fix(ai-chat): #381 PR2 review round 1 — F7 restart-survival + unmount abort + anchor-orphan

Do 1 [F7 regression]: транзиентный сбой attach больше не роняет строку и не
теряет ран. В transport fetch-wrapper `204 || !response.ok` оба зовут
onNoActiveStream (восстановить stripped-строку + invalidate + арм poll), и catch
зовёт его перед rethrow — раньше !ok/throw только сбрасывали флаг, и на 5xx/502/
network-blip in-progress ассистент-турн исчезал, durable-ран не отслеживался.
onNoActiveStream — суперсет (его часть-г всё ещё чистит флаг), идемпотентен.
Расширяет литеральный block-3 спеки (там был только сброс флага) — по ревью и
в согласии с интенцией окна «poll must survive a server restart».

Do 2 [stability]: attach-GET абортится при unmount + mount-гейтинг сайд-эффектов.
mountedRef: mount-эффект ре-армит true и в cleanup ставит false + abort
attachAbortRef; onNoActiveStream рано выходит на !mounted, onFinish-recovery
гейтится `wasResumed && mountedRef.current`. Снимает до-10-мин спурьёзный поллинг
+ чужую invalidateQueries + утёкший fetch на новооткрытом чате (и StrictMode
double-resume).

Do 3 [coherence]: anchor-mismatch не оставляет вечную dots-строку. Reconcile
после мержа хвоста мержит fresh-history версию stripped-строки, если её id !=
id хвоста — settl'ит осиротевшую streaming-A над раном B.

Тесты: F7 500 → restore+арм; F7 network-throw → restore+арм; unmount при pending
attach → abort + поздние колбэки не летят. vitest src/features/ai-chat 304
зелёных, grep-guard пуст.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

apps/client/src/features/ai-chat/components/ai-chat-window.tsx

@@ -41,17 +41,10 @@ import { extractPageSlugId } from "@/lib";
 import {
   AI_CHATS_RQ_KEY,
   AI_CHAT_MESSAGES_RQ_KEY,
-  AI_CHAT_RUN_RQ_KEY,
   useAiChatMessagesQuery,
-  useAiChatRunQuery,
   useAiChatsQuery,
   useAiRolesQuery,
 } from "@/features/ai-chat/queries/ai-chat-query.ts";
-import {
-  shouldClearLatchOnQueryError,
-  shouldClearStoppingLatch,
-  shouldObserveRun,
-} from "@/features/ai-chat/utils/run-polling.ts";
 import { workspaceAtom } from "@/features/user/atoms/current-user-atom";
 import ConversationList from "@/features/ai-chat/components/conversation-list.tsx";
 import ChatThread from "@/features/ai-chat/components/chat-thread.tsx";
@@ -85,6 +78,12 @@ const MIN_HEIGHT = 400;
 // Margin kept between the window and the viewport edges while dragging.
 const EDGE_MARGIN = 8;
 
+// #184 phase 1.5: hard cap on the degraded-poll fallback. The poll is armed when
+// a resume attempt could not attach to the live run and disarmed by the thread on
+// settle / local stream; this cap is the ONLY backstop against an endless tick
+// (a stuck 'streaming' row before the boot-sweep, or a user-tail 204 with no run).
+const DEGRADED_POLL_MAX_MS = 10 * 60_000;
+
 /** Compact token formatter: 1.2M / 3.4k / 950. */
 function formatTokens(n: number): string {
   if (n >= 1_000_000) return `${(n / 1_000_000).toFixed(1)}M`;
@@ -242,150 +241,62 @@ export default function AiChatWindow() {
     [roles],
   );
 
+  // #184 phase 1.5: degraded-poll fallback (replaces the F4/F5/F7 latches). When
+  // ChatThread could not attach to a still-running run it arms this via
+  // onResumeFallback(true); the thread disarms it on settle / local stream. The
+  // window only OWNS the timer (armedAtRef stamps when it was armed for the cap).
+  const [degradedPoll, setDegradedPoll] = useState(false);
+  const armedAtRef = useRef(0);
+  const onResumeFallback = useCallback((active: boolean): void => {
+    if (active) armedAtRef.current = Date.now();
+    setDegradedPoll(active);
+  }, []);
+  // Reset the degraded poll whenever the open chat changes: it is scoped to the
+  // resume attempt of the previously-open chat (invariant 8).
+  useEffect(() => {
+    setDegradedPoll(false);
+  }, [activeChatId]);
+
   const { data: messageRows, isLoading: messagesLoading } =
-    useAiChatMessagesQuery(activeChatId ?? undefined);
+    useAiChatMessagesQuery(
+      activeChatId ?? undefined,
+      // DELIBERATELY DUMB (invariant 8 / task 2.4): poll every 2.5s while armed
+      // and under the 10-min cap; otherwise off. NO error checks (TanStack v5
+      // resets fetchFailureCount each fetch, so consecutive errors are not
+      // expressible — and the poll must survive a server restart) and NO tail
+      // checks (the settled/local-stream semantics live in ChatThread, which
+      // disarms via onResumeFallback(false)). The time cap is the only backstop.
+      () =>
+        degradedPoll === true &&
+        Date.now() - armedAtRef.current < DEGRADED_POLL_MAX_MS
+          ? 2500
+          : false,
+    );
 
   // #184 reconnect-and-live-follow. Whether detached agent runs are enabled for
-  // this workspace. The reconnect endpoint itself is NOT flag-gated server-side
-  // (it is only owner-gated and returns `{ run: null }` when the chat has no
-  // run); but when the feature is off no runs are ever created, so polling it
-  // would always come back empty — we gate it off here to avoid pointless polls.
+  // this workspace. When the feature is off no runs are ever created, so the
+  // resume attempt would only ever 204; gating ChatThread's resume on it avoids a
+  // pointless attach round-trip.
   const workspace = useAtomValue(workspaceAtom);
   const autonomousRunsEnabled =
     workspace?.settings?.ai?.autonomousRuns === true;
 
-  // Whether THIS tab is the one actively streaming the open chat's run locally
-  // (it started the run here and holds the SSE). Reported up from ChatThread. We
-  // are the STREAMER while true and a passive OBSERVER while false — the basis of
-  // the observer-vs-streamer detection. Reset to false by the fresh ChatThread's
-  // mount effect on every chat switch.
-  const [localStreaming, setLocalStreaming] = useState(false);
-  const onStreamingChange = useCallback((streaming: boolean) => {
-    setLocalStreaming(streaming);
-  }, []);
-
-  // #184 Stop wiring. While a detached run is being stopped we SUPPRESS the
-  // observer merge so the stopping run's still-persisting output does not
-  // re-stream back into view between the moment the user pressed Stop and the run
-  // actually settling as 'aborted' server-side. Polling itself keeps running (so
-  // the terminal transition is still detected) — only the visual merge is gated.
-  // Cleared when the run is observed terminal (below) or the chat is switched.
-  const [stoppingRun, setStoppingRun] = useState(false);
-  // Reset the stopping latch whenever the open chat changes: it is scoped to the
-  // run of the previously-open chat.
-  useEffect(() => {
-    setStoppingRun(false);
-  }, [activeChatId]);
-
   // Authoritative stop of the open chat's detached run (the Stop button in
-  // autonomous mode). Latch "stopping" first (suppresses the re-stream flash),
-  // then request the server stop — the ONLY thing that ends a detached run; a mere
-  // local SSE abort is a client disconnect the server ignores. On failure we
-  // release the latch so the observer resumes (better to show the live run than to
-  // freeze the view) and surface the error.
+  // autonomous mode). Request the server stop — the ONLY thing that ends a
+  // detached run; a mere local SSE abort is a client disconnect the server
+  // ignores. On failure surface the error.
   const handleServerStop = useCallback(
     (chatId: string): void => {
-      setStoppingRun(true);
-      // #234 F4: drop the PREVIOUS turn's run from the cache so `run` becomes null
-      // until the CURRENT turn's run is fetched fresh. Without this, once the local
-      // stream aborts (localStreaming -> false) the run query re-enables and
-      // react-query SYNCHRONOUSLY returns the still-cached prior terminal run; the
-      // terminal effect would then clear the stopping latch against that STALE run
-      // before the current turn's (still-running, detached, growing) run is ever
-      // observed — re-opening the observer merge and flashing the growing output
-      // over the frozen row. With the cache cleared the terminal effect's
-      // `if (!run) return` holds the latch until the current run itself is observed
-      // terminal (see shouldClearStoppingLatch).
-      queryClient.removeQueries({ queryKey: AI_CHAT_RUN_RQ_KEY(chatId) });
       void stopRun(chatId).catch(() => {
-        setStoppingRun(false);
         notifications.show({
           message: t("Failed to stop the run"),
           color: "red",
         });
       });
     },
-    [t, queryClient],
+    [t],
   );
 
-  // Poll the latest run of the open chat ONLY when we are a passive observer:
-  // feature on, a chat is open, and we are NOT the local streamer (the streamer
-  // already has the live SSE — polling/merging too would double-render). The
-  // query's own status-keyed refetchInterval stops once the run is terminal.
-  const { data: runData, isError: runQueryFailed } = useAiChatRunQuery(
-    activeChatId ?? undefined,
-    autonomousRunsEnabled && !localStreaming,
-  );
-  const run = runData?.run ?? null;
-
-  // Safety net (#234 F4 review): after handleServerStop clears the run cache,
-  // `run` is null until the current turn's run is fetched fresh, and the terminal
-  // effect below holds the latch via `if (!run) return`. If that refetch instead
-  // ERRORS PERMANENTLY (the GET-run keeps failing) while we are no longer the
-  // streamer, the run stays null, its status-keyed refetchInterval is off, and
-  // nothing would ever observe a terminal run — freezing the view with the
-  // observer merge suppressed. Release the latch on that error so the live view
-  // resumes rather than stays stuck (the local stopRun may already have succeeded
-  // independently).
-  //
-  // #234 F7: this must NOT fire on a TRANSIENT error while `run` is still an
-  // ACTIVE held run. In TanStack Query v5 (retry:false) the query's `data` is
-  // RETAINED on error, so `runQueryFailed` can be true while `run` is still
-  // pending/running — releasing then would re-open the observer merge and flash
-  // the growing detached run over the frozen row (the very flash F4 prevents). The
-  // decision is the pure, unit-tested `shouldClearLatchOnQueryError`, which gates
-  // on the run NOT being active: it cures only the genuine permanent-null-freeze
-  // (`run === null`) and never releases against an active run.
-  useEffect(() => {
-    if (
-      shouldClearLatchOnQueryError({
-        stoppingRun,
-        isLocalStreaming: localStreaming,
-        runQueryFailed,
-        run,
-      })
-    )
-      setStoppingRun(false);
-  }, [stoppingRun, localStreaming, runQueryFailed, run]);
-  // The run's incrementally-persisted assistant message to merge into the thread,
-  // but only while we are an observer (never when we are the streamer — guards
-  // against a stale poll fighting the live stream). Includes a terminal run so the
-  // final persisted output is shown on reopen.
-  const observedRow =
-    shouldObserveRun(run, localStreaming) && !stoppingRun
-      ? (runData?.message ?? null)
-      : null;
-
-  // When the observed run reaches a terminal status, do a final messages refetch
-  // so the persisted final state (token/context badge, export source) is shown,
-  // then the query's refetchInterval has already stopped polling. Deduped per run
-  // id so it fires exactly once per run, not on every subsequent poll-less render.
-  const finalizedRunIdRef = useRef<string | null>(null);
-  useEffect(() => {
-    if (!run || !activeChatId) return;
-    if (run.status === "pending" || run.status === "running") {
-      // Active again (a new run) — re-arm so its terminal transition fires once.
-      finalizedRunIdRef.current = null;
-      return;
-    }
-    // Terminal: a stop we requested has landed (or the run finished on its own),
-    // so release the stopping latch — the observer merge can now show the final
-    // persisted (aborted/finished) output without any live re-stream. The decision
-    // is the pure, unit-tested `shouldClearStoppingLatch` (run-polling.ts): release
-    // ONLY when we requested a stop, this tab is no longer the streamer, AND the
-    // CURRENT run is terminal. The #234 F4 cache removal in handleServerStop makes
-    // `run` null (this branch's `if (!run) return` above holds) until the current
-    // turn's run is fetched fresh, so the latch can never clear against a stale
-    // cached run.
-    if (shouldClearStoppingLatch({ stoppingRun, run, isLocalStreaming: localStreaming }))
-      setStoppingRun(false);
-    if (finalizedRunIdRef.current === run.id) return;
-    finalizedRunIdRef.current = run.id;
-    queryClient.invalidateQueries({
-      queryKey: AI_CHAT_MESSAGES_RQ_KEY(activeChatId),
-    });
-  }, [run, activeChatId, queryClient, stoppingRun, localStreaming]);
-
   // The page the user is currently viewing. AiChatWindow lives in a pathless
   // parent layout route, so useParams() can't see :pageSlug. Match the full
   // pathname against the authenticated page route instead so "the current page"
@@ -1034,16 +945,13 @@ export default function AiChatWindow() {
               assistantName={currentRole?.name}
               onTurnFinished={onTurnFinished}
               onServerChatId={onServerChatId}
-              // #184: live-follow a still-running run when we reopened the chat as
-              // a passive observer; null when there is nothing to observe or this
-              // tab is the streamer. onStreamingChange lets the window stop polling
-              // while we are the streamer.
-              observedRow={observedRow}
-              onStreamingChange={onStreamingChange}
+              // #184 phase 1.5: arm/disarm the degraded-poll fallback when a
+              // resume attempt could not attach to the live run; the thread
+              // disarms it on settle / local stream.
+              onResumeFallback={onResumeFallback}
               // #184: in autonomous mode the Stop button must hit the authoritative
               // server stop (a local SSE abort is a client disconnect the server
-              // ignores). onServerStop also arms the "stopping" latch above so the
-              // stopped run's output does not re-stream via the observer merge.
+              // ignores).
               autonomousRunsEnabled={autonomousRunsEnabled}
               onServerStop={handleServerStop}
             />

apps/client/src/features/ai-chat/components/chat-thread.test.tsx

@@ -1,6 +1,13 @@
-import { describe, it, expect, beforeEach, vi } from "vitest";
-import { render, screen, fireEvent, act, cleanup } from "@testing-library/react";
+import { describe, it, expect, beforeEach, afterEach, vi } from "vitest";
+import {
+  render,
+  screen,
+  fireEvent,
+  act,
+  cleanup,
+} from "@testing-library/react";
 import { MantineProvider } from "@mantine/core";
+import { QueryClient, QueryClientProvider } from "@tanstack/react-query";
 
 // Shared, hoisted mock state so the @ai-sdk/react and "ai" module mocks (hoisted
 // above the imports) can expose the captured useChat callbacks / transport and
@@ -12,50 +19,61 @@ const h = vi.hoisted(() => ({
     sendMessage: vi.fn(),
     stop: vi.fn(),
     setMessages: vi.fn(),
+    resumeStream: vi.fn(),
+    // The messages array useChat was seeded with (to assert strip/seed behavior).
+    seededMessages: null as null | unknown[],
     transport: null as null | {
-      prepareSendMessagesRequest: (arg: {
+      prepareSendMessagesRequest?: (arg: {
         messages: unknown[];
         body: Record<string, unknown>;
       }) => { body: Record<string, unknown> };
+      prepareReconnectToStreamRequest?: () => { api?: string };
+      fetch?: (input: unknown, init?: { method?: string }) => Promise<unknown>;
     },
   },
 }));
 
-// Mock useChat: capture onFinish, return the spies and the controllable status.
+// Mock useChat: capture onFinish + seeded messages, return the spies and the
+// controllable status.
 vi.mock("@ai-sdk/react", () => ({
-  useChat: (opts: { onFinish?: (arg: Record<string, unknown>) => void }) => {
+  useChat: (opts: {
+    messages?: unknown[];
+    onFinish?: (arg: Record<string, unknown>) => void;
+  }) => {
     h.state.onFinish = opts.onFinish ?? null;
+    h.state.seededMessages = opts.messages ?? null;
     return {
       messages: [],
       sendMessage: h.state.sendMessage,
       status: h.state.status,
       stop: h.state.stop,
       error: null,
-      // #184: ChatThread reads setMessages to merge a polled observer run.
       setMessages: h.state.setMessages,
+      resumeStream: h.state.resumeStream,
     };
   },
 }));
 
 // Mock "ai": deterministic ids + a transport that records its options so the test
-// can invoke prepareSendMessagesRequest and assert the `interrupted` flag.
+// can invoke prepareSendMessagesRequest / prepareReconnectToStreamRequest / fetch.
 vi.mock("ai", () => {
   let counter = 0;
   return {
     generateId: () => `gid-${counter++}`,
     DefaultChatTransport: class {
-      constructor(opts: {
-        prepareSendMessagesRequest: (arg: {
-          messages: unknown[];
-          body: Record<string, unknown>;
-        }) => { body: Record<string, unknown> };
-      }) {
-        h.state.transport = opts;
+      constructor(opts: Record<string, unknown>) {
+        h.state.transport = opts as never;
       }
     },
   };
 });
 
+// Keep the ai-chat-query import light: ChatThread only needs the messages RQ key,
+// so stub the module to avoid pulling axios / i18n transitively.
+vi.mock("@/features/ai-chat/queries/ai-chat-query.ts", () => ({
+  AI_CHAT_MESSAGES_RQ_KEY: (chatId: string) => ["ai-chat-messages", chatId],
+}));
+
 // Stub the heavy children: MessageList (markdown/render) and ChatInput (the
 // composer). The ChatInput stub exposes a button that queues a message, the only
 // interaction this test needs to populate the queue while "streaming".
@@ -63,49 +81,90 @@ vi.mock("@/features/ai-chat/components/message-list.tsx", () => ({
   default: () => <div data-testid="message-list" />,
 }));
 vi.mock("@/features/ai-chat/components/chat-input.tsx", () => ({
-  default: ({ onQueue }: { onQueue: (text: string) => void }) => (
-    <button data-testid="queue-btn" onClick={() => onQueue("queued text")}>
-      queue
-    </button>
+  default: ({
+    onQueue,
+    onStop,
+  }: {
+    onQueue: (text: string) => void;
+    onStop: () => void;
+  }) => (
+    <>
+      <button data-testid="queue-btn" onClick={() => onQueue("queued text")}>
+        queue
+      </button>
+      <button aria-label="Stop" onClick={() => onStop()}>
+        stop
+      </button>
+    </>
   ),
 }));
 
 import ChatThread from "./chat-thread";
+import type { IAiChatMessageRow } from "@/features/ai-chat/types/ai-chat.types.ts";
 
-function renderThread() {
+function row(
+  id: string,
+  role: string,
+  status?: string,
+  text = "",
+): IAiChatMessageRow {
+  return { id, role, content: text, status, createdAt: "2026-01-01T00:00:00Z" };
+}
+
+function renderThread(props?: {
+  chatId?: string | null;
+  initialRows?: IAiChatMessageRow[];
+  autonomousRunsEnabled?: boolean;
+}) {
   const onTurnFinished = vi.fn();
-  render(
-    <MantineProvider>
-      <ChatThread chatId="c1" initialRows={[]} onTurnFinished={onTurnFinished} />
-    </MantineProvider>,
+  const onResumeFallback = vi.fn();
+  const onServerStop = vi.fn();
+  const queryClient = new QueryClient({
+    defaultOptions: { queries: { retry: false } },
+  });
+  const invalidateSpy = vi.spyOn(queryClient, "invalidateQueries");
+  const { unmount } = render(
+    <QueryClientProvider client={queryClient}>
+      <MantineProvider>
+        <ChatThread
+          chatId={props?.chatId === undefined ? "c1" : props.chatId}
+          initialRows={props?.initialRows ?? []}
+          autonomousRunsEnabled={props?.autonomousRunsEnabled}
+          onTurnFinished={onTurnFinished}
+          onResumeFallback={onResumeFallback}
+          onServerStop={onServerStop}
+        />
+      </MantineProvider>
+    </QueryClientProvider>,
   );
-  return { onTurnFinished };
+  return { onTurnFinished, onResumeFallback, onServerStop, invalidateSpy, unmount };
+}
+
+function resetState() {
+  h.state.status = "streaming";
+  h.state.onFinish = null;
+  h.state.seededMessages = null;
+  h.state.transport = null;
+  h.state.sendMessage.mockClear();
+  h.state.stop.mockClear();
+  h.state.setMessages.mockClear();
+  h.state.resumeStream.mockClear();
 }
 
 describe("ChatThread — send now (#198)", () => {
-  beforeEach(() => {
-    h.state.status = "streaming";
-    h.state.onFinish = null;
-    h.state.sendMessage.mockClear();
-    h.state.stop.mockClear();
-    h.state.transport = null;
-  });
+  beforeEach(resetState);
 
   it("aborts the current turn and resends the queued message on the abort", () => {
     renderThread();
 
-    // Queue a message while the turn is streaming.
     fireEvent.click(screen.getByTestId("queue-btn"));
     const sendNowBtn = screen.getByLabelText("Send now");
     expect(sendNowBtn).toBeTruthy();
 
-    // "Send now" interrupts the current turn (stop), but does NOT send yet —
-    // the resend happens once the abort lands in onFinish.
     fireEvent.click(sendNowBtn);
     expect(h.state.stop).toHaveBeenCalledTimes(1);
     expect(h.state.sendMessage).not.toHaveBeenCalled();
 
-    // The abort we triggered reaches onFinish: the promoted head is flushed.
     act(() => {
       h.state.onFinish?.({
         message: { id: "a", role: "assistant", parts: [] },
@@ -122,10 +181,8 @@ describe("ChatThread — send now (#198)", () => {
     fireEvent.click(screen.getByTestId("queue-btn"));
     fireEvent.click(screen.getByLabelText("Send now"));
 
-    const prep = h.state.transport!.prepareSendMessagesRequest;
-    // The send right after "send now" carries interrupted: true...
+    const prep = h.state.transport!.prepareSendMessagesRequest!;
     expect(prep({ messages: [], body: {} }).body.interrupted).toBe(true);
-    // ...and only that one (the flag is read-and-cleared).
     expect(prep({ messages: [], body: {} }).body.interrupted).toBe(false);
   });
 
@@ -136,42 +193,24 @@ describe("ChatThread — send now (#198)", () => {
     fireEvent.click(screen.getByTestId("queue-btn"));
     fireEvent.click(screen.getByLabelText("Send now"));
 
-    // No turn to interrupt: sent straight away, no abort, not flagged.
     expect(h.state.stop).not.toHaveBeenCalled();
     expect(h.state.sendMessage).toHaveBeenCalledWith({ text: "queued text" });
-    const prep = h.state.transport!.prepareSendMessagesRequest;
+    const prep = h.state.transport!.prepareSendMessagesRequest!;
     expect(prep({ messages: [], body: {} }).body.interrupted).toBe(false);
   });
 });
 
-// The turn-end decision lives in the `onFinish` handler: given the terminal
-// outcome of a turn (`isAbort` / `isDisconnect` / `isError`, or none = clean),
-// it decides whether to CONTINUE (flush the next queued message) or END (leave
-// the queue intact for the user), and which stop notice — if any — to show.
-// `sendNow` is exercised above; these tests pin down the plain outcomes.
 describe("ChatThread — turn-end decision (onFinish)", () => {
-  beforeEach(() => {
-    h.state.status = "streaming";
-    h.state.onFinish = null;
-    h.state.sendMessage.mockClear();
-    h.state.stop.mockClear();
-    h.state.transport = null;
-  });
+  beforeEach(resetState);
 
