Merge commit '666bbb0726' into dev

docs/architecture/chat.md

@@ -101,7 +101,7 @@ The list handed to the generator is `[system, *history within budget, user]`.
 - The runtime is scoped to the active thread; thread lists and workspaces stay application state. Sending from a new chat first creates a thread titled "New chat". The last open thread of each workspace is remembered in `localStorage`.
 - [`sse.ts`](../../surfsense_local/frontend/src/features/chat/sse.ts) parses the stream chunk-safely: frames split on blank lines, and a partial frame waits for the next read. This token stream is separate from the `/events` channel.
 - From send until the first token, the assistant turn shows a shimmering "Thinking" beside an orbiting-dots indicator ported from surfsense_web ([`thinking-indicator.tsx`](../../surfsense_local/frontend/src/features/chat/thinking-indicator.tsx)), so a model load or a long prompt is never a blank reply. It is one header from send to answer, whose state moves from waiting to thinking to done, so the indicator and shimmer never restart when the trace arrives. A thinking model's trace then streams open below it, above the answer, in a box 13rem tall that fades at its scrolled edges and follows the newest line until the reader scrolls up, and folds to "Thought for N seconds" when the answer starts, the indicator sliding away; the person's own open or close wins until the next reply, and holds the chat's scroll still while the trace slides, with assistant-ui's `useScrollLock` ([`reply-thinking.tsx`](../../surfsense_local/frontend/src/features/chat/reply-thinking.tsx)). The trace rides in the message's `metadata.custom`, beside the citations, not as a content part, so Copy takes the answer alone.
-- Sending is disabled without a usable model or while threads or messages load, and Stop aborts the request. A `409` "no chat model selected" opens model setup. An `error` frame attaches to its assistant turn: auth and not-found errors, and network errors from a remote endpoint, offer Model setup; a network error from the local runtime offers nothing, since no setting restarts it; the rest offer Retry, which resends the same text.
+- Sending is disabled without a usable model or while threads or messages load, and Stop aborts the request. A `409` "no chat model selected" opens model setup. An `error` frame attaches to its assistant turn: auth, not-found and model-cannot-run errors, and network errors from a remote endpoint, offer Model setup; a network error from the local runtime and a context-too-long error offer nothing, since no button fixes either; the rest offer Retry, which resends the same text.
 - Every turn sends the ids of the ready sources the user left included in the sources panel, so unticking a source takes it out of retrieval, and unticking all of them leaves the model with no context.
 
 ## Citation panel
@@ -128,7 +128,6 @@ The list handed to the generator is `[system, *history within budget, user]`.
 
 - A stream that ends without an error but yields no text still renames the thread and stores an empty assistant turn; the discard guard is `failed and not parts`.
 - `budget.py` prices the question at 1,024 tokens and says `MessageText` enforces that, but `MessageText` has no length limit, so a long question can push a turn past the model's window.
-- `model_cannot_run` and `context_too_long` offer Retry in `chat-error-notice.tsx`, which cannot fix either.
 - No live region announces streamed text, and focus does not move to the conversation heading after a thread switch; the dashboard design asks for both.
 - No test covers a client disconnecting mid-reply. The assistant's text is written only when generation ends, inside the stream, so whether a disconnected reply is kept is unverified.
 - With no hits, the system message is the instructions alone, which still ask for `[n]` labels. In the chat eval, Qwen3 1.7B wrote a `[1]` in all three such answers, which chat drops, and one of them never said the sources did not cover the question.

surfsense_local/frontend/src/features/chat/chat-error-notice.tsx

@@ -12,11 +12,16 @@ function actionFor(error: ChatTurnError): Action {
   switch (error.kind) {
     case "provider_auth":
     case "provider_not_found":
+    case "model_cannot_run":
       return "model-setup"
     case "network":
       // A bad base URL is a Model setup fix; a local runtime that isn't
       // running isn't — there's no settings action that starts it.
       return error.provider === "llamacpp" ? "none" : "model-setup"
+    case "context_too_long":
+      // The notice text already says to start a new chat or pick a model
+      // with a larger window — Retry would resend the same overlong turn.
+      return "none"
     default:
       return "retry"
   }

surfsense_local/frontend/src/features/dashboard/dashboard-page.test.tsx

@@ -891,6 +891,85 @@ describe("dashboard chat", () => {
     expect(screen.queryByRole("button", { name: "Model setup" })).toBeNull()
   })
 
+  it("offers no Retry for a context-too-long failure it cannot fix", async () => {
+    const fetchMock = vi.fn(
+      async (input: RequestInfo | URL, init?: RequestInit) => {
+        const path = String(input)
+        if (path === "/llm/providers") {
+          return Response.json([
+            { name: "llamacpp", healthy: true, can_download: true },
+          ])
+        }
+        if (
+          path === "/workspaces/1/documents?document_type=FILE&document_type=NOTE"
+        ) {
+          return Response.json([])
+        }
+        if (path === "/workspaces/1/chat/threads" && !init?.method) {
+          return Response.json([])
+        }
+        if (path === "/workspaces/1/chat/threads" && init?.method === "POST") {
+          return Response.json(
+            {
+              id: 10,
+              workspace_id: 1,
+              title: "Too long",
+              created_at: "2026-09-05T00:00:00Z",
+              updated_at: "2026-09-05T00:00:00Z",
+            },
+            { status: 201 }
+          )
+        }
+        if (path === "/chat/threads/10/messages" && init?.method === "POST") {
+          return new Response(
+            'data: {"type":"accepted","user_message_id":100,"assistant_message_id":101,"user_created_at":"2026-09-05T00:00:00Z"}\n\ndata: {"type":"error","kind":"context_too_long","message":"context window exceeded","provider":"llamacpp"}\n\ndata: [DONE]\n\n',
+            { headers: { "Content-Type": "text/event-stream" } }
+          )
+        }
+        if (path === "/chat/threads/10/messages") {
+          return Response.json([])
+        }
+        return Response.json({ detail: "not found" }, { status: 404 })
+      }
+    )
+    vi.stubGlobal("fetch", fetchMock)
+    const user = userEvent.setup()
+
+    render(
+      <TooltipProvider>
+        <DashboardPage
+          initialProviderAvailable={true}
+          selection={{
+            model_type: "text_gen",
+            provider: "llamacpp",
+            connection_id: null,
+            name: "llama3.2:1b",
+            updated_at: "2026-09-05T00:00:00Z",
+          }}
+          initialWorkspaces={[workspace]}
+          onModelSelected={vi.fn()}
+        />
+      </TooltipProvider>
+    )
+
+    await screen.findByRole("textbox", { name: "Message" })
+    await user.type(
+      screen.getByRole("textbox", { name: "Message" }),
+      "Too long"
+    )
+    await user.click(screen.getByRole("button", { name: "Send message" }))
+
+    expect(
+      await screen.findByText(
+        "This conversation is too long for the model’s context window. Start a new chat or pick a model with a larger window."
+      )
+    ).toBeTruthy()
+    // Retry resends the identical message to the identical model and fails
+    // identically — it cannot fix an oversized context window.
+    const retryButton = screen.queryByRole("button", { name: "Retry" })
+    expect(retryButton, "context_too_long must not offer Retry").toBeNull()
+  })
+
   it("aborts the active stream when stop is pressed", async () => {
     const captured: { signal: AbortSignal | null } = { signal: null }
     const fetchMock = vi.fn(