendingImage !== null);
@@ -260,6 +306,22 @@ export function AssistantPanel({ configured, userId, assistantName, assistantMod
           ) : null}
         </div>
         <div className="flex shrink-0 items-center gap-2">
+          <div className="relative max-w-32">
+            <select
+              aria-label="Assistant model route"
+              value={selectedRoute}
+              onChange={(event) => changeModelRoute(event.target.value)}
+              disabled={modelsLoading || savingModel || isPending}
+              className="h-9 w-full max-w-32 appearance-none truncate rounded-[min(var(--radius-md),10px)] border border-input bg-[var(--card)] py-0 pr-8 pl-3 text-[13px] leading-9 text-[var(--ink)] outline-none transition-colors focus-visible:border-ring focus-visible:ring-3 focus-visible:ring-ring/50 disabled:cursor-not-allowed disabled:opacity-50"
+            >
+              <option value="auto">Auto</option>
+              <option value="uncensored">Uncensored</option>
+            </select>
+            <ChevronDown
+              className="pointer-events-none absolute top-1/2 right-2 size-4 -translate-y-1/2 text-muted-foreground"
+              aria-hidden="true"
+            />
+          </div>
           {modelOptions.length > 0 ? (
             <div className="relative max-w-36">
               <select
@@ -287,7 +349,7 @@ export function AssistantPanel({ configured, userId, assistantName, assistantMod
             variant="outline"
             aria-label="Refresh assistant models"
             disabled={modelsLoading || savingModel || isPending}
-            onClick={() => void loadModels({ refresh: true })}
+            onClick={() => void loadModels({ route: selectedRoute, refresh: true })}
           >
             {modelsLoading ? (
               <Loader2 className="size-4 animate-spin" />

tests/e2e/assistant.spec.ts

@@ -26,7 +26,15 @@ test("assistant chat smoke after opt-in", async ({ page }) => {
   await page.goto("/");
   await page.getByRole("button", { name: "Open assistant" }).click();
   await expect(page.getByRole("dialog", { name: "Assistant" })).toBeVisible();
-  const modelSelector = page.getByRole("combobox", { name: "Assistant model" });
+  const routeSelector = page.getByRole("combobox", { name: "Assistant model route" });
+  await expect(routeSelector).toBeVisible();
+  await routeSelector.selectOption("auto");
+  await expect(routeSelector).toHaveValue("auto");
+  await expect(routeSelector).toBeEnabled();
+  await routeSelector.selectOption("uncensored");
+  await expect(routeSelector).toHaveValue("uncensored");
+  await expect(routeSelector).toBeEnabled();
+  const modelSelector = page.getByRole("combobox", { name: "Assistant model", exact: true });
   await expect(modelSelector).toBeVisible();
   const refreshModels = page.getByRole("button", { name: "Refresh assistant models" });
   await expect(refreshModels).toBeVisible();

tests/unit/llm-models.test.ts

@@ -1,6 +1,7 @@
 import assert from "node:assert/strict";
 import { describe, it } from "node:test";
 import {
+  isValidAssistantModelRoute,
   isValidLlmModelId,
   listLlmModels,
   normalizeLlmModelsPayload,
@@ -56,6 +57,18 @@ describe("isValidLlmModelId", () => {
   });
 });
 
+describe("isValidAssistantModelRoute", () => {
+  it("accepts the supported route families", () => {
+    assert.equal(isValidAssistantModelRoute("auto"), true);
+    assert.equal(isValidAssistantModelRoute("uncensored"), true);
+  });
+
+  it("rejects unsupported or padded route families", () => {
+    assert.equal(isValidAssistantModelRoute("bogus"), false);
+    assert.equal(isValidAssistantModelRoute(" uncensored "), false);
+  });
+});
+
 describe("listLlmModels", () => {
   it("fetches provider models with API key auth when the fallback is advertised", async () => {
     const requests: Request[] = [];
@@ -88,10 +101,11 @@ describe("listLlmModels", () => {
     assert.equal(result.degraded, true);
   });
 
-  it("fetches the typed uncensored catalog when uncensored is the fallback route", async () => {
+  it("fetches the typed uncensored catalog when uncensored is the selected route", async () => {
     const requests: Request[] = [];
     const result = await listLlmModels({
-      config: { ...openAiConfig, model: "uncensored" },
+      config: { ...openAiConfig, model: "auto" },
+      route: "uncensored",
       fetchImpl: async (input, init) => {
         requests.push(new Request(input, init));
         return Response.json({
@@ -111,6 +125,30 @@ describe("listLlmModels", () => {
       { id: "uncensored", label: "uncensored" },
     ]);
     assert.equal(result.fallbackModel, "uncensored");
+    assert.equal(result.route, "uncensored");
+    assert.equal(result.degraded, false);
+  });
+
+  it("fetches the default catalog when auto is selected over an uncensored deployment default", async () => {
+    const requests: Request[] = [];
+    const result = await listLlmModels({
+      config: { ...openAiConfig, model: "uncensored" },
+      route: "auto",
+      fetchImpl: async (input, init) => {
+        requests.push(new Request(input, init));
+        return Response.json({
+          data: [{ id: "auto" }, { id: "qwen3:8b" }],
+        });
+      },
+    });
+
+    assert.equal(requests[0]?.url, "https://llm.example.test/v1/models");
+    assert.deepEqual(result.models, [
+      { id: "auto", label: "auto" },
+      { id: "qwen3:8b", label: "qwen3:8b" },
+    ]);
+    assert.equal(result.fallbackModel, "auto");
+    assert.equal(result.route, "auto");
     assert.equal(result.degraded, false);
   });