-  // Drive a fresh onFinish with the given terminal flags after queueing a
-  // message, and report both what the parent was told and whether the queue was
-  // flushed (a resend to the sendMessage spy).
   function finishWith(flags: {
     isAbort?: boolean;
     isDisconnect?: boolean;
     isError?: boolean;
   }) {
-    // Tear down any prior render so the loop-driven "every outcome" case does
-    // not leave duplicate queue buttons in the DOM.
     cleanup();
     h.state.sendMessage.mockClear();
     const { onTurnFinished } = renderThread();
-    // Populate the queue while the turn is streaming.
     fireEvent.click(screen.getByTestId("queue-btn"));
     act(() => {
       h.state.onFinish?.({
@@ -187,16 +226,12 @@ describe("ChatThread — turn-end decision (onFinish)", () => {
 
   it("CONTINUES — flushes the next queued message on a clean finish", () => {
     finishWith({});
-    // Clean finish (no terminal flag): the queued message is auto-sent.
     expect(h.state.sendMessage).toHaveBeenCalledWith({ text: "queued text" });
-    // A clean finish shows no stop notice.
     expect(screen.queryByText("Response stopped.")).toBeNull();
   });
 
   it("ENDS — keeps the queue intact on a user abort and shows the stopped notice", () => {
     finishWith({ isAbort: true });
-    // A plain Stop (not the sendNow interrupt path) must NOT auto-resend: the
-    // queue is preserved for the user to decide.
     expect(h.state.sendMessage).not.toHaveBeenCalled();
     expect(screen.getByText("Response stopped.")).toBeTruthy();
   });
@@ -211,15 +246,11 @@ describe("ChatThread — turn-end decision (onFinish)", () => {
 
   it("ENDS — keeps the queue intact on a stream error (no auto-retry, no stopped notice)", () => {
     finishWith({ isError: true });
-    // Blindly retrying after a failure would be wrong; the queue is left alone.
     expect(h.state.sendMessage).not.toHaveBeenCalled();
-    // isError clears the neutral notice (the error banner covers this case).
     expect(screen.queryByText("Response stopped.")).toBeNull();
   });
 
   it("notifies the parent on EVERY terminal outcome", () => {
-    // The chat-list refresh / new-chat id adoption must run on success and on
-    // every failure path alike.
     for (const flags of [
       {},
       { isAbort: true },
@@ -232,55 +263,411 @@ describe("ChatThread — turn-end decision (onFinish)", () => {
   });
 });
 
-// #184 passive-observer merge: when reconnecting to a still-running run, the
-// parent feeds the polled run message via `observedRow`; ChatThread merges it via
-// setMessages — but ONLY when this tab is NOT itself streaming (the streamer's
-// SSE owns the view, so a stale observedRow must never overwrite it).
-describe("ChatThread — observer run merge (#184)", () => {
-  beforeEach(() => {
-    h.state.onFinish = null;
-    h.state.setMessages.mockReset();
+// #184 phase 1.5: the resumable-SSE client. A reopened tab resumes the live run
+// via the SDK's reconnect transport (attach: replay + tail) instead of polling.
+describe("ChatThread — resume (attach) machinery (#184)", () => {
+  const streamingTail = () => [
+    row("u1", "user", undefined, "hi"),
+    row("a1", "assistant", "streaming", "partial"),
+  ];
+  const settledTail = () => [
+    row("u1", "user", undefined, "hi"),
+    row("a1", "assistant", "succeeded", "done"),
+  ];
+  const userTail = () => [row("u1", "user", undefined, "hi")];
+
+  const visibleMsg = {
+    id: "a1",
+    role: "assistant",
+    parts: [{ type: "text", text: "streamed answer" }],
+  };
+  const emptyMsg = { id: "a1", role: "assistant", parts: [] };
+
+  beforeEach(resetState);
+  // NOTE: do NOT vi.unstubAllGlobals() here — vitest.setup.ts installs
+  // matchMedia/localStorage via vi.stubGlobal and unstubbing wipes them for the
+  // rest of the file. Fetch is re-stubbed per test that needs it.
+  afterEach(cleanup);
+
+  it("resumes on mount only when the flag is on, chatId is set, and the tail is not a settled assistant", () => {
+    // streaming tail -> resume
+    renderThread({ autonomousRunsEnabled: true, initialRows: streamingTail() });
+    expect(h.state.resumeStream).toHaveBeenCalledTimes(1);
+
+    // user tail -> resume (the assistant row may not be seeded yet)
+    cleanup();
+    h.state.resumeStream.mockClear();
+    renderThread({ autonomousRunsEnabled: true, initialRows: userTail() });
+    expect(h.state.resumeStream).toHaveBeenCalledTimes(1);
+
+    // settled assistant tail -> NO resume
+    cleanup();
+    h.state.resumeStream.mockClear();
+    renderThread({ autonomousRunsEnabled: true, initialRows: settledTail() });
+    expect(h.state.resumeStream).not.toHaveBeenCalled();
+
+    // flag off -> NO resume
+    cleanup();
+    h.state.resumeStream.mockClear();
+    renderThread({ autonomousRunsEnabled: false, initialRows: streamingTail() });
+    expect(h.state.resumeStream).not.toHaveBeenCalled();
+
+    // no chatId -> NO resume
+    cleanup();
+    h.state.resumeStream.mockClear();
+    renderThread({
+      autonomousRunsEnabled: true,
+      chatId: null,
+      initialRows: streamingTail(),
+    });
+    expect(h.state.resumeStream).not.toHaveBeenCalled();
   });
 
-  const observedRow = {
-    id: "a-run",
-    role: "assistant",
-    content: "step 1\nstep 2",
-    metadata: {
-      parts: [{ type: "text", text: "step 1\nstep 2" }],
-    },
-    createdAt: "2026-01-01T00:00:00Z",
-  } as const;
+  it("strips the streaming tail from the seed, but keeps a user tail whole", () => {
+    renderThread({ autonomousRunsEnabled: true, initialRows: streamingTail() });
+    // 2 rows in, streaming tail stripped -> 1 seeded message.
+    expect(h.state.seededMessages).toHaveLength(1);
 
-  function renderObserver(status: string) {
-    h.state.status = status;
-    render(
+    cleanup();
+    renderThread({ autonomousRunsEnabled: true, initialRows: userTail() });
+    // user tail is not stripped.
+    expect(h.state.seededMessages).toHaveLength(1);
+  });
+
+  it("builds the attach URL with expect=live&anchor only when the streaming tail was stripped", () => {
+    renderThread({ autonomousRunsEnabled: true, initialRows: streamingTail() });
+    expect(h.state.transport!.prepareReconnectToStreamRequest!().api).toBe(
+      "/api/ai-chat/runs/c1/stream?expect=live&anchor=a1",
+    );
+
+    cleanup();
+    renderThread({ autonomousRunsEnabled: true, initialRows: userTail() });
+    expect(h.state.transport!.prepareReconnectToStreamRequest!().api).toBe(
+      "/api/ai-chat/runs/c1/stream",
+    );
+  });
+
+  async function fetch204() {
+    vi.stubGlobal(
+      "fetch",
+      vi.fn().mockResolvedValue({ status: 204, ok: false }),
+    );
+    await act(async () => {
+      await h.state.transport!.fetch!("http://x", { method: "GET" });
+    });
+  }
+
+  it("204 on a user tail: no crash, no restore, reconcile+invalidate, onResumeFallback(true)", async () => {
+    const { onResumeFallback, invalidateSpy } = renderThread({
+      autonomousRunsEnabled: true,
+      initialRows: userTail(),
+    });
+    await fetch204();
+    // No stripped row -> no restore merge.
+    expect(h.state.setMessages).not.toHaveBeenCalled();
+    expect(invalidateSpy).toHaveBeenCalledWith({
+      queryKey: ["ai-chat-messages", "c1"],
+    });
+    expect(onResumeFallback).toHaveBeenCalledWith(true);
+  });
+
+  it("204 on a streaming tail: restore + invalidate + onResumeFallback(true)", async () => {
+    const { onResumeFallback, invalidateSpy } = renderThread({
+      autonomousRunsEnabled: true,
+      initialRows: streamingTail(),
+    });
+    await fetch204();
+    // Stripped row is restored to the store.
+    expect(h.state.setMessages).toHaveBeenCalledTimes(1);
+    expect(invalidateSpy).toHaveBeenCalledWith({
+      queryKey: ["ai-chat-messages", "c1"],
+    });
+    expect(onResumeFallback).toHaveBeenCalledWith(true);
+  });
+
+  it("F7 restart-survival: a 500 attach failure restores the stripped row AND arms the poll (not lost)", async () => {
+    const { onResumeFallback, invalidateSpy } = renderThread({
+      autonomousRunsEnabled: true,
+      initialRows: streamingTail(),
+    });
+    vi.stubGlobal(
+      "fetch",
+      vi.fn().mockResolvedValue({ status: 500, ok: false }),
+    );
+    await act(async () => {
+      await h.state.transport!.fetch!("http://x", { method: "GET" });
+    });
+    expect(h.state.setMessages).toHaveBeenCalledTimes(1); // stripped row restored
+    expect(invalidateSpy).toHaveBeenCalledWith({
+      queryKey: ["ai-chat-messages", "c1"],
+    });
+    expect(onResumeFallback).toHaveBeenCalledWith(true); // degraded poll armed
+  });
+
+  it("F7 restart-survival: a network throw restores the stripped row AND arms the poll", async () => {
+    const { onResumeFallback, invalidateSpy } = renderThread({
+      autonomousRunsEnabled: true,
+      initialRows: streamingTail(),
+    });
+    vi.stubGlobal(
+      "fetch",
+      vi.fn().mockRejectedValue(new Error("network down")),
+    );
+    await act(async () => {
+      await h.state
+        .transport!.fetch!("http://x", { method: "GET" })
+        .catch(() => undefined); // the wrapper rethrows; swallow here
+    });
+    expect(h.state.setMessages).toHaveBeenCalledTimes(1);
+    expect(invalidateSpy).toHaveBeenCalledWith({
+      queryKey: ["ai-chat-messages", "c1"],
+    });
+    expect(onResumeFallback).toHaveBeenCalledWith(true);
+  });
+
+  it("unmount during a pending attach aborts the controller and gates late callbacks", async () => {
+    const { onResumeFallback, invalidateSpy, unmount } = renderThread({
+      autonomousRunsEnabled: true,
+      initialRows: streamingTail(),
+    });
+    let abortSeen = false;
+    let resolveFetch!: (v: unknown) => void;
+    vi.stubGlobal(
+      "fetch",
+      vi.fn().mockImplementation((_input: unknown, init: RequestInit) => {
+        init.signal?.addEventListener("abort", () => {
+          abortSeen = true;
+        });
+        return new Promise((res) => {
+          resolveFetch = res;
+        });
+      }),
+    );
+    // Kick a reconnect GET (stays pending).
+    let pending!: Promise<unknown>;
+    act(() => {
+      pending = h.state.transport!.fetch!("http://x", { method: "GET" });
+    });
+    // Unmount: the cleanup aborts the in-flight attach.
+    unmount();
+    expect(abortSeen).toBe(true);
+    // A late 204 landing after unmount must NOT arm a poll / invalidate the (now
+    // different) chat.
+    onResumeFallback.mockClear();
+    invalidateSpy.mockClear();
+    await act(async () => {
+      resolveFetch({ status: 204, ok: false });
+      await pending;
+    });
+    expect(onResumeFallback).not.toHaveBeenCalledWith(true);
+    expect(invalidateSpy).not.toHaveBeenCalled();
+  });
+
+  it("a resume fetch error clears resumedTurn so the next local turn flushes the queue", async () => {
+    renderThread({ autonomousRunsEnabled: true, initialRows: streamingTail() });
+    h.state.status = "ready";
+    vi.stubGlobal(
+      "fetch",
+      vi.fn().mockResolvedValue({ status: 500, ok: false }),
+    );
+    await act(async () => {
+      await h.state.transport!.fetch!("http://x", { method: "GET" });
+    });
+    // Queue then clean-finish: suppression was cleared, so the queue flushes.
+    fireEvent.click(screen.getByTestId("queue-btn"));
+    act(() => {
+      h.state.onFinish?.({
+        message: visibleMsg,
+        isAbort: false,
+        isDisconnect: false,
+        isError: false,
+      });
+    });
+    expect(h.state.sendMessage).toHaveBeenCalledWith({ text: "queued text" });
+  });
+
+  it("a resumed turn's onFinish does NOT flush the queue", () => {
+    renderThread({ autonomousRunsEnabled: true, initialRows: streamingTail() });
+    fireEvent.click(screen.getByTestId("queue-btn"));
+    act(() => {
+      h.state.onFinish?.({
+        message: visibleMsg,
+        isAbort: false,
+        isDisconnect: false,
+        isError: false,
+      });
+    });
+    expect(h.state.sendMessage).not.toHaveBeenCalled();
+  });
+
+  it("a healthy resumed finish (visible content) arms nothing and keeps the store", () => {
+    h.state.status = "ready";
+    const { onResumeFallback } = renderThread({
+      autonomousRunsEnabled: true,
+      initialRows: streamingTail(),
+    });
+    h.state.setMessages.mockClear();
+    onResumeFallback.mockClear();
+    act(() => {
+      h.state.onFinish?.({
+        message: visibleMsg,
+        isAbort: false,
+        isDisconnect: false,
+        isError: false,
+      });
+    });
+    // No restore (would clobber the fuller streamed message), no poll arm.
+    expect(h.state.setMessages).not.toHaveBeenCalled();
+    expect(onResumeFallback).not.toHaveBeenCalledWith(true);
+  });
+
+  it("isDisconnect WITH visible content arms the poll but does NOT restore", () => {
+    h.state.status = "ready";
+    const { onResumeFallback, invalidateSpy } = renderThread({
+      autonomousRunsEnabled: true,
+      initialRows: streamingTail(),
+    });
+    h.state.setMessages.mockClear();
+    onResumeFallback.mockClear();
+    invalidateSpy.mockClear();
+    act(() => {
+      h.state.onFinish?.({
+        message: visibleMsg,
+        isAbort: false,
+        isDisconnect: true,
+        isError: false,
+      });
+    });
+    expect(onResumeFallback).toHaveBeenCalledWith(true);
+    expect(invalidateSpy).toHaveBeenCalledWith({
+      queryKey: ["ai-chat-messages", "c1"],
+    });
+    // Restore forbidden: the on-screen partial must not roll back.
+    expect(h.state.setMessages).not.toHaveBeenCalled();
+  });
+
+  it("an empty resumed message (starved replay) restores the stripped row AND arms the poll", () => {
+    h.state.status = "ready";
+    const { onResumeFallback } = renderThread({
+      autonomousRunsEnabled: true,
+      initialRows: streamingTail(),
+    });
+    h.state.setMessages.mockClear();
+    onResumeFallback.mockClear();
+    act(() => {
+      h.state.onFinish?.({
+        message: emptyMsg,
+        isAbort: false,
+        isDisconnect: false,
+        isError: false,
+      });
+    });
+    expect(h.state.setMessages).toHaveBeenCalledTimes(1); // restore
+    expect(onResumeFallback).toHaveBeenCalledWith(true); // arm
+  });
+
+  it("degraded-merge: merges the tail per initialRows update, and settles disarm the poll", async () => {
+    h.state.status = "ready";
+    const { rerender, onResumeFallback } = renderResumable(streamingTail());
+    // Arm reconcile via a 204.
+    vi.stubGlobal(
+      "fetch",
+      vi.fn().mockResolvedValue({ status: 204, ok: false }),
+    );
+    await act(async () => {
+      await h.state.transport!.fetch!("http://x", { method: "GET" });
+    });
+    h.state.setMessages.mockClear();
+    onResumeFallback.mockClear();
+
+    // A streaming-tail update: merge, poll stays armed.
+    rerender([
+      row("u1", "user", undefined, "hi"),
+      row("a1", "assistant", "streaming", "step 1\nstep 2"),
+    ]);
+    expect(h.state.setMessages).toHaveBeenCalledTimes(1);
+    expect(onResumeFallback).not.toHaveBeenCalledWith(false);
+
+    // A settled-tail update: merge + disarm.
+    h.state.setMessages.mockClear();
+    rerender([
+      row("u1", "user", undefined, "hi"),
+      row("a1", "assistant", "succeeded", "final"),
+    ]);
+    expect(h.state.setMessages).toHaveBeenCalledTimes(1);
+    expect(onResumeFallback).toHaveBeenCalledWith(false);
+  });
+
+  it("a local stream disarms both the merge and the poll", () => {
+    h.state.status = "streaming";
+    const { rerender, onResumeFallback } = renderResumable(streamingTail());
+    onResumeFallback.mockClear();
+    // A re-render while streaming: the reconciliation effect disarms.
+    rerender(streamingTail());
+    expect(onResumeFallback).toHaveBeenCalledWith(false);
+  });
+
+  it("Send now is hidden on a resumed turn but visible on a local stream", () => {
+    // Resumed turn: hidden.
+    renderThread({ autonomousRunsEnabled: true, initialRows: streamingTail() });
+    fireEvent.click(screen.getByTestId("queue-btn"));
+    expect(screen.queryByLabelText("Send now")).toBeNull();
+
+    // Local streaming turn (no resume): visible.
+    cleanup();
+    resetState();
+    renderThread({ initialRows: [] });
+    fireEvent.click(screen.getByTestId("queue-btn"));
+    expect(screen.getByLabelText("Send now")).toBeTruthy();
+  });
+
+  it("handleStop aborts the attach controller and calls onServerStop", async () => {
+    const { onServerStop } = renderThread({
+      autonomousRunsEnabled: true,
+      initialRows: streamingTail(),
+    });
+    // Establish an attach controller via a (pending) reconnect GET.
+    let abortSeen = false;
+    vi.stubGlobal(
+      "fetch",
+      vi.fn().mockImplementation((_input: unknown, init: RequestInit) => {
+        init.signal?.addEventListener("abort", () => {
+          abortSeen = true;
+        });
+        return new Promise(() => undefined); // never resolves
+      }),
+    );
+    act(() => {
+      void h.state.transport!.fetch!("http://x", { method: "GET" });
+    });
+    fireEvent.click(screen.getByLabelText("Stop"));
+    expect(abortSeen).toBe(true);
+    expect(onServerStop).toHaveBeenCalledWith("c1");
+  });
+});
+
+// Helper: render a resumable thread and expose a rerender that only swaps
+// initialRows (the degraded-merge effect depends on it).
+function renderResumable(initialRows: IAiChatMessageRow[]) {
+  const onResumeFallback = vi.fn();
+  const queryClient = new QueryClient({
+    defaultOptions: { queries: { retry: false } },
+  });
+  const Wrapper = ({ rows }: { rows: IAiChatMessageRow[] }) => (
+    <QueryClientProvider client={queryClient}>
       <MantineProvider>
         <ChatThread
           chatId="c1"
-          initialRows={[]}
+          initialRows={rows}
+          autonomousRunsEnabled
           onTurnFinished={vi.fn()}
-          observedRow={observedRow as never}
+          onResumeFallback={onResumeFallback}
         />
-      </MantineProvider>,
-    );
-  }
-
-  it("merges the polled run message when this tab is a passive observer", () => {
-    renderObserver("ready");
-    expect(h.state.setMessages).toHaveBeenCalledTimes(1);
-    // The updater replaces/append the observed assistant row by id.
-    const updater = h.state.setMessages.mock.calls[0][0] as (
-      prev: { id: string; parts: { text: string }[] }[],
-    ) => { id: string; parts: { text: string }[] }[];
-    const merged = updater([{ id: "u1", parts: [{ text: "hi" }] }]);
-    expect(merged).toHaveLength(2);
-    expect(merged[1].id).toBe("a-run");
-    expect(merged[1].parts[0].text).toBe("step 1\nstep 2");
-  });
-
-  it("does NOT merge while THIS tab is the streamer (no double-render)", () => {
-    renderObserver("streaming");
-    expect(h.state.setMessages).not.toHaveBeenCalled();
-  });
-});
+      </MantineProvider>
+    </QueryClientProvider>
+  );
+  const view = render(<Wrapper rows={initialRows} />);
+  const rerender = (rows: IAiChatMessageRow[]) =>
+    act(() => view.rerender(<Wrapper rows={rows} />));
+  return { rerender, onResumeFallback };
+}

apps/client/src/features/ai-chat/components/chat-thread.tsx

@@ -1,4 +1,5 @@
 import { useCallback, useEffect, useMemo, useRef, useState } from "react";
+import { useQueryClient } from "@tanstack/react-query";
 import { generateId } from "ai";
 import { ActionIcon, Box, Group, Stack, Text, Tooltip } from "@mantine/core";
 import {
@@ -24,7 +25,14 @@ import {
 } from "@/features/ai-chat/utils/role-launch.ts";
 import { describeChatError } from "@/features/ai-chat/utils/error-message.ts";
 import { extractServerChatId } from "@/features/ai-chat/utils/adopt-chat-id.ts";
-import { mergeObservedMessage } from "@/features/ai-chat/utils/run-polling.ts";
+import { assistantMessageHasVisibleContent } from "@/features/ai-chat/utils/message-content.ts";
+import {
+  isStreamingTail,
+  isSettledAssistantTail,
+  seedRows,
+  mergeById,
+} from "@/features/ai-chat/utils/resume-helpers.ts";
+import { AI_CHAT_MESSAGES_RQ_KEY } from "@/features/ai-chat/queries/ai-chat-query.ts";
 import {
   dequeue,
   enqueueMessage,
@@ -87,19 +95,13 @@ interface ChatThreadProps {
    *  Copy/export button available mid-stream). Distinct from onTurnFinished,
    *  which fires only at the terminal outcome. */
   onServerChatId?: (serverChatId?: string) => void;
-  /** #184 reconnect-and-live-follow. When THIS tab reopened a chat whose agent
-   *  run is still going (it is a PASSIVE OBSERVER — it did not start the run here),
-   *  the parent polls the reconnect endpoint and feeds the run's incrementally-
-   *  persisted assistant message here; we merge it into the live list so new
-   *  steps/tool-calls appear as they are persisted. Null when there is nothing to
-   *  observe (no run, feature off, or this tab IS the streamer). The merge is
-   *  ADDITIONALLY guarded by our own `isStreaming`, so a stale value can never
-   *  fight the local stream when we are the streamer. */
-  observedRow?: IAiChatMessageRow | null;
-  /** Report this tab's live streaming status up to the parent, so it can stop
-   *  polling the run while WE are the active streamer (the SSE owns the view) and
-   *  resume once we go idle. Called from an effect on every transition. */
-  onStreamingChange?: (streaming: boolean) => void;
+  /** #184 phase 1.5: arm/disarm the parent's degraded-poll fallback for THIS
+   *  chat's window. Called `true` when a resume attempt could not attach to the
+   *  live run (attach 204 / starved-or-torn resumed finish), so the window starts
+   *  a dumb timed poll of the message history to follow the detached run to settle;
+   *  called `false` the moment a local stream starts or the terminal settled row is
+   *  merged (invariant 8). The window owns the timer + its 10-min cap. */
+  onResumeFallback?: (active: boolean) => void;
   /** #184: whether detached/autonomous agent runs are enabled for this workspace.
    *  When true the Stop button must additionally hit the AUTHORITATIVE server stop
    *  (via onServerStop) — aborting only the local SSE is just a client disconnect,
@@ -155,15 +157,58 @@ export default function ChatThread({
   assistantName,
   onTurnFinished,
   onServerChatId,
-  observedRow,
-  onStreamingChange,
+  onResumeFallback,
   autonomousRunsEnabled,
   onServerStop,
 }: ChatThreadProps) {
   const { t } = useTranslation();
+  const queryClient = useQueryClient();
+
+  // resume machinery refs (#184 phase 1.5)
+  const attachAbortRef = useRef<AbortController | null>(null);
+  const reconcileTailRef = useRef(false);
+  const noStreamHandledRef = useRef(false);
+  const onNoActiveStreamRef = useRef<(() => void) | null>(null);
+  // Live mount flag. The attach GET and the resumed `onFinish` are async and can
+  // land AFTER this thread unmounts (the parent remounts per chat via `key`); with
+  // chatIdRef then pointing at the NEW chat, an ungated late callback would arm a
+  // spurious poll + foreign invalidation on the newly-opened chat. Every parent-
+  // facing resume side-effect is gated on this.
+  const mountedRef = useRef(true);
+  const [resumedTurn, setResumedTurn] = useState(false);
+  const resumedTurnRef = useRef(false);
+  // Identity-stable pair setter (bare useState setter + ref write): it is closed
+  // over by the transport useMemo([]), so it MUST NOT capture state.
+  const setResumedTurnPair = useCallback((v: boolean) => {
+    resumedTurnRef.current = v;
+    setResumedTurn(v);
+  }, []);
+
+  // Mount-time resume gating (in refs — computed once for this mount; the parent
+  // remounts per chat via `key`).
+  //
+  // Attempt resume for any non-settled tail: a streaming tail (strip + expect
+  // live replay) or a user tail (the run may exist but its assistant row is not
+  // seeded yet — attach to the pre-opened registry entry and wait for frames).
+  // A settled assistant tail must NEVER resume: replaying a finished run into a
+  // store that already contains its message duplicates parts (SDK text-start
+  // always pushes a new part).
+  const stripRef = useRef(chatId !== null && isStreamingTail(initialRows ?? []));
+  const attemptResumeRef = useRef(
+    autonomousRunsEnabled === true &&
+      chatId !== null &&
+      !isSettledAssistantTail(initialRows ?? []),
+  );
+  const strippedRowRef = useRef<IAiChatMessageRow | null>(
+    stripRef.current ? (initialRows ?? [])[initialRows!.length - 1] : null,
+  );
 
   const initialMessages = useMemo<UIMessage[]>(
-    () => (initialRows ?? []).map(rowToUiMessage),
+    () =>
+      seedRows(
+        initialRows ?? [],
+        attemptResumeRef.current && stripRef.current,
+      ).map(rowToUiMessage),
     [initialRows],
   );
 
@@ -261,9 +306,12 @@ export default function ChatThread({
     const { head, rest } = dequeue(queuedRef.current);
     if (!head) return false;
     setQueue(rest);
+    // Local send: clear any resume-suppression flag so this genuine local turn's
+    // onFinish flushes normally (invariant 8).
+    setResumedTurnPair(false);
     sendMessageRef.current?.({ text: head.text });
     return true;
-  }, [setQueue]);
+  }, [setQueue, setResumedTurnPair]);
 
   const enqueue = useCallback(
     (text: string) => {
@@ -283,6 +331,47 @@ export default function ChatThread({
       new DefaultChatTransport<UIMessage>({
         api: "/api/ai-chat/stream",
         credentials: "include",
+        prepareReconnectToStreamRequest: () => ({
+          // SDK default URL uses the useChat STORE id — always build from the real chat id.
+          // ?expect=live&anchor=<row id> ONLY when we stripped a streaming tail: expect=live
+          // is the only case where a finished-retained replay is safe (the row is stripped,
+          // replay rebuilds it), and the anchor pins the replay to OUR run — a mismatching
+          // (newer) run must 204 into the restore+poll path instead of replaying a foreign
+          // transcript into this store.
+          api: `/api/ai-chat/runs/${chatIdRef.current}/stream${
+            stripRef.current
+              ? `?expect=live&anchor=${strippedRowRef.current!.id}`
+              : ""
+          }`,
+        }),
+        fetch: async (input: RequestInfo | URL, init: RequestInit = {}) => {
+          if ((init.method ?? "GET") !== "GET") return fetch(input, init); // send path untouched
+          // Reconnect GET: the SDK passes no AbortSignal, so wire our own controller
+          // for observer Stop / unmount abort.
+          const controller = new AbortController();
+          attachAbortRef.current = controller;
+          try {
+            const response = await fetch(input, {
+              ...init,
+              signal: controller.signal,
+            });
+            // No onFinish will come for a 204 (silent no-op) OR any non-2xx
+            // (5xx/502 — a server restart mid-attach). Both run the same
+            // no-active-stream recovery: restore the stripped row, invalidate, and
+            // arm the degraded poll (idempotent via noStreamHandledRef; its part-d
+            // also clears the resumedTurn flag). This is the restart-survival path
+            // the removed F7 latch used to guard — a transient attach failure must
+            // NOT drop the in-progress row or stop tracking the durable run.
+            if (response.status === 204 || !response.ok)
+              onNoActiveStreamRef.current?.();
+            return response;
+          } catch (err) {
+            // Network throw: same no-onFinish recovery, then rethrow so the SDK
+            // still surfaces the error to its own machinery.
+            onNoActiveStreamRef.current?.();
+            throw err;
+          }
+        },
         // Inject the chat id and the currently-open page alongside the useChat
         // messages so the server can resolve an existing chat (or create one
         // when null) and tell the agent which page "this page" refers to. Both
@@ -312,7 +401,15 @@ export default function ChatThread({
     [],
   );
 
-  const { messages, sendMessage, status, stop, error, setMessages } = useChat({
+  const {
+    messages,
+    sendMessage,
+    status,
+    stop,
+    error,
+    setMessages,
+    resumeStream,
+  } = useChat({
     // Stable per-mount key. Existing chats use their real id; new chats use a
     // generated client id (never `undefined`) so the store is NOT re-created on
     // every render mid-stream (see `chatStoreId` above).
@@ -330,6 +427,38 @@ export default function ChatThread({
     // would be wrong, so on Stop/disconnect/error the queue is left intact for
     // the user to decide.
     onFinish: ({ message, isAbort, isDisconnect, isError }) => {
+      // (1) Capture whether THIS finish belongs to a resumed (attach) turn and
+      // immediately clear the flag so it can never suppress a LATER local turn.
+      const wasResumed = resumedTurnRef.current;
+      setResumedTurnPair(false);
+      // (2) Recovery after a starved/torn resumed finish (invariant 9). The arm
+      // and the stripped-row restore are gated DIFFERENTLY. Skip entirely once
+      // unmounted (an abort-triggered onFinish landing after a chat switch must
+      // not arm a poll / invalidate on the new chat).
+      if (wasResumed && mountedRef.current) {
+        const hasVisibleContent = assistantMessageHasVisibleContent(message);
+        // ARM the reconcile + degraded poll when the resumed message carries no
+        // visible content (starved replay) OR the connection dropped mid-run — in
+        // both cases the poll must drive the row to its real terminal state.
+        if (isDisconnect || !hasVisibleContent) {
+          reconcileTailRef.current = true;
+          queryClient.invalidateQueries({
+            queryKey: AI_CHAT_MESSAGES_RQ_KEY(chatIdRef.current),
+          });
+          onResumeFallback?.(true);
+        }
+        // RESTORE the stripped streaming row ONLY when the resumed message has no
+        // visible content. On isDisconnect WITH visible content restore is
+        // FORBIDDEN: the live stream may have advanced far past the mount-time
+        // snapshot, so restoring would clobber on-screen content (invariant 9) —
+        // the arm above suffices, the poll reaches the true terminal.
+        if (!hasVisibleContent && strippedRowRef.current) {
+          setMessages((prev) =>
+            mergeById(prev, rowToUiMessage(strippedRowRef.current!)),
+          );
+        }
+      }
+      // (3) Standard branches.
       // Forward the authoritative server chatId (streamed on the assistant
       // message metadata) so the parent adopts the REAL created chat id for a new
       // chat — see adopt-chat-id.ts for the full #137 design. `threadKey` lets the
@@ -342,6 +471,10 @@ export default function ChatThread({
       else if (isAbort) setStopNotice("manual");
       else if (isDisconnect) setStopNotice("disconnect");
       else setStopNotice(null);
+      // A resumed turn NEVER flushes the queue (invariant 7): skip BOTH the
+      // flush-on-abort branch and the plain flush. The local streamer is the only
+      // tab that owns the queue.
+      if (wasResumed) return;
       // "Send now": WE triggered this abort to interrupt the current turn and
       // immediately send the promoted head. Flush it even though the turn was
       // aborted (the normal abort path below keeps the queue intact). The
@@ -423,26 +556,98 @@ export default function ChatThread({
 
   const isStreaming = status === "submitted" || status === "streaming";
 
-  // #184: report our live streaming status up so the parent stops polling the run
-  // while WE are the streamer (the SSE owns the view) and resumes once we go idle.
-  // Effect (not render) so it never updates parent state during our own render;
-  // fires on mount with `false`, which also re-syncs the parent after a chat
-  // switch remounts this thread (a fresh mount is idle until the user sends).
-  useEffect(() => {
-    onStreamingChange?.(isStreaming);
-  }, [isStreaming, onStreamingChange]);
+  // 204-handler (`onNoActiveStream`): the attach returned 204 — nothing live to
+  // resume (overflow / begin-failure / after retention / anchor-mismatch). One-
+  // shot via noStreamHandledRef (we do NOT null onNoActiveStreamRef). Exactly four
+  // parts. Kept in a ref (read by the transport's fetch closure) and refreshed
+  // each render below.
+  const onNoActiveStream = useCallback(() => {
+    // A late attach outcome after unmount must not arm a poll / invalidate on the
+    // now-different chat this thread's refs were reused for.
+    if (!mountedRef.current) return;
+    if (noStreamHandledRef.current) return;
+    noStreamHandledRef.current = true;
+    // (a) Restore the stripped streaming row to the store — ONLY when we actually
+    // stripped one (a user-tail 204 does NOT reach here with a stripped row, so do
+    // not dereference null).
+    if (strippedRowRef.current) {
+      setMessages((prev) =>
+        mergeById(prev, rowToUiMessage(strippedRowRef.current!)),
+      );
+    }
+    // (b) Reconcile the tail from the message history + invalidate it so the
+    // degraded poll starts from a fresh fetch.
+    reconcileTailRef.current = true;
+    queryClient.invalidateQueries({
+      queryKey: AI_CHAT_MESSAGES_RQ_KEY(chatIdRef.current),
+    });
+    // (c) Arm the degraded poll (a dumb timer with a 10-min cap in the window);
+    // the thread disarms it via onResumeFallback(false) on settle / local stream.
+    onResumeFallback?.(true);
+    // (d) 204 means onFinish will NOT fire — clear the suppression flag so it
+    // cannot swallow the NEXT local turn's queue flush.
+    setResumedTurnPair(false);
+  }, [setMessages, queryClient, onResumeFallback, setResumedTurnPair]);
+  onNoActiveStreamRef.current = onNoActiveStream;
 
-  // #184 passive-observer merge: when the parent feeds a polled run message (we
-  // reopened a chat whose run is still going and did NOT start it here), merge it
-  // into the live list so new steps/tool-calls appear as they are persisted. Hard-
-  // gated by `!isStreaming`: if THIS tab is actually the streamer, the local SSE
-  // owns the view and a stale observedRow must never overwrite it. `observedRow`
-  // is a stable per-poll object, so this runs once per poll, not per render.
+  // Mount effect: kick off the resume attempt for a non-settled tail. Marking the
+  // turn as resumed BEFORE resumeStream so onFinish (invariant 7/8) sees it.
   useEffect(() => {
-    if (isStreaming || !observedRow) return;
-    const observed = rowToUiMessage(observedRow);
-    setMessages((prev) => mergeObservedMessage(prev, observed));
-  }, [observedRow, isStreaming, setMessages]);
+    // Re-arm on (re)mount — StrictMode dev-mounts twice, and the cleanup below
+    // flips this false between the two.
+    mountedRef.current = true;
+    if (attemptResumeRef.current) {
+      setResumedTurnPair(true);
+      void resumeStream();
+    }
+    // Unmount: mark unmounted (gates late attach/onFinish side-effects) and abort
+    // the in-flight attach GET so its callbacks don't fire against the next chat.
+    return () => {
+      mountedRef.current = false;
+      attachAbortRef.current?.abort();
+    };
+    // Mount-only by design; the parent remounts per chat via `key`.
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, []);
+
+  // Reconciliation + degraded-merge (invariant 8). Deps are EXACTLY
+  // [initialRows, isStreaming, setMessages].
+  useEffect(() => {
+    // A local stream owns the view: disarm BOTH the merge and the window poll.
+    if (isStreaming) {
+      reconcileTailRef.current = false;
+      onResumeFallback?.(false);
+      return;
+    }
+    if (!reconcileTailRef.current) return;
+    const rows = initialRows ?? [];
+    const tail = rows[rows.length - 1];
+    if (!tail || tail.role !== "assistant") return;
+    // Merge the polled assistant tail on EVERY initialRows update — while the
+    // degraded poll is active this IS the live per-step progress.
+    setMessages((prev) => mergeById(prev, rowToUiMessage(tail)));
+    // Anchor-mismatch coherence: when we restored a stripped streaming row A but a
+    // DIFFERENT run's row B is now the tail (A finished, B replaced the registry
+    // entry, so the attach 204'd), A would otherwise linger forever as an orphan
+    // jumping-dots row over the real run. Settle it from fresh history (where A is
+    // now persisted) so no phantom row survives. No-op in the common case where A
+    // IS the tail (id match).
+    const stripped = strippedRowRef.current;
+    if (stripped && stripped.id !== tail.id) {
+      const historical = rows.find((r) => r.id === stripped.id);
+      if (historical)
+        setMessages((prev) => mergeById(prev, rowToUiMessage(historical)));
+    }
+    // Settled: the terminal merge is done — disarm the flag AND the window poll
+    // explicitly (the window only has a time cap, it will not disarm itself).
+    if (tail.status !== "streaming") {
+      reconcileTailRef.current = false;
+      onResumeFallback?.(false);
+    }
+    // onResumeFallback intentionally omitted (parent-stable callback); deps are
+    // fixed by the resume design.
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [initialRows, isStreaming, setMessages]);
 
   // "Send now" on a queued message: interrupt the current turn and immediately
   // send THIS message, keeping the agent's partial output. Other queued messages
@@ -469,10 +674,12 @@ export default function ChatThread({
         const msg = queuedRef.current.find((m) => m.id === id);
         if (!msg) return;
         setQueue(removeQueuedById(queuedRef.current, id));
+        // Local send: clear any resume-suppression flag (invariant 8).
+        setResumedTurnPair(false);
         sendMessageRef.current?.({ text: msg.text });
       }
     },
-    [setQueue, stop],
+    [setQueue, stop, setResumedTurnPair],
   );
 
   // Stop the current turn. ALWAYS abort the local SSE (`stop()`) so the composer
@@ -485,6 +692,9 @@ export default function ChatThread({
   // is not known yet — a brand-new chat in the first moment of its first turn —
   // only the local abort happens (there is no server-side run handle to stop yet).
   const handleStop = useCallback(() => {
+    // Abort the resume/attach GET first: the SDK does not pass it a signal, so an
+    // observer's Stop would otherwise leave the attach fetch running.
+    attachAbortRef.current?.abort();
     stop();
     if (!autonomousRunsEnabled) return;
     if (chatIdRef.current) {
@@ -617,17 +827,23 @@ export default function ChatThread({
                 <Text size="xs" lineClamp={2} className={classes.queuedText}>
                   {m.text}
                 </Text>
-                <Tooltip label={t("Interrupt and send now")} withArrow>
-                  <ActionIcon
-                    size="xs"
-                    variant="subtle"
-                    color="blue"
-                    onClick={() => sendNow(m.id)}
-                    aria-label={t("Send now")}
-                  >
-                    <IconPlayerPlayFilled size={12} />
-                  </ActionIcon>
-                </Tooltip>
+                {/* "Send now" (interrupt) is hidden on a RESUMED turn: a local
+                    stop() does not abort the resumed attach fetch, so the click
+                    would be swallowed while flushOnAbortRef would fire minutes
+                    later on the natural finish. Only the remove affordance stays. */}
+                {!resumedTurn && (
+                  <Tooltip label={t("Interrupt and send now")} withArrow>
+                    <ActionIcon
+                      size="xs"
+                      variant="subtle"
+                      color="blue"
+                      onClick={() => sendNow(m.id)}
+                      aria-label={t("Send now")}
+                    >
+                      <IconPlayerPlayFilled size={12} />
+                    </ActionIcon>
+                  </Tooltip>
+                )}
                 <ActionIcon
                   size="xs"
                   variant="subtle"
@@ -642,7 +858,11 @@ export default function ChatThread({
           </Stack>
         )}
         <ChatInput
-          onSend={(text) => sendMessage({ text })}
+          onSend={(text) => {
+            // Local send: clear any resume-suppression flag (invariant 8).
+            setResumedTurnPair(false);
+            sendMessage({ text });
+          }}
           onQueue={enqueue}
           onStop={handleStop}
           isStreaming={isStreaming}

apps/client/src/features/ai-chat/queries/ai-chat-messages-query.test.tsx

@@ -0,0 +1,90 @@
+import { describe, it, expect, vi, beforeEach } from "vitest";
+import React from "react";
+import { renderHook, waitFor } from "@testing-library/react";
+import { QueryClient, QueryClientProvider } from "@tanstack/react-query";
+
+// react-i18next / notifications are pulled in transitively by ai-chat-query.ts
+// (the mutation hooks use them); stub so the module imports cleanly.
+vi.mock("react-i18next", () => ({
+  useTranslation: () => ({ t: (key: string) => key }),
+}));
+vi.mock("@mantine/notifications", () => ({
+  notifications: { show: vi.fn() },
+}));
+
+// Mock the service module; only getAiChatMessages is exercised, but the other
+// named exports must exist so ai-chat-query.ts imports resolve.
+vi.mock("@/features/ai-chat/services/ai-chat-service.ts", () => ({
+  getAiChatMessages: vi.fn(),
+  getAiChats: vi.fn(),
+  getAiRoleCatalog: vi.fn(),
+  getAiRoleCatalogBundle: vi.fn(),
+  getAiRoles: vi.fn(),
+  importAiRolesFromCatalog: vi.fn(),
+  createAiRole: vi.fn(),
+  deleteAiChat: vi.fn(),
+  deleteAiRole: vi.fn(),
+  renameAiChat: vi.fn(),
+  updateAiRole: vi.fn(),
+  updateAiRoleFromCatalog: vi.fn(),
+}));
+
+import { getAiChatMessages } from "@/features/ai-chat/services/ai-chat-service.ts";
+import { useAiChatMessagesQuery } from "@/features/ai-chat/queries/ai-chat-query.ts";
+
+const emptyPage = { items: [], meta: { hasNextPage: false, nextCursor: null } };
+
+function createWrapper() {
+  const queryClient = new QueryClient({
+    defaultOptions: { queries: { retry: false } },
+  });
+  return function Wrapper({ children }: { children: React.ReactNode }) {
+    return (
+      <QueryClientProvider client={queryClient}>{children}</QueryClientProvider>
+    );
+  };
+}
+
+// The degraded-poll fallback (#184 phase 1.5) is threaded into this query as a
+// `refetchInterval`; AiChatWindow supplies the deliberately-dumb callback. These
+// pin the plumbing the window depends on: the interval polls the message history,
+// and — critically — fetch ERRORS do NOT stop the tick (TanStack v5 resets the
+// failure count each fetch, so the poll must survive a server restart).
+describe("useAiChatMessagesQuery — degraded refetchInterval", () => {
+  beforeEach(() => {
+    vi.clearAllMocks();
+  });
+
+  it("re-polls at the interval while the callback returns a duration", async () => {
+    vi.mocked(getAiChatMessages).mockResolvedValue(emptyPage as never);
+    renderHook(() => useAiChatMessagesQuery("c1", () => 30), {
+      wrapper: createWrapper(),
+    });
+    await waitFor(() =>
+      expect(vi.mocked(getAiChatMessages).mock.calls.length).toBeGreaterThan(2),
+    );
+  });
+
+  it("does NOT re-poll when the callback returns false", async () => {
+    vi.mocked(getAiChatMessages).mockResolvedValue(emptyPage as never);
+    renderHook(() => useAiChatMessagesQuery("c1", () => false), {
+      wrapper: createWrapper(),
+    });
+    await waitFor(() =>
+      expect(vi.mocked(getAiChatMessages)).toHaveBeenCalledTimes(1),
+    );
+    // Give any errant interval a chance to fire, then assert it did not.
+    await new Promise((r) => setTimeout(r, 60));
+    expect(vi.mocked(getAiChatMessages)).toHaveBeenCalledTimes(1);
+  });
+
+  it("keeps ticking through fetch errors (errors do not gate the poll)", async () => {
+    vi.mocked(getAiChatMessages).mockRejectedValue(new Error("server down"));
+    renderHook(() => useAiChatMessagesQuery("c1", () => 30), {
+      wrapper: createWrapper(),
+    });
+    await waitFor(() =>
+      expect(vi.mocked(getAiChatMessages).mock.calls.length).toBeGreaterThan(2),
+    );
+  });
+});

apps/client/src/features/ai-chat/queries/ai-chat-query.ts

@@ -13,7 +13,6 @@ import {
   deleteAiChat,
   deleteAiRole,
   getAiChatMessages,
-  getAiChatRun,
   getAiChats,
   getAiRoleCatalog,
   getAiRoleCatalogBundle,
@@ -26,7 +25,6 @@ import {
 import {
   IAiChat,
   IAiChatMessageRow,
-  IAiChatRunResponse,
   IAiRole,
   IAiRoleCatalog,
   IAiRoleCatalogBundle,
@@ -37,7 +35,6 @@ import {
   IAiRoleUpdateFromCatalogResult,
 } from "@/features/ai-chat/types/ai-chat.types.ts";
 import { IPagination } from "@/lib/types.ts";
-import { runPollInterval } from "@/features/ai-chat/utils/run-polling.ts";
 
 export const AI_CHATS_RQ_KEY = ["ai-chats"];
 export const AI_ROLES_RQ_KEY = ["ai-roles"];
@@ -55,7 +52,6 @@ export const AI_CHAT_MESSAGES_RQ_KEY = (chatId: string) => [
   "ai-chat-messages",
   chatId,
 ];
-export const AI_CHAT_RUN_RQ_KEY = (chatId: string) => ["ai-chat-run", chatId];
 
 /** Paginated list of the current user's chats (auto-loads further pages). */
 export function useAiChatsQuery() {
@@ -89,7 +85,15 @@ export function useAiChatsQuery() {
  * Load all persisted messages of a chat (oldest first), flattening the
  * paginated server response. Used to seed `useChat` initial messages.
  */
-export function useAiChatMessagesQuery(chatId: string | undefined) {
+export function useAiChatMessagesQuery(
+  chatId: string | undefined,
+  // #184 phase 1.5: the degraded-poll fallback. When a tab could not attach to a
+  // still-running run (the attach returned 204 / the resumed stream ended with no
+  // terminal row), the window arms a dumb timed poll of the message history to
+  // follow the detached run to settle. The callback form lives in AiChatWindow;
+  // threaded here verbatim so this query owns the polling. Undefined => no poll.
+  refetchInterval?: number | false | (() => number | false),
+) {
   const query = useInfiniteQuery({
     queryKey: AI_CHAT_MESSAGES_RQ_KEY(chatId ?? ""),
     queryFn: ({ pageParam }) =>
@@ -100,6 +104,7 @@ export function useAiChatMessagesQuery(chatId: string | undefined) {
         ? (lastPage.meta.nextCursor ?? undefined)
         : undefined,
     enabled: !!chatId,
+    refetchInterval,
   });
 
   // useInfiniteQuery only fetches the first page on its own. The hook's contract
@@ -139,34 +144,6 @@ export function useAiChatMessagesQuery(chatId: string | undefined) {
   };
 }
 
-/**
- * Reconnect to a chat's latest agent run and LIVE-FOLLOW it (#184). While the run
- * is active the query re-polls every {@link runPollInterval} ms (driven off the
- * fetched `run.status`, the same status-keyed refetchInterval pattern as the
- * embeddings reindex polling); once the run reaches a terminal status — or there
- * is no run — the interval returns `false` and polling stops on its own. Polling
- * is thus naturally bounded by the run terminating; no separate timeout cap.
- *
- * `enabled` gates the whole thing: callers pass `false` when the autonomous-runs
- * feature is off (the endpoint is NOT flag-gated server-side, but with the feature
- * off the chat has no runs, so polling would only ever return `{ run: null }`) OR
- * when THIS tab is the one actively streaming the run (the live SSE owns the view,
- * so we must not also poll/merge). The global `retry: false` means a failed fetch
- * leaves `data` undefined, so refetchInterval(undefined run) returns false — a
- * failed fetch can never spin a tight loop.
- */
-export function useAiChatRunQuery(
-  chatId: string | undefined,
-  enabled: boolean,
-) {
-  return useQuery<IAiChatRunResponse, Error>({
-    queryKey: AI_CHAT_RUN_RQ_KEY(chatId ?? ""),
-    queryFn: () => getAiChatRun(chatId as string),
-    enabled: !!chatId && enabled,
-    refetchInterval: (query) => runPollInterval(query.state.data?.run),
-  });
-}
-
 export function useRenameAiChatMutation() {
   const queryClient = useQueryClient();
   const { t } = useTranslation();

apps/client/src/features/ai-chat/queries/ai-chat-run-query.test.tsx

@@ -1,92 +0,0 @@
-import { describe, it, expect, vi, beforeEach } from "vitest";
-import React from "react";
-import { renderHook, waitFor } from "@testing-library/react";
-import { QueryClient, QueryClientProvider } from "@tanstack/react-query";
-import type { IAiChatRunResponse } from "@/features/ai-chat/types/ai-chat.types.ts";
-
-// react-i18next is pulled in transitively by ai-chat-query.ts (the mutation hooks
-// use it); stub it so the module imports cleanly in this hook test.
-vi.mock("react-i18next", () => ({
-  useTranslation: () => ({ t: (key: string) => key }),
-}));
-
-vi.mock("@mantine/notifications", () => ({
-  notifications: { show: vi.fn() },
-}));
-
-// Mock the whole service module; only getAiChatRun is exercised here, but the
-// other named exports must exist so ai-chat-query.ts imports resolve.
-vi.mock("@/features/ai-chat/services/ai-chat-service.ts", () => ({
-  getAiChatRun: vi.fn(),
-  getAiChatMessages: vi.fn(),
-  getAiChats: vi.fn(),
-  getAiRoleCatalog: vi.fn(),
-  getAiRoleCatalogBundle: vi.fn(),
-  getAiRoles: vi.fn(),
-  importAiRolesFromCatalog: vi.fn(),
-  createAiRole: vi.fn(),
-  deleteAiChat: vi.fn(),
-  deleteAiRole: vi.fn(),
-  renameAiChat: vi.fn(),
-  updateAiRole: vi.fn(),
-  updateAiRoleFromCatalog: vi.fn(),
-}));
-
-import { getAiChatRun } from "@/features/ai-chat/services/ai-chat-service.ts";
-import { useAiChatRunQuery } from "@/features/ai-chat/queries/ai-chat-query.ts";
-
-function createWrapper() {
-  const queryClient = new QueryClient({
-    defaultOptions: { queries: { retry: false } },
-  });
-  return function Wrapper({ children }: { children: React.ReactNode }) {
-    return (
-      <QueryClientProvider client={queryClient}>{children}</QueryClientProvider>
-    );
-  };
-}
-
-const runningResponse: IAiChatRunResponse = {
-  run: { id: "run-1", chatId: "c1", status: "running" },
-  message: {
-    id: "a1",
-    role: "assistant",
-    content: "working...",
-    createdAt: "2026-01-01T00:00:00Z",
-  },
-};
-
-describe("useAiChatRunQuery — enable gating", () => {
-  beforeEach(() => {
-    vi.clearAllMocks();
-  });
-
-  it("fetches the run when enabled (passive observer, feature on)", async () => {
-    vi.mocked(getAiChatRun).mockResolvedValue(runningResponse);
-    const { result } = renderHook(() => useAiChatRunQuery("c1", true), {
-      wrapper: createWrapper(),
-    });
-    await waitFor(() => expect(result.current.isSuccess).toBe(true));
-    expect(getAiChatRun).toHaveBeenCalledWith("c1");
-    expect(result.current.data?.run?.status).toBe("running");
-  });
-
-  it("does NOT fetch when disabled (this tab is the streamer / feature off)", async () => {
-    vi.mocked(getAiChatRun).mockResolvedValue(runningResponse);
-    renderHook(() => useAiChatRunQuery("c1", false), {
-      wrapper: createWrapper(),
-    });
-    // Give any errant fetch a chance to fire, then assert none did.
-    await new Promise((r) => setTimeout(r, 20));
-    expect(getAiChatRun).not.toHaveBeenCalled();
-  });
-
-  it("does NOT fetch when there is no chat id", async () => {
-    vi.mocked(getAiChatRun).mockResolvedValue(runningResponse);
-    renderHook(() => useAiChatRunQuery(undefined, true), {
-      wrapper: createWrapper(),
-    });
-    await new Promise((r) => setTimeout(r, 20));
-    expect(getAiChatRun).not.toHaveBeenCalled();
-  });
-});

apps/client/src/features/ai-chat/services/ai-chat-service.ts

@@ -5,7 +5,6 @@ import {
   IAiChatListParams,
   IAiChatMessageRow,
   IAiChatMessagesParams,
-  IAiChatRunResponse,
   IAiRole,
   IAiRoleCatalog,
   IAiRoleCatalogBundle,
@@ -43,23 +42,6 @@ export async function getAiChatMessages(
   return req.data;
 }
 
-/**
- * Reconnect to the latest agent run of a chat (#184). Returns the run's
- * persisted lifecycle state and the assistant message it materializes (the
- * partial output while the run is in-flight, the final output once it finished).
- * The DB is the source of truth, so this works for an in-flight run (the browser
- * dropped, the run kept going) and a finished one alike; `{ run: null }` when the
- * chat has never had a run. Owner-gated server-side (the requesting user must own
- * the chat); it is NOT flag-gated — when the feature is off the chat simply has no
- * runs, so the endpoint returns `{ run: null }`.
- */
-export async function getAiChatRun(
-  chatId: string,
-): Promise<IAiChatRunResponse> {
-  const req = await api.post<IAiChatRunResponse>("/ai-chat/run", { chatId });
-  return req.data;
-}
-
 /**
  * Explicitly STOP the active agent run of a chat (#184). This is the ONLY thing
  * that ends a DETACHED run — a mere browser disconnect (aborting the local SSE)

apps/client/src/features/ai-chat/types/ai-chat.types.ts

@@ -210,41 +210,14 @@ export interface IAiChatMessageRow {
     // renders a "stopped" marker on interrupted turns.
     finishReason?: string;
   } | null;
+  // Persisted lifecycle status of the row's turn, carried on the wire by
+  // `baseFields`. 'streaming' marks a still-in-progress assistant row (used by
+  // the resume machinery to decide whether a tail is a live stream to attach to
+  // or a settled row that must not be replayed).
+  status?: string;
   createdAt: string;
 }
 
-/**
- * A persisted agent-run row (#184), mirroring the `ai_chat_runs` fields the
- * client reads from `POST /ai-chat/run`. Only `status` is load-bearing for the
- * reconnect-and-live-update UX (it drives the poll cadence); the rest are carried
- * for display/diagnostics. The DB is the source of truth, so this resolves for an
- * in-flight run (the browser dropped, the run kept going) and a finished one.
- */
-export interface IAiChatRun {
-  id: string;
-  chatId: string;
-  // 'pending' | 'running' | 'succeeded' | 'failed' | 'aborted'. The first two are
-  // ACTIVE (keep polling); the rest are TERMINAL (stop polling).
-  status: "pending" | "running" | "succeeded" | "failed" | "aborted" | string;
-  error?: string | null;
-  stepCount?: number;
-  assistantMessageId?: string | null;
-  startedAt?: string | null;
-  finishedAt?: string | null;
-  createdAt?: string;
-  updatedAt?: string;
-}
-
-/**
- * Response of `POST /ai-chat/run` (#184): the latest run of a chat and the
- * assistant message it materializes (the partial/final output, projected from the
- * persisted rows). Both are `null` when the chat has never had a run.
- */
-export interface IAiChatRunResponse {
-  run: IAiChatRun | null;
-  message: IAiChatMessageRow | null;
-}
-
 export interface IAiChatListParams extends QueryParams {}
 
 export interface IAiChatMessagesParams {

apps/client/src/features/ai-chat/utils/resume-helpers.test.ts

@@ -0,0 +1,112 @@
+import { describe, it, expect } from "vitest";
+import type { UIMessage } from "@ai-sdk/react";
+import type { IAiChatMessageRow } from "@/features/ai-chat/types/ai-chat.types.ts";
+import {
+  isStreamingTail,
+  isSettledAssistantTail,
+  seedRows,
+  mergeById,
+} from "./resume-helpers.ts";
+
+function row(
+  id: string,
+  role: string,
+  status?: string,
+): IAiChatMessageRow {
+  return { id, role, content: "", status, createdAt: "2026-01-01T00:00:00Z" };
+}
+
+function makeMsg(id: string, text: string): UIMessage {
+  return {
+    id,
+    role: "assistant",
+    parts: [{ type: "text", text }],
+  } as UIMessage;
+}
+
+describe("isStreamingTail", () => {
+  it("is true when the last row is a streaming assistant row", () => {
+    expect(
+      isStreamingTail([row("u1", "user"), row("a1", "assistant", "streaming")]),
+    ).toBe(true);
+  });
+
+  it("is false for a settled assistant tail", () => {
+    expect(isStreamingTail([row("a1", "assistant", "succeeded")])).toBe(false);
+    expect(isStreamingTail([row("a1", "assistant")])).toBe(false);
+  });
+
+  it("is false when the tail is a user row or the list is empty", () => {
+    expect(isStreamingTail([row("u1", "user")])).toBe(false);
+    expect(isStreamingTail([])).toBe(false);
+  });
+});
+
+describe("isSettledAssistantTail", () => {
+  it("is true for an assistant tail whose status is not streaming", () => {
+    expect(isSettledAssistantTail([row("a1", "assistant", "succeeded")])).toBe(
+      true,
+    );
+    expect(isSettledAssistantTail([row("a1", "assistant")])).toBe(true);
+    expect(isSettledAssistantTail([row("a1", "assistant", "aborted")])).toBe(
+      true,
+    );
+  });
+
+  it("is false for a streaming assistant tail", () => {
+    expect(isSettledAssistantTail([row("a1", "assistant", "streaming")])).toBe(
+      false,
+    );
+  });
+
+  it("is false when the tail is a user row or the list is empty", () => {
+    expect(isSettledAssistantTail([row("u1", "user")])).toBe(false);
+    expect(isSettledAssistantTail([])).toBe(false);
+  });
+});
+
+describe("seedRows", () => {
+  const rows = [row("u1", "user"), row("a1", "assistant", "streaming")];
+
+  it("returns the rows unchanged when not stripping", () => {
+    expect(seedRows(rows, false)).toBe(rows);
+  });
+
+  it("drops the last row when stripping", () => {
+    const seeded = seedRows(rows, true);
+    expect(seeded).toHaveLength(1);
+    expect(seeded[0].id).toBe("u1");
+  });
+
+  it("returns an empty list when stripping a single-row list", () => {
+    expect(seedRows([row("a1", "assistant", "streaming")], true)).toHaveLength(
+      0,
+    );
+  });
+});
+
+describe("mergeById", () => {
+  it("replaces the message with the same id in place (per-step growth)", () => {
+    const prev = [makeMsg("u1", "hi"), makeMsg("a1", "step 1")];
+    const incoming = makeMsg("a1", "step 1\nstep 2");
+    const next = mergeById(prev, incoming);
+    expect(next).toHaveLength(2);
+    expect(next[1]).toBe(incoming);
+    expect(next[0]).toBe(prev[0]); // untouched
+    expect(next).not.toBe(prev); // new array (never mutates input)
+  });
+
+  it("appends when the incoming message is not yet present", () => {
+    const prev = [makeMsg("u1", "hi")];
+    const incoming = makeMsg("a1", "first token");
+    const next = mergeById(prev, incoming);
+    expect(next).toHaveLength(2);
+    expect(next[1]).toBe(incoming);
+  });
+
+  it("returns the original list unchanged when there is nothing to merge", () => {
+    const prev = [makeMsg("u1", "hi")];
+    expect(mergeById(prev, null)).toBe(prev);
+    expect(mergeById(prev, undefined)).toBe(prev);
+  });
+});

apps/client/src/features/ai-chat/utils/resume-helpers.ts

@@ -0,0 +1,62 @@
+import type { UIMessage } from "@ai-sdk/react";
+import type { IAiChatMessageRow } from "@/features/ai-chat/types/ai-chat.types.ts";
+
+/**
+ * Pure decisions for the resumable-SSE resume machinery (#184 phase 1.5). A tab
+ * that reopens a chat whose agent run is still going attaches to the server's
+ * run-stream registry (replay + live tail) instead of polling snapshots; these
+ * small predicates decide WHICH tail is safe to resume and how to seed the store,
+ * extracted so they can be unit-tested in isolation.
+ */
+
+/**
+ * A STREAMING tail: the last persisted row is an assistant row still marked
+ * `status === 'streaming'`. Such a tail is stripped from the seed and rebuilt by
+ * the replay (`expect=live`), since the SDK's `text-start` always pushes a new
+ * part and replaying over a seeded in-progress row would duplicate its text.
+ */
+export function isStreamingTail(rows: IAiChatMessageRow[]): boolean {
+  const tail = rows[rows.length - 1];
+  return !!tail && tail.role === "assistant" && tail.status === "streaming";
+}
+
+/**
+ * A SETTLED assistant tail: the last row is an assistant row whose status is
+ * anything OTHER than 'streaming'. A settled assistant tail must NEVER resume —
+ * replaying a finished run into a store that already holds its message duplicates
+ * parts (`text-start` always pushes a new part).
+ */
+export function isSettledAssistantTail(rows: IAiChatMessageRow[]): boolean {
+  const tail = rows[rows.length - 1];
+  return !!tail && tail.role === "assistant" && tail.status !== "streaming";
+}
+
+/**
+ * Seed rows for `useChat`: return the rows unchanged, or without the last row when
+ * `strip` is set (the streaming tail is stripped so the live replay rebuilds it
+ * without duplicating parts).
+ */
+export function seedRows(
+  rows: IAiChatMessageRow[],
+  strip: boolean,
+): IAiChatMessageRow[] {
+  return strip ? rows.slice(0, -1) : rows;
+}
+
+/**
+ * Merge an assistant message into the rendered list by id: replace the message
+ * with the same id in place (the in-progress assistant row is already seeded from
+ * history, so per-step growth replaces it), or append it when absent. Returns a
+ * new array; the input is never mutated.
+ */
+export function mergeById(
+  messages: UIMessage[],
+  incoming: UIMessage | null | undefined,
+): UIMessage[] {
+  if (!incoming) return messages;
+  const idx = messages.findIndex((m) => m.id === incoming.id);
+  if (idx === -1) return [...messages, incoming];
+  const next = messages.slice();
+  next[idx] = incoming;
+  return next;
+}

apps/client/src/features/ai-chat/utils/run-polling.test.ts

@@ -1,303 +0,0 @@
-import { describe, it, expect } from "vitest";
-import type { UIMessage } from "@ai-sdk/react";
-import type { IAiChatRun } from "@/features/ai-chat/types/ai-chat.types.ts";
-import {
-  RUN_POLL_INTERVAL_MS,
-  isRunActive,
-  runPollInterval,
-  shouldObserveRun,
-  shouldClearStoppingLatch,
-  shouldClearLatchOnQueryError,
-  mergeObservedMessage,
-} from "./run-polling.ts";
-
-function makeRun(status: string): IAiChatRun {
-  return { id: "run-1", chatId: "c1", status };
-}
-
-function makeMsg(id: string, text: string): UIMessage {
-  return {
-    id,
-    role: "assistant",
-    parts: [{ type: "text", text }],
-  } as UIMessage;
-}
-
-describe("isRunActive", () => {
-  it("treats pending and running as active", () => {
-    expect(isRunActive(makeRun("pending"))).toBe(true);
-    expect(isRunActive(makeRun("running"))).toBe(true);
-  });
-
-  it("treats terminal / unknown / nullish as not active", () => {
-    expect(isRunActive(makeRun("succeeded"))).toBe(false);
-    expect(isRunActive(makeRun("failed"))).toBe(false);
-    expect(isRunActive(makeRun("aborted"))).toBe(false);
-    expect(isRunActive(makeRun("weird-future-status"))).toBe(false);
-    expect(isRunActive(null)).toBe(false);
-    expect(isRunActive(undefined)).toBe(false);
-  });
-});
-
-describe("runPollInterval (the refetchInterval helper)", () => {
-  it("returns 2000ms while the run is pending/running", () => {
-    expect(runPollInterval(makeRun("pending"))).toBe(RUN_POLL_INTERVAL_MS);
-    expect(runPollInterval(makeRun("running"))).toBe(RUN_POLL_INTERVAL_MS);
-    expect(RUN_POLL_INTERVAL_MS).toBe(2000);
-  });
-
-  it("returns false (stop polling) once the run is terminal", () => {
-    expect(runPollInterval(makeRun("succeeded"))).toBe(false);
-    expect(runPollInterval(makeRun("failed"))).toBe(false);
-    expect(runPollInterval(makeRun("aborted"))).toBe(false);
-  });
-
-  it("returns false (no polling) when there is no run", () => {
-    expect(runPollInterval(null)).toBe(false);
-    expect(runPollInterval(undefined)).toBe(false);
-  });
-});
-
-describe("shouldObserveRun (observer-vs-streamer decision)", () => {
-  it("observes an active run when this tab is NOT the local streamer", () => {
-    expect(shouldObserveRun(makeRun("running"), false)).toBe(true);
-    expect(shouldObserveRun(makeRun("pending"), false)).toBe(true);
-  });
-
-  it("observes a terminal run too (so the final output shows on reopen)", () => {
-    expect(shouldObserveRun(makeRun("succeeded"), false)).toBe(true);
-  });
-
-  it("does NOT observe when this tab IS the streamer (no double-render)", () => {
-    expect(shouldObserveRun(makeRun("running"), true)).toBe(false);
-    expect(shouldObserveRun(makeRun("succeeded"), true)).toBe(false);
-  });
-
-  it("does NOT observe when there is no run", () => {
-    expect(shouldObserveRun(null, false)).toBe(false);
-    expect(shouldObserveRun(undefined, false)).toBe(false);
-  });
-});
-
-describe("shouldClearStoppingLatch (#234 latch-release decision)", () => {
-  // The one case the latch SHOULD clear: we requested a stop, we are the passive
-  // observer (not streaming), and the CURRENT run is terminal.
-  it("clears only when stopping, observing, and the run is terminal", () => {
-    expect(
-      shouldClearStoppingLatch({
-        stoppingRun: true,
-        run: makeRun("aborted"),
-        isLocalStreaming: false,
-      }),
-    ).toBe(true);
-    expect(
-      shouldClearStoppingLatch({
-        stoppingRun: true,
-        run: makeRun("succeeded"),
-        isLocalStreaming: false,
-      }),
-    ).toBe(true);
-    expect(
-      shouldClearStoppingLatch({
-        stoppingRun: true,
-        run: makeRun("failed"),
-        isLocalStreaming: false,
-      }),
-    ).toBe(true);
-  });
-
-  // Round-3 regression: clearing while THIS tab is still the local streamer would
-  // re-open the flash for the current turn the moment we switch to observer role.
-  // A predicate lacking the streaming gate would (wrongly) return true here.
-  it("does NOT clear while this tab is the local streamer", () => {
-    expect(
-      shouldClearStoppingLatch({
-        stoppingRun: true,
-        run: makeRun("aborted"),
-        isLocalStreaming: true,
-      }),
-    ).toBe(false);
-    expect(
-      shouldClearStoppingLatch({
-        stoppingRun: true,
-        run: makeRun("succeeded"),
-        isLocalStreaming: true,
-      }),
-    ).toBe(false);
-  });
-
-  // The detached run keeps growing after a local abort — while it is still
-  // active the latch MUST hold so the observer merge stays suppressed.
-  it("does NOT clear while the run is still active", () => {
-    expect(
-      shouldClearStoppingLatch({
-        stoppingRun: true,
-        run: makeRun("running"),
-        isLocalStreaming: false,
-      }),
-    ).toBe(false);
-    expect(
-      shouldClearStoppingLatch({
-        stoppingRun: true,
-        run: makeRun("pending"),
-        isLocalStreaming: false,
-      }),
-    ).toBe(false);
-  });
-
-  // #234 F4: on Stop the stale PREVIOUS-turn run is removed from the cache, so the
-  // observed `run` is null until the current turn's run is fetched fresh. A null
-  // run HOLDS the latch — it can never clear against the just-removed stale run,
-  // only against the current turn's own terminal run once observed.
-  it("does NOT clear against a removed/absent run (F4 stale-run guard)", () => {
-    expect(
-      shouldClearStoppingLatch({
-        stoppingRun: true,
-        run: null,
-        isLocalStreaming: false,
-      }),
-    ).toBe(false);
-    expect(
-      shouldClearStoppingLatch({
-        stoppingRun: true,
-        run: undefined,
-        isLocalStreaming: false,
-      }),
-    ).toBe(false);
-  });
-
-  it("does NOT clear when no stop was requested", () => {
-    expect(
-      shouldClearStoppingLatch({
-        stoppingRun: false,
-        run: makeRun("aborted"),
-        isLocalStreaming: false,
-      }),
-    ).toBe(false);
-  });
-});
-
-describe("shouldClearLatchOnQueryError (#234 F7 error-safety-net decision)", () => {
-  // This guards the REAL anti-flash decision the component's run-query-error
-  // safety-net effect uses (ai-chat-window.tsx wires the effect to THIS helper,
-  // not a copy — so the test is non-vacuous vs the live code).
-
-  // (b) The F7 hole: a TRANSIENT run-query error while `run` is STILL ACTIVE must
-  // NOT clear the latch. TanStack Query v5 retains `data` on error, so
-  // runQueryFailed can be true while the held run is still pending/running.
-  // Against the PRE-F7 condition (without `!isRunActive(run)`) this would return
-  // true — so this assertion fails on the buggy code (non-vacuous).
-  it("does NOT clear on a transient error while the run is still ACTIVE (F7)", () => {
-    expect(
-      shouldClearLatchOnQueryError({
-        stoppingRun: true,
-        isLocalStreaming: false,
-        runQueryFailed: true,
-        run: makeRun("running"),
-      }),
-    ).toBe(false);
-    expect(
-      shouldClearLatchOnQueryError({
-        stoppingRun: true,
-        isLocalStreaming: false,
-        runQueryFailed: true,
-        run: makeRun("pending"),
-      }),
-    ).toBe(false);
-  });
-
-  // (a) The genuine permanent-null-freeze: run cache cleared by removeQueries +
-  // the refetch keeps ERRORING, so `run === null`. This is the ONLY case the
-  // safety-net exists to cure — it MUST clear so the frozen view resumes.
-  it("clears on a permanent error when the run is null (permanent-null-freeze)", () => {
-    expect(
-      shouldClearLatchOnQueryError({
-        stoppingRun: true,
-        isLocalStreaming: false,
-        runQueryFailed: true,
-        run: null,
-      }),
-    ).toBe(true);
-    expect(
-      shouldClearLatchOnQueryError({
-        stoppingRun: true,
-        isLocalStreaming: false,
-        runQueryFailed: true,
-        run: undefined,
-      }),
-    ).toBe(true);
-  });
-
-  // A TERMINAL run also satisfies `!isRunActive`; clearing then is harmless — the
-  // terminal effect (shouldClearStoppingLatch) already clears for a terminal run,
-  // so this only ever agrees with it. Asserted so the (c) reasoning is pinned.
-  it("clears on an error when the run is terminal (harmless, agrees with terminal effect)", () => {
-    expect(
-      shouldClearLatchOnQueryError({
-        stoppingRun: true,
-        isLocalStreaming: false,
-        runQueryFailed: true,
-        run: makeRun("aborted"),
-      }),
-    ).toBe(true);
-  });
-
-  it("does NOT clear without an actual query error", () => {
-    expect(
-      shouldClearLatchOnQueryError({
-        stoppingRun: true,
-        isLocalStreaming: false,
-        runQueryFailed: false,
-        run: null,
-      }),
-    ).toBe(false);
-  });
-
-  it("does NOT clear while this tab is the local streamer", () => {
-    expect(
-      shouldClearLatchOnQueryError({
-        stoppingRun: true,
-        isLocalStreaming: true,
-        runQueryFailed: true,
-        run: null,
-      }),
-    ).toBe(false);
-  });
-
-  it("does NOT clear when no stop was requested", () => {
-    expect(
-      shouldClearLatchOnQueryError({
-        stoppingRun: false,
-        isLocalStreaming: false,
-        runQueryFailed: true,
-        run: null,
-      }),
-    ).toBe(false);
-  });
-});
-
-describe("mergeObservedMessage", () => {
-  it("replaces the message with the same id in place (per-step growth)", () => {
-    const prev = [makeMsg("u1", "hi"), makeMsg("a1", "step 1")];
-    const observed = makeMsg("a1", "step 1\nstep 2");
-    const next = mergeObservedMessage(prev, observed);
-    expect(next).toHaveLength(2);
-    expect(next[1]).toBe(observed);
-    expect(next[0]).toBe(prev[0]); // untouched
-    expect(next).not.toBe(prev); // new array (never mutates input)
-  });
-
-  it("appends when the observed message is not yet present", () => {
-    const prev = [makeMsg("u1", "hi")];
-    const observed = makeMsg("a1", "first token");
-    const next = mergeObservedMessage(prev, observed);
-    expect(next).toHaveLength(2);
-    expect(next[1]).toBe(observed);
-  });
-
-  it("returns the original list unchanged when there is nothing to merge", () => {
-    const prev = [makeMsg("u1", "hi")];
-    expect(mergeObservedMessage(prev, null)).toBe(prev);
-    expect(mergeObservedMessage(prev, undefined)).toBe(prev);
-  });
-});

apps/client/src/features/ai-chat/utils/run-polling.ts

@@ -1,151 +0,0 @@
-import type { UIMessage } from "@ai-sdk/react";
-import type { IAiChatRun } from "@/features/ai-chat/types/ai-chat.types.ts";
-
-/**
- * Reconnect-and-live-follow helpers (#184). When a chat is reopened while its
- * agent run is STILL going, this tab is a PASSIVE OBSERVER: it did not start the
- * run here (no local SSE stream), so it catches up by POLLING the reconnect
- * endpoint (`POST /ai-chat/run`) and merging the run's incrementally-persisted
- * assistant message into the rendered thread. These are the small pure decisions
- * that machinery hangs off, extracted so they can be unit-tested in isolation
- * (mirrors how reindex polling / editor-sync-state are tested).
- */
-
-/** How often to re-poll the reconnect endpoint while a run is ACTIVE. */
-export const RUN_POLL_INTERVAL_MS = 2000;
-
-// 'pending' and 'running' are the two ACTIVE statuses; 'succeeded' | 'failed' |
-// 'aborted' are TERMINAL (and any unknown future status is treated as terminal,
-// so a stale/odd value never polls forever).
-const ACTIVE_STATUSES = new Set(["pending", "running"]);
-
-/** Whether a run is still going (worth polling / merging live updates from). */
-export function isRunActive(run: IAiChatRun | null | undefined): boolean {
-  return !!run && ACTIVE_STATUSES.has(run.status);
-}
-
-/**
- * The TanStack Query `refetchInterval` value for the run query: poll every
- * {@link RUN_POLL_INTERVAL_MS} while the run is active, and `false` (stop) once
- * it is terminal or there is no run. Polling is thus naturally bounded by the run
- * reaching a terminal status — no separate timeout cap is needed.
- */
-export function runPollInterval(
-  run: IAiChatRun | null | undefined,
-): number | false {
-  return isRunActive(run) ? RUN_POLL_INTERVAL_MS : false;
-}
-
-/**
- * Observer-vs-streamer decision. We render the polled run message (catch up +
- * keep advancing) ONLY when this tab is a passive observer: there IS a run AND
- * this tab is NOT the one locally streaming it (we reconnected, we didn't start
- * it here). When this tab is the streamer, the live SSE stream owns the view, so
- * we neither poll nor merge — avoiding a double-render fight. Terminal runs still
- * merge (so the final persisted output is shown on reopen); the poll itself is
- * stopped separately by {@link runPollInterval}.
- */
-export function shouldObserveRun(
-  run: IAiChatRun | null | undefined,
-  localStreaming: boolean,
-): boolean {
-  return !!run && !localStreaming;
-}
-
-/**
- * Should the "stopping" latch — which suppresses the observer re-stream flash
- * after the user pressed Stop — be RELEASED now? All three must hold:
- *  - `stoppingRun`: we actually requested a stop (otherwise nothing to release);
- *  - `!isLocalStreaming`: this tab is NOT the local streamer. While we are the
- *    streamer the run query is disabled, so the observed `run` is not the run we
- *    are following — releasing the latch then would re-open the flash for the
- *    current turn the instant we switch to observer role;
- *  - the observed `run` EXISTS and has reached a TERMINAL status.
- *
- * The null / still-active `run` case is the #234 F4 invariant. On Stop the stale
- * PREVIOUS-turn run is removed from the query cache (`removeQueries`), so `run`
- * is null until the CURRENT turn's run is re-fetched fresh; a null or active run
- * therefore HOLDS the latch, so it can only ever clear against the current turn's
- * OWN terminal run — never a stale cached one. (The cache removal itself is
- * integration-level in AiChatWindow; this predicate encodes the decision given
- * whatever run is currently observed, and a stale terminal run is
- * indistinguishable from a current terminal run at the predicate level — hence
- * the cache removal is what guarantees only the current run is ever passed here.)
- */
-export function shouldClearStoppingLatch(args: {
-  stoppingRun: boolean;
-  run: IAiChatRun | null | undefined;
-  isLocalStreaming: boolean;
-}): boolean {
-  const { stoppingRun, run, isLocalStreaming } = args;
-  if (!stoppingRun || isLocalStreaming) return false;
-  return !!run && !isRunActive(run);
-}
-
-/**
- * Should the "stopping" latch be RELEASED by the run-query ERROR safety-net?
- * (#234 F7 — a NEW path of the same re-stream flash the F4 latch exists to
- * prevent.) After Stop, `handleServerStop` clears the run cache; the terminal
- * effect then holds the latch via `if (!run) return` until the CURRENT turn's run
- * is fetched fresh. If that refetch instead ERRORS permanently, `run` stays null,
- * its status-keyed refetchInterval is off, and nothing would ever observe a
- * terminal run — freezing the view with the observer merge suppressed. This
- * safety-net cures ONLY that genuine permanent-null-freeze.
- *
- * All four must hold:
- *  - `stoppingRun`: we actually requested a stop (otherwise nothing to release);
- *  - `!isLocalStreaming`: this tab is NOT the local streamer (same reason as
- *    {@link shouldClearStoppingLatch});
- *  - `runQueryFailed`: the run query is in its error state (TanStack Query v5 with
- *    retry:false — isError);
- *  - `!isRunActive(run)`: the observed `run` is NOT an active (pending/running)
- *    held run. This is the F7 gate. In TanStack Query v5 the query's `data` is
- *    RETAINED on error, so `runQueryFailed` can be true while `run` is STILL an
- *    ACTIVE run (a single transient GET-run failure in the window between Stop and
- *    settle). Without this gate a transient error would release the latch early —
- *    re-opening the observer merge and flashing the growing detached run over the
- *    frozen row (exactly the F4 flash). Gating on the run NOT being active means we
- *    only ever cure the permanent-null-freeze (`run === null`, so
- *    `isRunActive(null)` is false), never release against an active run.
- *
- * (A terminal `run` also satisfies `!isRunActive(run)`; clearing then is harmless
- * — the terminal effect's {@link shouldClearStoppingLatch} already clears the
- * latch for a terminal run, so this only ever agrees with it, never conflicts.)
- *
- * INVARIANT (do not break): clearing the latch on the `run === null` branch is safe
- * ONLY because the run query's `refetchInterval` (see {@link runPollInterval}) stops
- * polling when the data is empty — so after we clear on null+error there is no
- * subsequent auto-poll that could return a still-active detached run and re-open the
- * merge. If `refetchInterval` is ever changed to keep polling on `run === null`/on
- * error, this null-branch clear would re-open the F7 flash through the null path.
- * Do not change the run query's refetchInterval without re-checking this path.
- */
-export function shouldClearLatchOnQueryError(args: {
-  stoppingRun: boolean;
-  isLocalStreaming: boolean;
-  runQueryFailed: boolean;
-  run: IAiChatRun | null | undefined;
-}): boolean {
-  const { stoppingRun, isLocalStreaming, runQueryFailed, run } = args;
-  return (
-    stoppingRun && !isLocalStreaming && runQueryFailed && !isRunActive(run)
-  );
-}
-
-/**
- * Merge an observed assistant message into the rendered list: replace the message
- * with the same id in place (the in-progress assistant row is already seeded from
- * history, so per-step growth replaces it), or append it when absent. Returns a
- * new array; the input is never mutated.
- */
-export function mergeObservedMessage(
-  messages: UIMessage[],
-  observed: UIMessage | null | undefined,
-): UIMessage[] {
-  if (!observed) return messages;
-  const idx = messages.findIndex((m) => m.id === observed.id);
-  if (idx === -1) return [...messages, observed];
-  const next = messages.slice();
-  next[idx] = observed;
-  return next;
-}

apps/server/src/core/ai-chat/ai-chat-stream-registry.service.ts

@@ -0,0 +1,329 @@
+import { Injectable, Logger, OnModuleDestroy } from '@nestjs/common';
+
+/**
+ * In-memory run-stream registry (#184 phase 1.5). A durable agent run tees its
+ * SSE frames here (via `pipeUIMessageStreamToResponse({ consumeSseStream })`)
+ * so a LATE tab — one that reloaded, or opened after the starter dropped — can
+ * attach through `GET /ai-chat/runs/:chatId/stream`, replay the frames buffered
+ * so far, and then follow the live tail as a normal streamer.
+ *
+ * This is deliberately single-process and best-effort: it holds nothing the DB
+ * does not (the run + assistant row are the source of truth), so a process
+ * restart simply drops in-flight entries and the client falls back to its
+ * restore + degraded-poll path. The async `attach` return type is the seam for a
+ * future phase-2 cross-process backend (Redis) — the interface does not change.
+ */
+
+/** How long a finished entry is retained for late attach (replay + immediate end). */
+export const RUN_STREAM_RETAIN_FINISHED_MS = 30_000;
+
+/** Per-run replay buffer cap. Past this the buffer is dropped (attach -> 204). */
+export const RUN_STREAM_MAX_BUFFER_BYTES = 4 * 1024 * 1024;
+
+// 2x the replay cap: a just-written 4MB replay burst alone can never trip the
+// per-subscriber cap (see controller); only a genuinely stalled socket can.
+export const SUBSCRIBER_MAX_BUFFERED_BYTES = 2 * RUN_STREAM_MAX_BUFFER_BYTES;
+
+export interface RunStreamCallbacks {
+  onFrame: (frame: string) => void;
+  onEnd: () => void;
+}
+
+export interface RunStreamAttachment {
+  replay: string[];
+  finished: boolean;
+  start(): void; // drain pending frames (order preserved) and go live
+  unsubscribe(): void; // safe to call at any point, idempotent
+}
+
+interface Subscriber extends RunStreamCallbacks {
+  started: boolean;
+  pending: string[];
+  // Byte size of `pending`, capped at SUBSCRIBER_MAX_BUFFERED_BYTES. `start()` is
+  // called in the SAME tick as `attach()` today (see attach), so `pending` never
+  // holds more than one microtask of frames — but the async `attach` signature is
+  // a phase-2 seam: an await between attach and start would let a stalled paused
+  // subscriber buffer the WHOLE run here. The cap is the structural backstop.
+  pendingBytes: number;
+  overflowed: boolean;
+  pendingEnd: boolean;
+}
+
+interface Entry {
+  runId: string;
+  // The persisted assistant row id of this run (set at bind; undefined if the
+  // seed failed). Used by the attach anchor check (invariant 6).
+  assistantMessageId?: string;
+  frames: string[];
+  bytes: number;
+  overflowed: boolean;
+  finished: boolean;
+  subscribers: Set<Subscriber>;
+  retainTimer?: NodeJS.Timeout;
+}
+
+@Injectable()
+export class AiChatStreamRegistryService implements OnModuleDestroy {
+  private readonly logger = new Logger(AiChatStreamRegistryService.name);
+  private readonly entries = new Map<string, Entry>(); // key: chatId
+
+  /**
+   * Register a fresh entry at the START of a run (before any frame), so a tab
+   * that attaches in the begin->seed window finds an entry to wait on. If an
+   * entry already exists for this chat (a previous, possibly still-live run whose
+   * tee loop is draining), it is terminated MIRRORING the done-path (invariant 3)
+   * so its subscribers are released and its retention timer is cleared; a late
+   * `done` from that old tee then fires against the closed-over old reference and,
+   * thanks to identity checks, never touches this new entry.
+   */
+  open(chatId: string, runId: string): void {
+    const existing = this.entries.get(chatId);
+    if (existing) {
+      if (existing.retainTimer) {
+        clearTimeout(existing.retainTimer);
+        existing.retainTimer = undefined;
+      }
+      // Started subscribers get exactly one onEnd() and are removed; paused ones
+      // are marked pendingEnd (their start() will end them). finished=true guards
+      // any later done from the old tee loop from double-notifying.
+      this.terminateSubscribers(existing);
+    }
+    this.entries.set(chatId, {
+      runId,
+      frames: [],
+      bytes: 0,
+      overflowed: false,
+      finished: false,
+      subscribers: new Set<Subscriber>(),
+    });
+  }
+
+  /**
+   * Tee a run's SSE frame stream into its entry (called from consumeSseStream).
+   * No-op with a warning when there is no entry or the entry belongs to a
+   * different run (invariant 1). The reader loop is fire-and-forget: the tee
+   * branch outlives the client socket by design.
+   */
+  bind(
+    chatId: string,
+    runId: string,
+    assistantMessageId: string | undefined,
+    stream: ReadableStream<string>,
+  ): void {
+    const entry = this.entries.get(chatId);
+    if (!entry || entry.runId !== runId) {
+      // Invariant 1: only the matching run may mutate the entry.
+      this.logger.warn(
+        `bind: no matching run-stream entry for chat=${chatId} run=${runId}`,
+      );
+      return;
+    }
+    entry.assistantMessageId = assistantMessageId;
+    const reader = stream.getReader();
+    const pump = async (): Promise<void> => {
+      try {
+        for (;;) {
+          const { done, value } = await reader.read();
+          if (done) break;
+          this.ingestFrame(entry, value);
+        }
+        this.finalizeEntry(chatId, entry);
+      } catch {
+        // A read error is a terminal event too — release subscribers.
+        this.finalizeEntry(chatId, entry);
+      }
+    };
+    void pump();
+  }
+
+  /**
+   * Terminate a run's entry from the OUTER catch of the stream method (a failure
+   * before/while wiring the pipe, so `done` will never arrive). Identity-checked
+   * on runId (invariant 1); the shared terminal path is idempotent.
+   */
+  abortEntry(chatId: string, runId: string): void {
+    const entry = this.entries.get(chatId);
+    if (!entry || entry.runId !== runId) return;
+    this.finalizeEntry(chatId, entry);
+  }
+
+  /**
+   * Attach to a run's stream. Async only for the phase-2 Redis seam — the body
+   * runs synchronously so the replay snapshot and the subscriber registration
+   * happen in ONE tick with no await between them (invariant 4): a frame ingested
+   * concurrently cannot slip into the gap and be lost or duplicated.
+   *
+   * Returns null (-> the caller answers 204) when:
+   *  - there is no entry, or it overflowed (replay is gone);
+   *  - expect=live with an anchor that does not match this run's assistant id
+   *    (invariant 6: a stripped tab must never replay a FOREIGN run's transcript);
+   *  - the run finished and the caller did not expect a live tail.
+   * A finished run with expect=live yields a replay-only attachment (no
+   * subscriber registered). Otherwise a paused subscriber is registered and the
+   * caller replays `replay`, then calls start() to drain and go live.
+   */
+  async attach(
+    chatId: string,
+    expectLive: boolean,
+    anchor: string | undefined,
+    cb: RunStreamCallbacks,
+  ): Promise<RunStreamAttachment | null> {
+    const entry = this.entries.get(chatId);
+    if (!entry || entry.overflowed) return null;
+    // Invariant 6: cross-run replay is forbidden. Before bind, assistantMessageId
+    // is undefined and mismatches any anchor -> 204 -> client restore+poll path.
+    if (expectLive && anchor && entry.assistantMessageId !== anchor) return null;
+    if (entry.finished && !expectLive) return null;
+    if (entry.finished && expectLive) {
+      // Replay-only: the run is done, no subscriber is registered.
+      return {
+        replay: entry.frames.slice(),
+        finished: true,
+        start: () => undefined,
+        unsubscribe: () => undefined,
+      };
+    }
+
+    const sub: Subscriber = {
+      onFrame: cb.onFrame,
+      onEnd: cb.onEnd,
+      started: false,
+      pending: [],
+      pendingBytes: 0,
+      overflowed: false,
+      pendingEnd: false,
+    };
+    entry.subscribers.add(sub);
+    // Snapshot in the SAME synchronous block as the registration (invariant 4).
+    const replay = entry.frames.slice();
+    // CONTRACT: the caller MUST call start() in the SAME tick as this attach()
+    // returns — no await between them. While a subscriber is paused, every frame
+    // is buffered in sub.pending; a delayed start() lets a whole run accumulate
+    // there. The pendingBytes cap (see ingestFrame) is the structural backstop if
+    // that contract is ever broken (e.g. the phase-2 Redis await seam).
+    return {
+      replay,
+      finished: false,
+      start: () => {
+        if (sub.overflowed) {
+          // The pending buffer overflowed while paused: end the stream instead of
+          // replaying a partial (a 204-equivalent post-attach degrade).
+          try {
+            sub.onEnd();
+          } catch {
+            // The socket is gone; nothing to end.
+          }
+          entry.subscribers.delete(sub);
+          return;
+        }
+        // Deliver frames buffered while paused, in order, then go live.
+        for (const frame of sub.pending) {
+          try {
+            sub.onFrame(frame);
+          } catch {
+            entry.subscribers.delete(sub);
+            return;
+          }
+        }
+        sub.pending = [];
+        sub.started = true;
+        if (sub.pendingEnd) {
+          try {
+            sub.onEnd();
+          } catch {
+            // The socket is gone; nothing to end.
+          }
+          entry.subscribers.delete(sub);
+        }
+      },
+      unsubscribe: () => {
+        entry.subscribers.delete(sub);
+      },
+    };
+  }
+
+  onModuleDestroy(): void {
+    for (const entry of this.entries.values()) {
+      if (entry.retainTimer) clearTimeout(entry.retainTimer);
+    }
+    this.entries.clear();
+  }
+
+  /** Buffer + fan-out a single frame. See invariant/overflow semantics inline. */
+  private ingestFrame(entry: Entry, frame: string): void {
+    entry.bytes += Buffer.byteLength(frame);
+    if (!entry.overflowed) {
+      entry.frames.push(frame);
+      if (entry.bytes > RUN_STREAM_MAX_BUFFER_BYTES) {
+        // The crossing frame was already counted AND (below) fanned out; only the
+        // replay buffer is dropped. After overflow no more frames are buffered,
+        // but live fan-out continues.
+        entry.overflowed = true;
+        entry.frames = [];
+        this.logger.warn(
+          `run-stream buffer overflow for run=${entry.runId}; ` +
+            `late attach will 204 until the run ends`,
+        );
+      }
+    }
+    for (const sub of entry.subscribers) {
+      if (sub.started) {
+        try {
+          sub.onFrame(frame);
+        } catch {
+          entry.subscribers.delete(sub);
+        }
+      } else {
+        sub.pending.push(frame);
+        sub.pendingBytes += Buffer.byteLength(frame);
+        if (sub.pendingBytes > SUBSCRIBER_MAX_BUFFERED_BYTES) {
+          // The paused subscriber's buffer overflowed — only possible if start()
+          // was delayed past the same-tick contract (the phase-2 await seam).
+          // Drop it rather than buffer the whole run; on start() it degrades to an
+          // immediate end (a 204-equivalent) instead of replaying a partial.
+          sub.overflowed = true;
+          sub.pending = [];
+          entry.subscribers.delete(sub);
+        }
+      }
+    }
+  }
+
+  /**
+   * Shared terminal path for done / read-error / external-abort. Idempotent: a
+   * second call (already finished) is a no-op, so an open()-replaced or
+   * abort-then-done entry is never double-armed or double-ended.
+   */
+  private finalizeEntry(chatId: string, entry: Entry): void {
+    if (entry.finished) return;
+    this.terminateSubscribers(entry);
+    const timer = setTimeout(() => {
+      // Invariant 2: only delete OUR entry (a replacement may already own the key).
+      if (this.entries.get(chatId) === entry) this.entries.delete(chatId);
+    }, RUN_STREAM_RETAIN_FINISHED_MS);
+    timer.unref?.();
+    entry.retainTimer = timer;
+  }
+
+  /**
+   * Mark the entry finished and release its subscribers, mirroring the done-path:
+   * started subscribers get exactly one onEnd() and are removed; paused ones are
+   * flagged pendingEnd so their start() ends them. Deleting the current element
+   * during Set iteration is safe.
+   */
+  private terminateSubscribers(entry: Entry): void {
+    entry.finished = true;
+    for (const sub of entry.subscribers) {
+      if (sub.started) {
+        try {
+          sub.onEnd();
+        } catch {
+          // The socket is gone; nothing to end.
+        }
+        entry.subscribers.delete(sub);
+      } else {
+        sub.pendingEnd = true;
+      }
+    }
+  }
+}

apps/server/src/core/ai-chat/ai-chat-stream-registry.spec.ts

@@ -0,0 +1,388 @@
+import {
+  AiChatStreamRegistryService,
+  RUN_STREAM_MAX_BUFFER_BYTES,
+  RUN_STREAM_RETAIN_FINISHED_MS,
+  RunStreamCallbacks,
+} from './ai-chat-stream-registry.service';
+
+/**
+ * Unit tests for the in-memory run-stream registry (#184 phase 1.5). The registry
+ * is the whole of the resumable-transport contract: replay ordering, paused ->
+ * live hand-off, overflow, retention, the anchor check (invariant 6), and the
+ * mirror-the-done-path replace semantics (invariant 3). Every enumerated case in
+ * the issue's task 1.5 has a test here.
+ */
+
+// A ReadableStream whose frames the test pushes explicitly, plus close/error.
+function makePushStream(): {
+  stream: ReadableStream<string>;
+  push: (f: string) => void;
+  close: () => void;
+  error: (e?: unknown) => void;
+} {
+  let controller!: ReadableStreamDefaultController<string>;
+  const stream = new ReadableStream<string>({
+    start(c) {
+      controller = c;
+    },
+  });
+  return {
+    stream,
+    push: (f) => controller.enqueue(f),
+    close: () => controller.close(),
+    error: (e) => controller.error(e ?? new Error('read error')),
+  };
+}
+
+// Let the fire-and-forget pump drain queued frames (reader.read() resolves on a
+// macrotask boundary for an already-enqueued value).
+const flush = (): Promise<void> => new Promise((r) => setTimeout(r, 0));
+
+function collector(): {
+  cb: RunStreamCallbacks;
+  frames: string[];
+  ended: () => number;
+} {
+  const frames: string[] = [];
+  let ends = 0;
+  return {
+    frames,
+    ended: () => ends,
+    cb: {
+      onFrame: (f) => frames.push(f),
+      onEnd: () => {
+        ends += 1;
+      },
+    },
+  };
+}
+
+describe('AiChatStreamRegistryService', () => {
+  const CHAT = 'chat-1';
+  let registry: AiChatStreamRegistryService;
+
+  beforeEach(() => {
+    registry = new AiChatStreamRegistryService();
+    jest.spyOn((registry as any).logger, 'warn').mockImplementation(() => {});
+  });
+
+  afterEach(() => {
+    registry.onModuleDestroy();
+  });
+
+  it('replays frames in arrival order (live attach)', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push('a');
+    src.push('b');
+    src.push('c');
+    await flush();
+
+    const c = collector();
+    const att = await registry.attach(CHAT, false, undefined, c.cb);
+    expect(att).not.toBeNull();
+    expect(att!.replay).toEqual(['a', 'b', 'c']);
+    expect(att!.finished).toBe(false);
+  });
+
+  it('late attach gets the full prefix as replay plus the live tail', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push('a');
+    src.push('b');
+    await flush();
+
+    const c = collector();
+    const att = (await registry.attach(CHAT, false, undefined, c.cb))!;
+    expect(att.replay).toEqual(['a', 'b']);
+    att.start();
+    // Live tail arrives after start().
+    src.push('c');
+    src.push('d');
+    await flush();
+    expect(c.frames).toEqual(['c', 'd']);
+  });
+
+  it('a paused subscriber receives frames buffered during pause in order, then live (no loss/reorder)', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push('a');
+    await flush();
+
+    const c = collector();
+    // Attach (paused). Frames that arrive BEFORE start() must queue, not drop.
+    const att = (await registry.attach(CHAT, false, undefined, c.cb))!;
+    expect(att.replay).toEqual(['a']);
+    src.push('b'); // arrives while paused -> pending
+    src.push('c');
+    await flush();
+    expect(c.frames).toEqual([]); // nothing delivered yet (paused)
+    att.start(); // drains pending in order
+    expect(c.frames).toEqual(['b', 'c']);
+    src.push('d'); // now live
+    await flush();
+    expect(c.frames).toEqual(['b', 'c', 'd']);
+  });
+
+  it('a run that finishes while a subscriber is paused ends it on start()', async () => {
+    registry.open(CHAT, 'run-1');
+    const c = collector();
+    const att = (await registry.attach(CHAT, false, undefined, c.cb))!;
+    // Terminate the run while the subscriber is still paused.
+    registry.abortEntry(CHAT, 'run-1');
+    expect(c.ended()).toBe(0); // paused: not ended yet
+    att.start();
+    expect(c.ended()).toBe(1); // start() drains + ends
+  });
+
+  it('finished + expect=live returns a replay WITHOUT registering a subscriber', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push('a');
+    src.push('b');
+    src.close();
+    await flush();
+
+    const c = collector();
+    const att = (await registry.attach(CHAT, true, undefined, c.cb))!;
+    expect(att.finished).toBe(true);
+    expect(att.replay).toEqual(['a', 'b']);
+    // No subscriber registered: start()/unsubscribe are no-ops and the entry has
+    // zero subscribers.
+    const entry = (registry as any).entries.get(CHAT);
+    expect(entry.subscribers.size).toBe(0);
+    att.start();
+    expect(c.frames).toEqual([]);
+  });
+
+  it('finished WITHOUT expect=live returns null', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push('a');
+    src.close();
+    await flush();
+
+    const c = collector();
+    expect(await registry.attach(CHAT, false, undefined, c.cb)).toBeNull();
+  });
+
+  it('anchor mismatch with expect=live returns null (and null before bind sets assistantMessageId)', async () => {
+    registry.open(CHAT, 'run-1');
+    const c = collector();
+    // Before bind: assistantMessageId is undefined -> mismatches any anchor.
+    expect(
+      await registry.attach(CHAT, true, 'assist-1', c.cb),
+    ).toBeNull();
+
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push('a');
+    await flush();
+    // Wrong anchor -> null (cross-run replay forbidden, invariant 6).
+    expect(await registry.attach(CHAT, true, 'other-id', c.cb)).toBeNull();
+  });
+
+  it('matching anchor with expect=live attaches', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push('a');
+    await flush();
+
+    const c = collector();
+    const att = await registry.attach(CHAT, true, 'assist-1', c.cb);
+    expect(att).not.toBeNull();
+    expect(att!.replay).toEqual(['a']);
+  });
+
+  it('overflow: attach returns null, but the LIVE subscriber keeps receiving (incl. the crossing frame)', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+
+    // A live (started) subscriber attached before the flood.
+    const c = collector();
+    const att = (await registry.attach(CHAT, false, undefined, c.cb))!;
+    att.start();
+
+    const oneMb = 'x'.repeat(1024 * 1024);
+    // 5 x 1MB = 5MB > 4MB cap; the 5th frame is the one that crosses.
+    for (let i = 0; i < 5; i++) src.push(oneMb + i);
+    await flush();
+
+    const entry = (registry as any).entries.get(CHAT);
+    expect(entry.overflowed).toBe(true);
+    expect(entry.bytes).toBeGreaterThan(RUN_STREAM_MAX_BUFFER_BYTES);
+    // The live subscriber received ALL 5 frames, including the crossing one.
+    expect(c.frames).toHaveLength(5);
+    expect(c.frames[4]).toBe(oneMb + 4);
+
+    // A NEW attach after overflow gets null (replay buffer is gone).
+    const c2 = collector();
+    expect(await registry.attach(CHAT, false, undefined, c2.cb)).toBeNull();
+  });
+
+  it('a paused subscriber whose pending buffer overflows is dropped and ends on start(); other subscribers keep receiving', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+
+    // A: paused (start() deliberately delayed to simulate the phase-2 await seam).
+    const a = collector();
+    const attA = (await registry.attach(CHAT, false, undefined, a.cb))!;
+    // B: live (started) — its delivery must be unaffected by A's overflow.
+    const b = collector();
+    const attB = (await registry.attach(CHAT, false, undefined, b.cb))!;
+    attB.start();
+
+    const oneMb = 'x'.repeat(1024 * 1024);
+    // 9 x 1MB = 9MB > 8MB per-subscriber cap; A's pending overflows, B streams live.
+    for (let i = 0; i < 9; i++) src.push(oneMb + i);
+    await flush();
+
+    const entry = (registry as any).entries.get(CHAT);
+    // A was dropped from the subscriber set on overflow; B (started) remains.
+    expect(entry.subscribers.size).toBe(1);
+    expect(a.frames).toEqual([]); // paused + overflowed: nothing was delivered
+    // B received every frame live (delivery unaffected by A's overflow).
+    expect(b.frames).toHaveLength(9);
+
+    // A's start() (arriving late) degrades to an immediate end, not a partial replay.
+    attA.start();
+    expect(a.frames).toEqual([]);
+    expect(a.ended()).toBe(1);
+  });
+
+  it('open() over a LIVE entry ends started subscribers exactly once and a late done does not touch the new entry (invariant 3)', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+    src.push('a');
+    await flush();
+
+    const c = collector();
+    const att = (await registry.attach(CHAT, false, undefined, c.cb))!;
+    att.start(); // started subscriber on run-1
+
+    // run-2 starts on the same chat while run-1's tee is still reading.
+    registry.open(CHAT, 'run-2');
+    expect(c.ended()).toBe(1); // exactly one onEnd from the replace
+
+    const newEntry = (registry as any).entries.get(CHAT);
+    expect(newEntry.runId).toBe('run-2');
+    expect(newEntry.finished).toBe(false);
+
+    // The old tee now completes: its late done must NOT double-end nor delete the
+    // new entry.
+    src.push('b');
+    src.close();
+    await flush();
+    expect(c.ended()).toBe(1); // still exactly one
+    const still = (registry as any).entries.get(CHAT);
+    expect(still).toBe(newEntry);
+    expect(still.runId).toBe('run-2');
+  });
+
+  it('bind with a foreign runId is a no-op (invariant 1)', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'WRONG-run', 'assist-x', src.stream);
+    src.push('a');
+    await flush();
+    const entry = (registry as any).entries.get(CHAT);
+    // Frames were NOT ingested (bind bailed), assistantMessageId untouched.
+    expect(entry.frames).toEqual([]);
+    expect(entry.assistantMessageId).toBeUndefined();
+  });
+
+  it('abortEntry with a foreign runId is a no-op (invariant 1)', async () => {
+    registry.open(CHAT, 'run-1');
+    registry.abortEntry(CHAT, 'WRONG-run');
+    const entry = (registry as any).entries.get(CHAT);
+    expect(entry.finished).toBe(false);
+  });
+
+  it('a throwing onFrame ejects only that subscriber; the ingest loop stays alive', async () => {
+    registry.open(CHAT, 'run-1');
+    const src = makePushStream();
+    registry.bind(CHAT, 'run-1', 'assist-1', src.stream);
+
+    const bad = collector();
+    const badAtt = (await registry.attach(CHAT, false, undefined, {
+      onFrame: () => {
+        throw new Error('boom');
+      },
+      onEnd: bad.cb.onEnd,
+    }))!;
+    badAtt.start();
+
+    const good = collector();
+    const goodAtt = (await registry.attach(CHAT, false, undefined, good.cb))!;
+    goodAtt.start();
+
+    src.push('a'); // bad throws on this frame -> ejected
+    src.push('b'); // good still receives both
+    await flush();
+
+    const entry = (registry as any).entries.get(CHAT);
+    expect(entry.subscribers.size).toBe(1); // bad ejected, good remains
+    expect(good.frames).toEqual(['a', 'b']);
+  });
+});
+
+/**
+ * Retention + replace timer behavior. Fake timers, and entries are finalized via
+ * the synchronous abortEntry() path so no stream pump / microtask juggling is
+ * needed.
+ */
+describe('AiChatStreamRegistryService retention timers', () => {
+  const CHAT = 'chat-r';
+  let registry: AiChatStreamRegistryService;
+
+  beforeEach(() => {
+    jest.useFakeTimers();
+    registry = new AiChatStreamRegistryService();
+    jest.spyOn((registry as any).logger, 'warn').mockImplementation(() => {});
+  });
+
+  afterEach(() => {
+    registry.onModuleDestroy();
+    jest.useRealTimers();
+  });
+
+  it('a finished entry is removed after the retention window', () => {
+    registry.open(CHAT, 'run-1');
+    registry.abortEntry(CHAT, 'run-1'); // finalize -> retention armed
+    expect((registry as any).entries.get(CHAT)).toBeDefined();
+    jest.advanceTimersByTime(RUN_STREAM_RETAIN_FINISHED_MS + 1);
+    expect((registry as any).entries.get(CHAT)).toBeUndefined();
+  });
+
+  it('retention deletes ONLY its own entry (invariant 2)', () => {
+    registry.open(CHAT, 'run-1');
+    registry.abortEntry(CHAT, 'run-1'); // arm retention for entry A
+    // Simulate the race where the key was replaced without clearing A's timer.
+    const sentinel = { marker: true };
+    (registry as any).entries.set(CHAT, sentinel);
+    jest.advanceTimersByTime(RUN_STREAM_RETAIN_FINISHED_MS + 1);
+    // A's timer saw entries.get(CHAT) !== A, so it did NOT delete the successor.
+    expect((registry as any).entries.get(CHAT)).toBe(sentinel);
+  });
+
+  it('open() over a retained entry clears its timer and the successor survives', () => {
+    registry.open(CHAT, 'run-1');
+    registry.abortEntry(CHAT, 'run-1'); // retained, timer armed
+    const clearSpy = jest.spyOn(global, 'clearTimeout');
+    registry.open(CHAT, 'run-2'); // must clear run-1's retain timer
+    expect(clearSpy).toHaveBeenCalled();
+    jest.advanceTimersByTime(RUN_STREAM_RETAIN_FINISHED_MS + 1);
+    const entry = (registry as any).entries.get(CHAT);
+    expect(entry).toBeDefined();
+    expect(entry.runId).toBe('run-2');
+  });
+});

apps/server/src/core/ai-chat/ai-chat.controller.attach.spec.ts

@@ -0,0 +1,423 @@
+import { ForbiddenException } from '@nestjs/common';
+import { AiChatController } from './ai-chat.controller';
+import type {
+  RunStreamAttachment,
+  RunStreamCallbacks,
+} from './ai-chat-stream-registry.service';
+import { SUBSCRIBER_MAX_BUFFERED_BYTES } from './ai-chat-stream-registry.service';
+import type { User, Workspace } from '@docmost/db/types/entity.types';
+
+/**
+ * Wiring spec for the #184 phase 1.5 attach endpoint
+ * (`GET /ai-chat/runs/:chatId/stream`). Owner-gated via assertOwnedChat; the
+ * registry is mocked so this exercises ONLY the controller's replay/live/204/
+ * cleanup wiring against a fake raw socket. Constructor order is (aiChatService,
+ * aiChatRunService, aiChatRepo, aiChatMessageRepo, aiTranscription, pageRepo,
+ * streamRegistry, environment).
+ */
+describe('AiChatController attach endpoint (#184 phase 1.5)', () => {
+  const user = { id: 'u1' } as User;
+  const workspace = { id: 'ws1' } as Workspace;
+
+  function makeRawRes() {
+    const raw: any = {
+      writableEnded: false,
+      writableLength: 0,
+      destroyed: false,
+      written: [] as string[],
+      head: null as any,
+      write: jest.fn((f: string) => {
+        raw.written.push(f);
+        return true;
+      }),
+      writeHead: jest.fn((code: number, headers: any) => {
+        raw.head = { code, headers };
+        return raw;
+      }),
+      flushHeaders: jest.fn(),
+      end: jest.fn(() => {
+        raw.writableEnded = true;
+      }),
+      destroy: jest.fn(() => {
+        raw.destroyed = true;
+      }),
+      on: jest.fn(),
+      once: jest.fn(),
+    };
+    const res: any = {
+      raw,
+      status: jest.fn(() => res),
+      send: jest.fn(),
+      hijack: jest.fn(),
+    };
+    return { res, raw };
+  }
+
+  function makeReq(destroyed = false) {
+    const handlers: Record<string, () => void> = {};
+    const raw: any = {
+      destroyed,
+      once: jest.fn((ev: string, fn: () => void) => {
+        handlers[ev] = fn;
+      }),
+    };
+    return { req: { raw } as any, raw, fireClose: () => handlers['close']?.() };
+  }
+
+  function makeAttachment(
+    over: Partial<RunStreamAttachment> = {},
+  ): RunStreamAttachment {
+    return {
+      replay: [],
+      finished: false,
+      start: jest.fn(),
+      unsubscribe: jest.fn(),
+      ...over,
+    };
+  }
+
+  function makeController(opts: {
+    chat?: unknown;
+    attachment?: RunStreamAttachment | null;
+  }) {
+    const aiChatRepo = { findById: jest.fn().mockResolvedValue(opts.chat) };
+    let capturedCb: RunStreamCallbacks | undefined;
+    const streamRegistry = {
+      attach: jest.fn(
+        (
+          _chatId: string,
+          _live: boolean,
+          _anchor: string | undefined,
+          cb: RunStreamCallbacks,
+        ) => {
+          capturedCb = cb;
+          return Promise.resolve(
+            opts.attachment === undefined ? makeAttachment() : opts.attachment,
+          );
+        },
+      ),
+    };
+    const environment = { isAiChatResumableStreamEnabled: () => true };
+    const controller = new AiChatController(
+      {} as never, // aiChatService
+      {} as never, // aiChatRunService
+      aiChatRepo as never,
+      {} as never, // aiChatMessageRepo
+      {} as never, // aiTranscription
+      {} as never, // pageRepo
+      streamRegistry as never,
+      environment as never,
+    );
+    return {
+      controller,
+      aiChatRepo,
+      streamRegistry,
+      getCb: () => capturedCb!,
+    };
+  }
+
+  const owned = { id: 'c1', creatorId: 'u1' };
+
+  it('owner-gates: a foreign chat throws ForbiddenException and never attaches', async () => {
+    const { controller, streamRegistry } = makeController({
+      chat: { id: 'c1', creatorId: 'someone-else' },
+    });
+    const { res } = makeRawRes();
+    const { req } = makeReq();
+    await expect(
+      controller.attachRunStream(
+        'c1',
+        undefined,
+        undefined,
+        req,
+        res,
+        user,
+        workspace,
+      ),
+    ).rejects.toBeInstanceOf(ForbiddenException);
+    expect(streamRegistry.attach).not.toHaveBeenCalled();
+  });
+
+  it('answers 204 when the registry has nothing to resume (no entry / finished / anchor-mismatch)', async () => {
+    const { controller } = makeController({ chat: owned, attachment: null });
+    const { res } = makeRawRes();
+    const { req } = makeReq();
+    await controller.attachRunStream(
+      'c1',
+      undefined,
+      undefined,
+      req,
+      res,
+      user,
+      workspace,
+    );
+    expect(res.status).toHaveBeenCalledWith(204);
+    expect(res.send).toHaveBeenCalled();
+    expect(res.hijack).not.toHaveBeenCalled();
+  });
+
+  it('threads expect=live and anchor through to the registry', async () => {
+    const { controller, streamRegistry } = makeController({
+      chat: owned,
+      attachment: null,
+    });
+    const { res } = makeRawRes();
+    const { req } = makeReq();
+    await controller.attachRunStream(
+      'c1',
+      'live',
+      'anchor-1',
+      req,
+      res,
+      user,
+      workspace,
+    );
+    expect(streamRegistry.attach).toHaveBeenCalledWith(
+      'c1',
+      true,
+      'anchor-1',
+      expect.anything(),
+    );
+  });
+
+  it('passes expect=false when the query is absent', async () => {
+    const { controller, streamRegistry } = makeController({
+      chat: owned,
+      attachment: null,
+    });
+    const { res } = makeRawRes();
+    const { req } = makeReq();
+    await controller.attachRunStream(
+      'c1',
+      undefined,
+      undefined,
+      req,
+      res,
+      user,
+      workspace,
+    );
+    expect(streamRegistry.attach).toHaveBeenCalledWith(
+      'c1',
+      false,
+      undefined,
+      expect.anything(),
+    );
+  });
+
+  it('hijacks, writes headers + replay, registers a close cleanup, then goes live', async () => {
+    const start = jest.fn();
+    const attachment = makeAttachment({
+      replay: ['f1', 'f2'],
+      finished: false,
+      start,
+    });
+    const { controller } = makeController({ chat: owned, attachment });
+    const { res, raw } = makeRawRes();
+    const { req } = makeReq();
+    await controller.attachRunStream(
+      'c1',
+      undefined,
+      undefined,
+      req,
+      res,
+      user,
+      workspace,
+    );
+    expect(res.hijack).toHaveBeenCalled();
+    expect(raw.writeHead).toHaveBeenCalledWith(
+      200,
+      expect.objectContaining({ 'content-type': 'text/event-stream' }),
+    );
+    expect(raw.written).toEqual(['f1', 'f2']); // replay
+    expect(start).toHaveBeenCalled(); // go live after replay
+    expect(req.raw.once).toHaveBeenCalledWith('close', expect.any(Function));
+  });
+
+  it('finished replay ends the response immediately without going live', async () => {
+    const start = jest.fn();
+    const attachment = makeAttachment({
+      replay: ['f1'],
+      finished: true,
+      start,
+    });
+    const { controller } = makeController({ chat: owned, attachment });
+    const { res, raw } = makeRawRes();
+    const { req } = makeReq();
+    await controller.attachRunStream(
+      'c1',
+      'live',
+      'a1',
+      req,
+      res,
+      user,
+      workspace,
+    );
+    expect(raw.written).toEqual(['f1']);
+    expect(raw.end).toHaveBeenCalled();
+    expect(start).not.toHaveBeenCalled(); // finished -> returns before start()
+  });
+
+  it('a close during the awaits (req.raw.destroyed) unsubscribes and writes nothing', async () => {
+    const attachment = makeAttachment({ replay: ['f1'] });
+    const { controller } = makeController({ chat: owned, attachment });
+    const { res, raw } = makeRawRes();
+    const { req } = makeReq(true); // destroyed already at registration time
+    await controller.attachRunStream(
+      'c1',
+      undefined,
+      undefined,
+      req,
+      res,
+      user,
+      workspace,
+    );
+    expect(attachment.unsubscribe).toHaveBeenCalled();
+    expect(raw.writeHead).not.toHaveBeenCalled();
+    expect(raw.written).toEqual([]);
+  });
+
+  it('the registered close handler unsubscribes the attachment', async () => {
+    const attachment = makeAttachment({ replay: [] });
+    const { controller } = makeController({ chat: owned, attachment });
+    const { res } = makeRawRes();
+    const { req, fireClose } = makeReq();
+    await controller.attachRunStream(
+      'c1',
+      undefined,
+      undefined,
+      req,
+      res,
+      user,
+      workspace,
+    );
+    expect(attachment.unsubscribe).not.toHaveBeenCalled();
+    fireClose(); // socket closed
+    expect(attachment.unsubscribe).toHaveBeenCalled();
+  });
+
+  it('onFrame destroys the socket when the buffered length exceeds the cap', async () => {
+    const attachment = makeAttachment({ replay: [] });
+    const { controller, getCb } = makeController({ chat: owned, attachment });
+    const { res, raw } = makeRawRes();
+    const { req } = makeReq();
+    await controller.attachRunStream(
+      'c1',
+      undefined,
+      undefined,
+      req,
+      res,
+      user,
+      workspace,
+    );
+    const cb = getCb();
+    // Normal live frame writes.
+    raw.writableLength = 0;
+    cb.onFrame('live-1');
+    expect(raw.written).toContain('live-1');
+    // A stalled socket over the cap is destroyed instead of buffering.
+    raw.writableLength = SUBSCRIBER_MAX_BUFFERED_BYTES + 1;
+    raw.write.mockClear();
+    cb.onFrame('too-much');
+    expect(raw.destroy).toHaveBeenCalled();
+    expect(raw.write).not.toHaveBeenCalled();
+  });
+});
+
+/**
+ * The begin-hook `open()` flag gate (#184 phase 1.5). `open()` lives ONLY in the
+ * stream() begin-hook, gated on the resumable flag. If it regressed, a flag-off
+ * turn would create an EMPTY registry entry (never bound, never finished) and a
+ * later attach would find a non-null paused attachment -> a hung SSE that never
+ * gets a frame and never ends, instead of a clean 204. These drive stream() only
+ * far enough to capture the runHooks it hands to the service, then invoke the
+ * begin-hook and assert whether the registry was opened.
+ */
+describe('AiChatController begin-hook open() flag gate (#184 phase 1.5)', () => {
+  const user = { id: 'u1' } as User;
+  const workspace = {
+    id: 'ws1',
+    settings: { ai: { chat: true, autonomousRuns: true } },
+  } as unknown as Workspace;
+
+  function makeController(opts: { resumable: boolean }) {
+    let capturedArgs: any;
+    const aiChatService = {
+      resolveRoleForRequest: jest.fn(async () => null),
+      getChatModel: jest.fn(async () => ({})),
+      stream: jest.fn(async (args: any) => {
+        capturedArgs = args;
+      }),
+    };
+    const aiChatRunService = {
+      getActiveForChat: jest.fn(async () => undefined),
+      beginRun: jest.fn(async () => ({
+        runId: 'run-1',
+        signal: new AbortController().signal,
+      })),
+    };
+    const streamRegistry = { open: jest.fn(), attach: jest.fn() };
+    const environment = {
+      isAiChatResumableStreamEnabled: () => opts.resumable,
+    };
+    const controller = new AiChatController(
+      aiChatService as never,
+      aiChatRunService as never,
+      {} as never, // aiChatRepo
+      {} as never, // aiChatMessageRepo
+      {} as never, // aiTranscription
+      {} as never, // pageRepo
+      streamRegistry as never,
+      environment as never,
+    );
+    return {
+      controller,
+      streamRegistry,
+      aiChatRunService,
+      getRunHooks: () => capturedArgs?.runHooks,
+    };
+  }
+
+  function makeReqRes() {
+    const req: any = {
+      raw: { sessionId: 'sess-1', once: jest.fn() },
+      body: {
+        messages: [
+          { id: 'm1', role: 'user', parts: [{ type: 'text', text: 'hi' }] },
+        ],
+      },
+    };
+    const res: any = {
+      raw: { once: jest.fn(), on: jest.fn(), headersSent: false, writableEnded: false },
+      hijack: jest.fn(),
+    };
+    return { req, res };
+  }
+
+  it('flag OFF: the begin-hook does NOT open a registry entry (no hung empty entry)', async () => {
+    const { controller, streamRegistry, aiChatRunService, getRunHooks } =
+      makeController({ resumable: false });
+    const { req, res } = makeReqRes();
+    await controller.stream(req, res, user, workspace);
+
+    const runHooks = getRunHooks();
+    expect(runHooks).toBeDefined();
+    const handle = await runHooks.begin('chat-1');
+    // The run still begins (the durable-run feature is independent of resume)...
+    expect(aiChatRunService.beginRun).toHaveBeenCalled();
+    expect(handle).toEqual({ runId: 'run-1', signal: expect.anything() });
+    // ...but with the flag off the registry entry is NEVER opened.
+    expect(streamRegistry.open).not.toHaveBeenCalled();
+  });
+
+  it('flag ON: the begin-hook opens the registry entry with (chatId, runId)', async () => {
+    const { controller, streamRegistry, getRunHooks } = makeController({
+      resumable: true,
+    });
+    const { req, res } = makeReqRes();
+    await controller.stream(req, res, user, workspace);
+
+    const runHooks = getRunHooks();
+    await runHooks.begin('chat-1');
+    expect(streamRegistry.open).toHaveBeenCalledWith('chat-1', 'run-1');
+  });
+});

apps/server/src/core/ai-chat/ai-chat.controller.ts

@@ -4,11 +4,15 @@ import {
   ConflictException,
   Controller,
   ForbiddenException,
+  Get,
   HttpCode,
   HttpException,
   HttpStatus,
   Logger,
+  Param,
+  ParseUUIDPipe,
   Post,
+  Query,
   Req,
   Res,
   ServiceUnavailableException,
@@ -54,6 +58,12 @@ import {
 } from './dto/ai-chat.dto';
 import { describeProviderError } from '../../integrations/ai/ai-error.util';
 import { buildChatMarkdown } from './chat-markdown.util';
+import {
+  AiChatStreamRegistryService,
+  SUBSCRIBER_MAX_BUFFERED_BYTES,
+} from './ai-chat-stream-registry.service';
+import { startSseHeartbeat } from './sse-resilience';
+import { EnvironmentService } from '../../integrations/environment/environment.service';
 
 /**
  * Per-user AI chat API (§6.1). Routes are POST to match this codebase's
@@ -72,6 +82,11 @@ export class AiChatController {
     private readonly aiChatMessageRepo: AiChatMessageRepo,
     private readonly aiTranscription: AiTranscriptionService,
     private readonly pageRepo: PageRepo,
+    // #184 phase 1.5. OPTIONAL so existing positional constructions (controller
+    // specs) compile unchanged; Nest always injects the real providers in
+    // production. Only touched on the resumable-stream (flag-on) path.
+    private readonly streamRegistry?: AiChatStreamRegistryService,
+    private readonly environment?: EnvironmentService,
   ) {}
 
   /** List the requesting user's chats in this workspace (paginated). */
@@ -233,6 +248,102 @@ export class AiChatController {
     return { stopped };
   }
 
+  /**
+   * Attach to a chat's live run stream (#184 phase 1.5). A late/reloaded tab
+   * replays the frames buffered so far and then follows the live tail as a normal
+   * streamer. Owner-gated via assertOwnedChat (same gate as getRun). When there is
+   * nothing to resume — no entry, a finished run without expect=live, an
+   * overflowed buffer, or an anchor that pins a DIFFERENT run — the endpoint
+   * answers 204, the ONLY "nothing to resume" signal the AI SDK's reconnect
+   * accepts (it maps 204 to a silent no-op). With AI_CHAT_RESUMABLE_STREAM off the
+   * registry is never populated, so attach always 204s.
+   *
+   * `expect=live` opts into replaying a finished-but-retained run (safe only when
+   * the client stripped the streaming tail); `anchor` is the client's assistant
+   * row id, which must match this run's (invariant 6) or a foreign run's
+   * transcript would be replayed into the store.
+   */
+  @SkipTransform()
+  @UseGuards(JwtAuthGuard, UserThrottlerGuard)
+  @Throttle({ [AI_CHAT_THROTTLER]: { limit: 60, ttl: 60000 } })
+  @Get('runs/:chatId/stream')
+  async attachRunStream(
+    @Param('chatId', new ParseUUIDPipe()) chatId: string,
+    @Query('expect') expect: string | undefined,
+    @Query('anchor') anchor: string | undefined,
+    @Req() req: FastifyRequest,
+    @Res() res: FastifyReply,
+    @AuthUser() user: User,
+    @AuthWorkspace() workspace: Workspace,
+  ): Promise<void> {
+    await this.assertOwnedChat(chatId, user, workspace); // same gate as getRun
+    let stopHeartbeat: () => void = () => undefined;
+    const attachment = await this.streamRegistry?.attach(
+      chatId,
+      expect === 'live',
+      anchor,
+      {
+        onFrame: (frame) => {
+          // Backpressure guard: 2x the replay cap, so the initial replay burst
+          // alone can never trip it; only a genuinely stalled socket can.
+          try {
+            if (res.raw.writableLength > SUBSCRIBER_MAX_BUFFERED_BYTES) {
+              res.raw.destroy(); // 'close' fires -> unsubscribe below
+              return;
+            }
+            if (!res.raw.writableEnded) res.raw.write(frame);
+          } catch {
+            res.raw.destroy();
+          }
+        },
+        onEnd: () => {
+          stopHeartbeat();
+          if (!res.raw.writableEnded) res.raw.end();
+        },
+      },
+    );
+    if (!attachment) {
+      res.status(204).send(); // the ONLY "nothing to resume" signal the SDK accepts
+      return;
+    }
+    res.hijack();
+    // Cleanup BEFORE any write (invariant 5): a torn-down socket must not orphan
+    // a paused subscriber whose pending queue would buffer the whole run.
+    req.raw.once('close', () => {
+      attachment.unsubscribe();
+      stopHeartbeat();
+    });
+    // A close emitted DURING the awaits above was missed by the listener — check.
+    // (Healthy pending GETs have req.raw.destroyed === false, so no false
+    // positives; returning without end() is fine — the socket is gone.)
+    if (req.raw.destroyed) {
+      attachment.unsubscribe();
+      return;
+    }
+    res.raw.on('error', () => undefined);
+    try {
+      res.raw.writeHead(200, {
+        'content-type': 'text/event-stream',
+        'cache-control': 'no-cache',
+        'x-vercel-ai-ui-message-stream': 'v1',
+        'x-accel-buffering': 'no',
+        // deliberately NO Connection/Keep-Alive (hop-by-hop; Safari/HTTP2)
+      });
+      res.raw.flushHeaders?.();
+      for (const frame of attachment.replay) res.raw.write(frame);
+      if (attachment.finished) {
+        res.raw.end();
+        return;
+      }
+      stopHeartbeat = startSseHeartbeat(res.raw, 15_000);
+      attachment.start(); // drain pending accumulated during replay, go live
+    } catch {
+      attachment.unsubscribe();
+      stopHeartbeat();
+      res.raw.destroy();
+    }
+  }
+
   /** Rename a chat. */
   @HttpCode(HttpStatus.OK)
   @Post('rename')
@@ -344,13 +455,25 @@ export class AiChatController {
     // its progress, and settle its terminal status — see AiChatRunService.
     const runHooks: AiChatRunHooks | undefined = autonomousRuns
       ? {
-          begin: (chatId) =>
-            this.aiChatRunService.beginRun({
+          begin: async (chatId) => {
+            const handle = await this.aiChatRunService.beginRun({
               chatId,
               workspaceId: workspace.id,
               userId: user.id,
               trigger: 'user',
-            }),
+            });
+            // #184 phase 1.5: register the run-stream entry at BEGIN (before any
+            // frame) so a tab that attaches in the begin->seed window finds an
+            // entry to wait on. Gated on AI_CHAT_RESUMABLE_STREAM: with the flag
+            // off nothing is registered and attach always 204s.
+            if (
+              handle?.runId &&
+              this.environment?.isAiChatResumableStreamEnabled?.()
+            ) {
+              this.streamRegistry?.open(chatId, handle.runId);
+            }
+            return handle;
+          },
           onAssistantSeeded: (runId, messageId) =>
             this.aiChatRunService.linkAssistantMessage(
               runId,

apps/server/src/core/ai-chat/ai-chat.module.ts

@@ -4,6 +4,7 @@ import { TokenModule } from '../auth/token.module';
 import { AiChatController } from './ai-chat.controller';
 import { AiChatService } from './ai-chat.service';
 import { AiChatRunService } from './ai-chat-run.service';
+import { AiChatStreamRegistryService } from './ai-chat-stream-registry.service';
 import { AiTranscriptionService } from './ai-transcription.service';
 import { AiChatToolsService } from './tools/ai-chat-tools.service';
 import { EmbeddingModule } from './embedding/embedding.module';
@@ -44,6 +45,7 @@ import { PublicShareChatToolsService } from './tools/public-share-chat-tools.ser
   providers: [
     AiChatService,
     AiChatRunService,
+    AiChatStreamRegistryService,
     AiTranscriptionService,
     AiChatToolsService,
     PublicShareChatService,

apps/server/src/core/ai-chat/ai-chat.service.spec.ts

@@ -1,4 +1,13 @@
-import { ForbiddenException } from '@nestjs/common';
+import { ForbiddenException, Logger } from '@nestjs/common';
+// Mock ONLY streamText so a driven stream() call can capture the pipe-options
+// object (consumeSseStream / generateMessageId). Everything else in the AI SDK
+// stays REAL (requireActual), so the pure-helper suites in this file are
+// unaffected — none of them call stream()/streamText.
+jest.mock('ai', () => ({
+  ...jest.requireActual('ai'),
+  streamText: jest.fn(),
+}));
+import { streamText } from 'ai';
 import {
   AiChatService,
   compactToolOutput,
@@ -1059,3 +1068,181 @@ describe('isInterruptResume', () => {
     expect(isInterruptResume(withPrev(null), true)).toBe(false);
   });
 });
+
+/**
+ * #184 phase 1.5 — the run-wrapped pipe options (unit). Drives stream() to the
+ * pipe call with streamText mocked, capturing the options object, and asserts:
+ *  - flag OFF while a runId IS present -> the LEGACY option shape (no
+ *    consumeSseStream, no generateMessageId), and the registry is never touched.
+ *    This is the exact dormancy guarantee this PR rests on.
+ *  - flag ON + runId -> consumeSseStream tees into the registry and
+ *    generateMessageId returns the seeded assistant DB row id.
+ *  - flag ON but no runHooks (runId undefined) -> legacy (the runId gate).
+ *  - flag ON + runId -> the outer catch releases the entry via abortEntry.
+ */
+describe('AiChatService.stream — resumable pipe options (#184 phase 1.5)', () => {
+  const streamTextMock = streamText as unknown as jest.Mock;
+  let pipeMock: jest.Mock;
+
+  beforeEach(() => {
+    streamTextMock.mockReset();
+    pipeMock = jest.fn();
+    streamTextMock.mockReturnValue({
+      consumeStream: jest.fn(),
+      pipeUIMessageStreamToResponse: pipeMock,
+    });
+    // Silence the service's diagnostic logging for a clean test run.
+    jest.spyOn(Logger.prototype, 'log').mockImplementation(() => undefined as never);
+    jest
+      .spyOn(Logger.prototype, 'error')
+      .mockImplementation(() => undefined as never);
+    jest
+      .spyOn(Logger.prototype, 'warn')
+      .mockImplementation(() => undefined as never);
+  });
+
+  afterEach(() => jest.restoreAllMocks());
+
+  // A raw-response stub sufficient for the post-streamText wiring.
+  function makeRes() {
+    return {
+      raw: {
+        writeHead: jest.fn(),
+        write: jest.fn(),
+        once: jest.fn(),
+        on: jest.fn(),
+        flushHeaders: jest.fn(),
+        writableEnded: false,
+        destroyed: false,
+      },
+    };
+  }
+
+  // Wire only the deps reached on the way to the pipe call, plus a spy registry.
+  function makeService(opts: { resumable: boolean }) {
+    const aiChatRepo = {
+      findById: jest.fn(async () => ({ id: 'chat-1', workspaceId: 'ws-1' })),
+      insert: jest.fn(),
+    };
+    const aiChatMessageRepo = {
+      // Both the user insert and the assistant seed return the same row id.
+      insert: jest.fn(async () => ({ id: 'msg-1' })),
+      findAllByChat: jest.fn(async () => []),
+      update: jest.fn(async () => ({ id: 'msg-1' })),
+    };
+    const aiSettings = { resolve: jest.fn(async () => ({})) };
+    const tools = { forUser: jest.fn(async () => ({})) };
+    const mcpClients = {
+      toolsFor: jest.fn(async () => ({
+        tools: {},
+        clients: [],
+        outcomes: [],
+        instructions: [],
+      })),
+    };
+    const streamRegistry = {
+      open: jest.fn(),
+      bind: jest.fn(),
+      abortEntry: jest.fn(),
+    };
+    const svc = new AiChatService(
+      {} as never, // ai (model is injected)
+      aiChatRepo as never,
+      aiChatMessageRepo as never,
+      {} as never, // aiChatPageSnapshotRepo
+      aiSettings as never,
+      tools as never,
+      mcpClients as never,
+      {} as never, // aiAgentRoleRepo
+      {} as never, // pageRepo (openPage undefined -> never touched)
+      {} as never, // pageAccess
+      {
+        isAiChatDeferredToolsEnabled: () => false,
+        isAiChatResumableStreamEnabled: () => opts.resumable,
+      } as never,
+      streamRegistry as never,
+    );
+    return { svc, streamRegistry };
+  }
+
+  const body = {
+    chatId: 'chat-1',
+    messages: [
+      { id: 'm1', role: 'user', parts: [{ type: 'text', text: 'hi' }] },
+    ],
+  };
+
+  const makeRunHooks = () => ({
+    begin: jest.fn(async () => ({
+      runId: 'run-1',
+      signal: new AbortController().signal,
+    })),
+    onAssistantSeeded: jest.fn(),
+    onStep: jest.fn(),
+    onSettled: jest.fn(),
+  });
+
+  async function drive(svc: AiChatService, hooks: unknown): Promise<void> {
+    await svc.stream({
+      user: { id: 'u1' } as never,
+      workspace: { id: 'ws-1' } as never,
+      sessionId: 's1',
+      body: body as never,
+      res: makeRes() as never,
+      signal: new AbortController().signal,
+      model: {} as never,
+      role: null,
+      runHooks: hooks as never,
+    });
+  }
+
+  it('flag OFF + runId present: LEGACY option shape (no consumeSseStream / generateMessageId); registry untouched', async () => {
+    const { svc, streamRegistry } = makeService({ resumable: false });
+    await drive(svc, makeRunHooks());
+    expect(pipeMock).toHaveBeenCalledTimes(1);
+    const options = pipeMock.mock.calls[0][1];
+    // The dormancy guarantee: a live run with the flag off tees NOTHING and does
+    // not stamp a message id — byte-for-byte the pre-1.5 wire.
+    expect(options.consumeSseStream).toBeUndefined();
+    expect(options.generateMessageId).toBeUndefined();
+    expect(streamRegistry.bind).not.toHaveBeenCalled();
+    expect(streamRegistry.abortEntry).not.toHaveBeenCalled();
+  });
+
+  it('flag ON + runId: consumeSseStream tees into the registry; generateMessageId returns the seeded row id', async () => {
+    const { svc, streamRegistry } = makeService({ resumable: true });
+    await drive(svc, makeRunHooks());
+    const options = pipeMock.mock.calls[0][1];
+    expect(typeof options.consumeSseStream).toBe('function');
+    expect(typeof options.generateMessageId).toBe('function');
+    // generateMessageId stamps the seeded assistant DB row id.
+    expect(options.generateMessageId()).toBe('msg-1');
+    // consumeSseStream binds the tee: (chatId, runId, assistantId, stream).
+    const fakeStream = {} as ReadableStream<string>;
+    options.consumeSseStream({ stream: fakeStream });
+    expect(streamRegistry.bind).toHaveBeenCalledWith(
+      'chat-1',
+      'run-1',
+      'msg-1',
+      fakeStream,
+    );
+  });
+
+  it('flag ON but NO runHooks (runId undefined): pipe options stay legacy (the runId gate)', async () => {
+    const { svc, streamRegistry } = makeService({ resumable: true });
+    await drive(svc, undefined);
+    const options = pipeMock.mock.calls[0][1];
+    expect(options.consumeSseStream).toBeUndefined();
+    expect(options.generateMessageId).toBeUndefined();
+    expect(streamRegistry.bind).not.toHaveBeenCalled();
+  });
+
+  it('flag ON + runId: the outer catch calls abortEntry when the stream throws', async () => {
+    const { svc, streamRegistry } = makeService({ resumable: true });
+    streamTextMock.mockImplementation(() => {
+      throw new Error('boom');
+    });
+    await expect(drive(svc, makeRunHooks())).rejects.toThrow('boom');
+    expect(streamRegistry.abortEntry).toHaveBeenCalledWith('chat-1', 'run-1');
+  });
+});

apps/server/src/core/ai-chat/ai-chat.service.ts

@@ -32,6 +32,7 @@ import {
 import { AiChatToolsService } from './tools/ai-chat-tools.service';
 import { McpClientsService } from './external-mcp/mcp-clients.service';
 import { EnvironmentService } from '../../integrations/environment/environment.service';
+import { AiChatStreamRegistryService } from './ai-chat-stream-registry.service';
 import { buildSystemPrompt } from './ai-chat.prompt';
 import {
   CORE_TOOL_KEYS,
@@ -276,6 +277,10 @@ export class AiChatService implements OnModuleInit {
     // Reads the AI_CHAT_DEFERRED_TOOLS toggle (#332). Injected last so existing
     // positional constructor callers (tests) only append one stub.
     private readonly environment: EnvironmentService,
+    // #184 phase 1.5 run-stream registry. OPTIONAL so existing positional
+    // constructions (int-specs) compile unchanged; Nest always injects the real
+    // provider in production. Only ever touched on the run-wrapped + flag-on path.
+    private readonly streamRegistry?: AiChatStreamRegistryService,
   ) {}
 
   /**
@@ -1174,6 +1179,35 @@ export class AiChatService implements OnModuleInit {
         // as the cumulative authoritative usage so the client never jumps DOWN.
         let cumulativeStepUsage: ChatStreamUsage | undefined;
         result.pipeUIMessageStreamToResponse(res.raw, {
+          // #184 phase 1.5: run-wrapped mode only — the legacy path (flag off) stays
+          // byte-for-byte identical, including the absence of start.messageId. Both
+          // fields are gated on `runId` (present only for a durable run) AND the
+          // AI_CHAT_RESUMABLE_STREAM flag; the seed `assistantId` is unconditional,
+          // so gating on `assistantId` alone would change the legacy wire.
+          ...(runId && this.environment?.isAiChatResumableStreamEnabled?.()
+            ? {
+                // Tee the SSE frames into the run-stream registry so late tabs can
+                // attach (replay + live tail).
+                consumeSseStream: ({
+                  stream,
+                }: {
+                  stream: ReadableStream<string>;
+                }) =>
+                  this.streamRegistry?.bind(
+                    chatId,
+                    runId!,
+                    assistantId,
+                    stream,
+                  ),
+                // Stamp the persisted assistant row's DB id onto the streamed
+                // message so every tab renders the SAME id as the DB row (id-based
+                // reconciliation). Seeding is best-effort: when it failed, let the
+                // client generate the id.
+                ...(assistantId
+                  ? { generateMessageId: () => assistantId }
+                  : {}),
+              }
+            : {}),
           headers: { 'X-Accel-Buffering': 'no' },
           // Surface the authoritative chatId on the streamed assistant UI message so
           // the client adopts the REAL id of the row we created, instead of guessing
@@ -1239,6 +1273,12 @@ export class AiChatService implements OnModuleInit {
       // finalizeRun (onSettled) is idempotent — a settle here and a settle from a
       // streamText callback collapse to a single terminal write.
       if (runId) {
+        // #184 phase 1.5: a failure here means the tee `done` will never arrive,
+        // so release the registry entry's subscribers explicitly — otherwise an
+        // attached tab hangs forever. Same flag gate as the tee wiring above.
+        if (this.environment?.isAiChatResumableStreamEnabled?.()) {
+          this.streamRegistry?.abortEntry(chatId, runId);
+        }
         await runHooks?.onSettled?.(
           runId,
           'error',

apps/server/src/integrations/environment/environment.service.ts

@@ -292,6 +292,23 @@ export class EnvironmentService {
     return enabled === 'true';
   }
 
+  /**
+   * Resumable SSE transport for durable agent runs (#184 phase 1.5). When
+   * enabled, a run tees its SSE frames into the in-memory run-stream registry so
+   * a late/reloaded tab can attach (replay + live tail) via
+   * `GET /ai-chat/runs/:chatId/stream`. Defaults to DISABLED: PR 1 ships the
+   * server code dormant — with the flag off, `open`/`bind`/`generateMessageId`
+   * are never called and attach always answers 204, so the legacy and #184
+   * phase-1 wire paths stay byte-for-byte identical. Set
+   * AI_CHAT_RESUMABLE_STREAM=true to activate it (paired with the PR 2 client).
+   */
+  isAiChatResumableStreamEnabled(): boolean {
+    const enabled = this.configService
+      .get<string>('AI_CHAT_RESUMABLE_STREAM', 'false')
+      .toLowerCase();
+    return enabled === 'true';
+  }
+
   getPostHogHost(): string {
     return this.configService.get<string>('POSTHOG_HOST');
   }

apps/server/test/integration/ai-chat-attach.int-spec.ts

@@ -0,0 +1,564 @@
+import * as http from 'node:http';
+import { Kysely } from 'kysely';
+import {
+  MockLanguageModelV3,
+  convertArrayToReadableStream,
+} from 'ai/test';
+import { AiChatRepo } from '@docmost/db/repos/ai-chat/ai-chat.repo';
+import { AiChatMessageRepo } from '@docmost/db/repos/ai-chat/ai-chat-message.repo';
+import { AiChatRunRepo } from '@docmost/db/repos/ai-chat/ai-chat-run.repo';
+import { AiChatService } from 'src/core/ai-chat/ai-chat.service';
+import { AiChatRunService } from 'src/core/ai-chat/ai-chat-run.service';
+import {
+  AiChatStreamRegistryService,
+  RunStreamCallbacks,
+} from 'src/core/ai-chat/ai-chat-stream-registry.service';
+import {
+  getTestDb,
+  destroyTestDb,
+  createWorkspace,
+  createUser,
+  createChat,
+} from './db';
+
+/**
+ * #184 phase 1.5 — the resumable transport end to end against REAL Postgres,
+ * the REAL `streamText` (seeded via MockLanguageModelV3) and a REAL Node
+ * ServerResponse, driving the REAL `AiChatService.stream` run-wrapped path with a
+ * REAL `AiChatStreamRegistryService`. The run-hooks mirror the controller: they
+ * begin a durable run and `open()` the registry entry at begin, and the service
+ * tees the SSE frames into it via `consumeSseStream` while stamping the DB row id
+ * via `generateMessageId` (both gated on runId + the resumable flag).
+ *
+ * Proven here: a finished run's replay is the full frame sequence incl `[DONE]`
+ * with `start.messageId` == the seeded DB row id; the anchor check (invariant 6);
+ * an attach opened BEFORE the first frame follows the live stream from frame 0; an
+ * explicit stop surfaces `{"type":"abort"}` + `[DONE]` + end to the subscriber;
+ * and the legacy (non-run) path tees nothing.
+ */
+
+const sleep = (ms: number) => new Promise((r) => setTimeout(r, ms));
+
+async function waitFor(
+  cond: () => Promise<boolean> | boolean,
+  { timeoutMs = 15_000, stepMs = 25 } = {},
+): Promise<void> {
+  const start = Date.now();
+  while (Date.now() - start < timeoutMs) {
+    if (await cond()) return;
+    await sleep(stepMs);
+  }
+  throw new Error('waitFor: condition not met within timeout');
+}
+
+// A real Node ServerResponse wired to a live socket (as in the stream int-spec).
+function makeRealResponse(): Promise<{
+  res: http.ServerResponse;
+  cleanup: () => Promise<void>;
+}> {
+  return new Promise((resolve) => {
+    const server = http.createServer((_req, res) => {
+      resolve({
+        res,
+        cleanup: () =>
+          new Promise<void>((done) => {
+            try {
+              if (!res.writableEnded) res.end();
+            } catch {
+              /* socket already gone */
+            }
+            server.close(() => done());
+          }),
+      });
+    });
+    server.listen(0, () => {
+      const port = (server.address() as any).port;
+      const creq = http.request({ port, method: 'GET' }, (cres) => {
+        cres.resume();
+      });
+      creq.on('error', () => undefined);
+      creq.end();
+    });
+  });
+}
+
+// A full, successful single-step turn.
+function successStream() {
+  return convertArrayToReadableStream([
+    { type: 'stream-start', warnings: [] },
+    { type: 'text-start', id: 't1' },
+    { type: 'text-delta', id: 't1', delta: 'Hello' },
+    { type: 'text-delta', id: 't1', delta: ' there' },
+    { type: 'text-end', id: 't1' },
+    {
+      type: 'finish',
+      finishReason: 'stop',
+      usage: { inputTokens: 10, outputTokens: 5, totalTokens: 15 },
+    },
+  ] as any);
+}
+
+// A stream the test feeds chunk by chunk (to attach mid-flight / drive a stop).
+function makeControlledStream() {
+  let controller!: ReadableStreamDefaultController<any>;
+  const stream = new ReadableStream<any>({
+    start(c) {
+      controller = c;
+    },
+  });
+  return {
+    stream,
+    emit: (chunk: any) => controller.enqueue(chunk),
+    close: () => controller.close(),
+  };
+}
+
+// Collect replay + live frames from an attachment.
+function liveSink(): {
+  cb: RunStreamCallbacks;
+  frames: string[];
+  ended: () => boolean;
+} {
+  const frames: string[] = [];
+  let ended = false;
+  return {
+    frames,
+    ended: () => ended,
+    cb: {
+      onFrame: (f) => frames.push(f),
+      onEnd: () => {
+        ended = true;
+      },
+    },
+  };
+}
+
+// The SSE `start` frame carries the message id; pull it out of a `data: {...}`.
+function parseStartMessageId(frames: string[]): string | undefined {
+  for (const f of frames) {
+    const m = /^data: (\{.*\})\s*$/m.exec(f.trim());
+    if (!m) continue;
+    try {
+      const json = JSON.parse(m[1]);
+      if (json.type === 'start') return json.messageId;
+    } catch {
+      /* not this frame */
+    }
+  }
+  return undefined;
+}
+
+describe('AiChatService run-stream attach [integration]', () => {
+  let db: Kysely<any>;
+  let aiChatRepo: AiChatRepo;
+  let msgRepo: AiChatMessageRepo;
+  let runRepo: AiChatRunRepo;
+  let workspaceId: string;
+  let userId: string;
+
+  const mcpClients = {
+    toolsFor: async () => ({
+      tools: {},
+      clients: [],
+      outcomes: [],
+      instructions: [],
+    }),
+  };
+
+  // Build the service with the run-stream registry wired and the resumable flag
+  // ON (the property under test). Deferred tools OFF (irrelevant here).
+  function buildService(registry: AiChatStreamRegistryService): AiChatService {
+    return new AiChatService(
+      { getChatModel: async () => null } as any,
+      aiChatRepo,
+      msgRepo,
+      {} as any,
+      { resolve: async () => null } as any,
+      { forUser: async () => ({}) } as any,
+      mcpClients as any,
+      {} as any,
+      {} as any,
+      {} as any,
+      {
+        isAiChatDeferredToolsEnabled: () => false,
+        isAiChatResumableStreamEnabled: () => true,
+      } as any,
+      registry,
+    );
+  }
+
+  // Run-hooks mirroring the controller: begin the durable run AND open() the
+  // registry entry at begin. Captures the runId so a test can stop it.
+  function makeRunHooks(
+    runService: AiChatRunService,
+    registry: AiChatStreamRegistryService,
+    box: { runId?: string },
+  ) {
+    return {
+      begin: async (chatId: string) => {
+        const handle = await runService.beginRun({
+          chatId,
+          workspaceId,
+          userId,
+          trigger: 'user',
+        });
+        box.runId = handle.runId;
+        registry.open(chatId, handle.runId);
+        return handle;
+      },
+      onAssistantSeeded: (runId: string, messageId: string) =>
+        runService.linkAssistantMessage(runId, workspaceId, messageId),
+      onStep: (runId: string, n: number) =>
+        void runService.recordStep(runId, workspaceId, n),
+      onSettled: (runId: string, status: any, error?: string) =>
+        runService.finalizeRun(runId, workspaceId, status, error),
+    };
+  }
+
+  function userUiMessage(text: string) {
+    return {
+      id: `u-${Math.random()}`,
+      role: 'user',
+      parts: [{ type: 'text', text }],
+    };
+  }
+
+  async function startRun(opts: {
+    registry: AiChatStreamRegistryService;
+    runService?: AiChatRunService;
+    model: MockLanguageModelV3;
+    chatId: string;
+    body: any;
+    box?: { runId?: string };
+  }): Promise<{ res: http.ServerResponse; cleanup: () => Promise<void> }> {
+    const service = buildService(opts.registry);
+    const { res, cleanup } = await makeRealResponse();
+    const runHooks = opts.runService
+      ? makeRunHooks(opts.runService, opts.registry, opts.box ?? {})
+      : undefined;
+    await service.stream({
+      user: { id: userId, workspaceId } as any,
+      workspace: { id: workspaceId, name: 'WS' } as any,
+      sessionId: 'sess-1',
+      body: opts.body,
+      res: { raw: res } as any,
+      signal: new AbortController().signal,
+      model: opts.model as any,
+      role: null,
+      runHooks,
+    } as any);
+    return { res, cleanup };
+  }
+
+  async function assistantRowId(chatId: string): Promise<string> {
+    const rows = await msgRepo.findAllByChat(chatId, workspaceId);
+    const row = rows.find((r: any) => r.role === 'assistant');
+    return row!.id as string;
+  }
+
+  beforeAll(async () => {
+    db = getTestDb();
+    aiChatRepo = new AiChatRepo(db as any);
+    msgRepo = new AiChatMessageRepo(db as any);
+    runRepo = new AiChatRunRepo(db as any);
+    workspaceId = (await createWorkspace(db)).id;
+    userId = (await createUser(db, workspaceId)).id;
+  });
+
+  afterAll(async () => {
+    await destroyTestDb();
+  });
+
+  it('run-wrapped: replay is the full frame sequence incl [DONE], start.messageId == the seeded DB row id', async () => {
+    const chatId = (await createChat(db, { workspaceId, creatorId: userId })).id;
+    const registry = new AiChatStreamRegistryService();
+    const runService = new AiChatRunService(runRepo, {
+      isCloud: () => false,
+    } as never);
+    const model = new MockLanguageModelV3({
+      doStream: async () => ({ stream: successStream() }),
+    } as any);
+
+    const box: { runId?: string } = {};
+    const { cleanup } = await startRun({
+      registry,
+      runService,
+      model,
+      chatId,
+      body: { chatId, messages: [userUiMessage('Hi')] },
+      box,
+    });
+    try {
+      // Wait for the assistant row to settle (terminal callbacks run async).
+      await waitFor(async () => {
+        const rows = await msgRepo.findAllByChat(chatId, workspaceId);
+        return rows.some(
+          (r: any) =>
+            r.role === 'assistant' &&
+            ['completed', 'error', 'aborted'].includes(r.status),
+        );
+      });
+      const rowId = await assistantRowId(chatId);
+
+      // Finished-run replay with expect=live + the correct anchor.
+      const sink = liveSink();
+      const att = await registry.attach(chatId, true, rowId, sink.cb);
+      expect(att).not.toBeNull();
+      expect(att!.finished).toBe(true);
+      // The tee captured frames (consumeSseStream was wired).
+      expect(att!.replay.length).toBeGreaterThan(0);
+      // generateMessageId stamped the DB row id onto the streamed start frame.
+      expect(parseStartMessageId(att!.replay)).toBe(rowId);
+      // The full sequence includes the streamed text and the terminal marker.
+      const joined = att!.replay.join('');
+      expect(joined).toContain('Hello');
+      expect(att!.replay.some((f) => f.includes('[DONE]'))).toBe(true);
+    } finally {
+      registry.onModuleDestroy();
+      await cleanup();
+    }
+  });
+
+  it('anchor mismatch with expect=live returns null (invariant 6)', async () => {
+    const chatId = (await createChat(db, { workspaceId, creatorId: userId })).id;
+    const registry = new AiChatStreamRegistryService();
+    const runService = new AiChatRunService(runRepo, {
+      isCloud: () => false,
+    } as never);
+    const model = new MockLanguageModelV3({
+      doStream: async () => ({ stream: successStream() }),
+    } as any);
+
+    const { cleanup } = await startRun({
+      registry,
+      runService,
+      model,
+      chatId,
+      body: { chatId, messages: [userUiMessage('Hi')] },
+    });
+    try {
+      await waitFor(async () => {
+        const rows = await msgRepo.findAllByChat(chatId, workspaceId);
+        return rows.some(
+          (r: any) => r.role === 'assistant' && r.status === 'completed',
+        );
+      });
+      const sink = liveSink();
+      // A foreign anchor must NOT replay this run's transcript.
+      expect(
+        await registry.attach(chatId, true, 'a-different-run-row', sink.cb),
+      ).toBeNull();
+    } finally {
+      registry.onModuleDestroy();
+      await cleanup();
+    }
+  });
+
+  it('an attach opened BEFORE the first frame follows the live stream from frame 0', async () => {
+    const chatId = (await createChat(db, { workspaceId, creatorId: userId })).id;
+    const registry = new AiChatStreamRegistryService();
+    const runService = new AiChatRunService(runRepo, {
+      isCloud: () => false,
+    } as never);
+    const controlled = makeControlledStream();
+    const model = new MockLanguageModelV3({
+      doStream: async () => ({ stream: controlled.stream }),
+    } as any);
+
+    const { cleanup } = await startRun({
+      registry,
+      runService,
+      model,
+      chatId,
+      body: { chatId, messages: [userUiMessage('Slow please')] },
+    });
+    try {
+      // Attach while the entry exists (opened at begin) but before any frame.
+      const sink = liveSink();
+      const att = (await registry.attach(chatId, false, undefined, sink.cb))!;
+      expect(att.replay).toEqual([]); // nothing streamed yet -> replay from 0
+      att.start(); // go live (drains nothing, then follows)
+
+      // Now emit the whole turn.
+      controlled.emit({ type: 'stream-start', warnings: [] });
+      controlled.emit({ type: 'text-start', id: 't1' });
+      controlled.emit({ type: 'text-delta', id: 't1', delta: 'Zero' });
+      controlled.emit({ type: 'text-end', id: 't1' });
+      controlled.emit({
+        type: 'finish',
+        finishReason: 'stop',
+        usage: { inputTokens: 1, outputTokens: 1, totalTokens: 2 },
+      });
+      controlled.close();
+
+      await waitFor(() => sink.frames.some((f) => f.includes('[DONE]')));
+      // The subscriber saw the stream from the very first frame (`start`) through
+      // the terminal marker, with the streamed text present.
+      expect(sink.frames.some((f) => f.includes('"type":"start"'))).toBe(true);
+      expect(sink.frames.join('')).toContain('Zero');
+      expect(sink.frames[sink.frames.length - 1]).toContain('[DONE]');
+      expect(sink.ended()).toBe(true);
+    } finally {
+      registry.onModuleDestroy();
+      await cleanup();
+    }
+  });
+
+  it('requestStop surfaces {"type":"abort"} + [DONE] + end to the attached subscriber', async () => {
+    const chatId = (await createChat(db, { workspaceId, creatorId: userId })).id;
+    const registry = new AiChatStreamRegistryService();
+    const runService = new AiChatRunService(runRepo, {
+      isCloud: () => false,
+    } as never);
+    // An abort-AWARE model: it streams some partial output, then errors the model
+    // stream with an AbortError when the run signal aborts — exactly as a real
+    // provider network stream is torn down on abort (a plain in-memory stream
+    // would just stall, so streamText would never observe the stop).
+    const model = new MockLanguageModelV3({
+      doStream: async ({ abortSignal }: any) => {
+        const stream = new ReadableStream<any>({
+          start(controller) {
+            controller.enqueue({ type: 'stream-start', warnings: [] });
+            controller.enqueue({ type: 'text-start', id: 't1' });
+            controller.enqueue({ type: 'text-delta', id: 't1', delta: 'partial' });
+            abortSignal?.addEventListener('abort', () => {
+              try {
+                controller.error(
+                  new DOMException('Aborted', 'AbortError'),
+                );
+              } catch {
+                /* already errored/closed */
+              }
+            });
+          },
+        });
+        return { stream };
+      },
+    } as any);
+
+    const box: { runId?: string } = {};
+    const { cleanup } = await startRun({
+      registry,
+      runService,
+      model,
+      chatId,
+      body: { chatId, messages: [userUiMessage('Start then stop')] },
+      box,
+    });
+    try {
+      const sink = liveSink();
+      const att = (await registry.attach(chatId, false, undefined, sink.cb))!;
+      att.start();
+
+      // Give streamText a beat to begin consuming the partial output.
+      await sleep(250);
+      // User presses Stop -> the run signal aborts -> the SDK emits an abort chunk.
+      await runService.requestStop(box.runId!, workspaceId);
+
+      await waitFor(() => sink.ended());
+      expect(sink.frames.some((f) => f.includes('"type":"abort"'))).toBe(true);
+      expect(sink.frames.some((f) => f.includes('[DONE]'))).toBe(true);
+      expect(sink.ended()).toBe(true);
+    } finally {
+      registry.onModuleDestroy();
+      await cleanup();
+    }
+  });
+
+  it('the outer catch calls abortEntry so an open entry is released (finished)', async () => {
+    const chatId = (await createChat(db, { workspaceId, creatorId: userId })).id;
+    const registry = new AiChatStreamRegistryService();
+    const runService = new AiChatRunService(runRepo, {
+      isCloud: () => false,
+    } as never);
+    const model = new MockLanguageModelV3({
+      doStream: async () => ({ stream: successStream() }),
+    } as any);
+
+    // A msgRepo whose user-row insert throws: the turn fails AFTER begin (the
+    // entry is already open) but BEFORE the pipe, exercising the outer catch.
+    const throwingMsgRepo = {
+      insert: async () => {
+        throw new Error('db boom');
+      },
+      findAllByChat: (...a: any[]) => (msgRepo as any).findAllByChat(...a),
+      update: (...a: any[]) => (msgRepo as any).update(...a),
+      findById: (...a: any[]) => (msgRepo as any).findById(...a),
+    };
+    const service = new AiChatService(
+      { getChatModel: async () => null } as any,
+      aiChatRepo,
+      throwingMsgRepo as any,
+      {} as any,
+      { resolve: async () => null } as any,
+      { forUser: async () => ({}) } as any,
+      mcpClients as any,
+      {} as any,
+      {} as any,
+      {} as any,
+      {
+        isAiChatDeferredToolsEnabled: () => false,
+        isAiChatResumableStreamEnabled: () => true,
+      } as any,
+      registry,
+    );
+    const { res, cleanup } = await makeRealResponse();
+    const box: { runId?: string } = {};
+    try {
+      await expect(
+        service.stream({
+          user: { id: userId, workspaceId } as any,
+          workspace: { id: workspaceId, name: 'WS' } as any,
+          sessionId: 'sess-1',
+          body: { chatId, messages: [userUiMessage('will throw')] },
+          res: { raw: res } as any,
+          signal: new AbortController().signal,
+          model: model as any,
+          role: null,
+          runHooks: makeRunHooks(runService, registry, box),
+        } as any),
+      ).rejects.toThrow();
+      // The entry opened at begin was terminated by abortEntry (from the catch),
+      // so it is finished and a plain attach returns null instead of hanging.
+      const entry = (registry as any).entries.get(chatId);
+      expect(entry).toBeDefined();
+      expect(entry.finished).toBe(true);
+      const sink = liveSink();
+      expect(await registry.attach(chatId, false, undefined, sink.cb)).toBeNull();
+    } finally {
+      registry.onModuleDestroy();
+      await cleanup();
+    }
+  });
+
+  it('legacy (no run-hooks): the registry is never populated', async () => {
+    const chatId = (await createChat(db, { workspaceId, creatorId: userId })).id;
+    const registry = new AiChatStreamRegistryService();
+    const model = new MockLanguageModelV3({
+      doStream: async () => ({ stream: successStream() }),
+    } as any);
+
+    // No runHooks -> runId undefined -> the run-wrapped tee is never wired.
+    const { cleanup } = await startRun({
+      registry,
+      model,
+      chatId,
+      body: { chatId, messages: [userUiMessage('Legacy hi')] },
+    });
+    try {
+      await waitFor(async () => {
+        const rows = await msgRepo.findAllByChat(chatId, workspaceId);
+        return rows.some(
+          (r: any) => r.role === 'assistant' && r.status === 'completed',
+        );
+      });
+      const sink = liveSink();
+      // No entry was ever opened; attach always yields null.
+      expect(await registry.attach(chatId, false, undefined, sink.cb)).toBeNull();
+      expect(await registry.attach(chatId, true, 'anything', sink.cb)).toBeNull();
+    } finally {
+      registry.onModuleDestroy();
+      await cleanup();
+    }
+  });
+});