Align Assistant provider retry with OpenCode

docs/release-control/v6/internal/subsystems/ai-runtime.md

@@ -417,102 +417,55 @@ leave the transcript without exposing hidden provider/tool metadata.
    retryable in-memory assistant error may replay the original user turn's
    structured mentions, finding id, approval override, handoff resources,
    handoff actions, and handoff metadata, but must not reconstruct scoped
-   context from prompt history or saved transcript prose.
-   Failed-turn recovery must also expose model-route switching through the
-   existing drawer model selector so operators can move from a blocked direct
-   provider route to a configured gateway or alternate model without losing the
-   draft or creating a parallel picker. When the same model is available through
-   another configured provider route, the failed turn must offer that route as a
-   direct one-click route-and-retry action, but route recovery must not loop
-   between providers that already failed in the same transcript; once equivalent
-   routes are exhausted, the failed turn must fall back to another configured
-   notable model/provider before falling back to the general model selector.
-   Retry remains available, but it must not be the only visible action when a
-   failed Assistant turn is shown.
-   Provider recovery before visible output is backend-owned chat-runtime
-   behavior. When a selected route fails before streaming content, tool
-   progress, approval, or question events, `internal/ai/chat` may try the next
-   configured provider/model route in the same user turn and must emit a
-   `provider_fallback` workflow-state event that identifies the failed and next
-   provider/model route. This fallback must use chat-suitable model resolution
-   from `internal/ai/modelresolution`, skipping obvious non-chat endpoints such
-   as realtime, audio, moderation, embedding, and content-safety catalog
-   entries. Gateway-equivalent routes, such as a direct provider model exposed
-   through a configured gateway, must be produced by that model-resolution
-   policy rather than hardcoded in chat execution or frontend UI logic.
-   Fallback planning must not block the selected provider's first
-   attempt on live catalog reads for every other configured provider; the hot
-   path may queue only explicit provider preferences or stable provider defaults
-   and defer fallback provider construction until the selected route actually
-   fails before visible output. Primary interactive chat model resolution is
-   governed by that same hot-path rule: it must use the explicit configured
+   context from prompt history or saved transcript prose. Provider/model route
+   recovery is explicit user-visible recovery, not hidden chat execution. A
+   failed turn may expose route-and-retry actions through the existing drawer
+   model selector so operators can move from a blocked direct provider route to a
+   configured gateway or alternate model without losing the draft or creating a
+   parallel picker, but `/api/ai/chat` must not automatically switch provider or
+   model routes inside a single Assistant turn after the selected route fails.
+   That prohibition includes same-model gateway equivalents and configured
+   provider defaults: those routes may be offered as explicit recovery actions,
+   but the selected turn owns its selected route until it completes or fails.
+   The referenced OpenCode source at fetched `dev` commit
+   `e82542b8023a8374f29c23b70ec019c8f256354e` models this boundary by keeping
+   same-route retry visible in `packages/opencode/src/session/retry.ts`,
+   publishing retry status from `packages/opencode/src/session/processor.ts`,
+   rendering retry state in `packages/ui/src/components/session-retry.tsx`, and
+   treating provider/model changes as explicit session events in
+   `packages/core/src/session/event.ts` and
+   `packages/opencode/src/session/prompt.ts`. Pulse adapts that behavior for the
+   drawer instead of implementing automatic cross-provider fallback.
+   Primary interactive chat model resolution must use the explicit configured
    chat route or a stable provider default without calling provider model
    catalogs before the selected stream starts. Catalog-backed recommendation
-   belongs to settings/model-list flows, not `/api/ai/chat` first-response
-   startup. When the selected direct-provider route fails before visible output,
-   fallback planning may first try configured gateway-equivalent candidates
-   returned by `internal/ai/modelresolution` when they preserve the same
-   provider/model family through a deterministic route string, before falling
-   back to unrelated configured provider defaults. That same-model gateway
-   planning must remain catalog-free, capability-driven, and may continue to the
-   next configured provider if a gateway attempt also fails before visible
-   output; chat execution and frontend UI must never encode user-specific
-   network assumptions or named-provider failover rules. Once visible output has
-   streamed, Pulse must
-   not silently switch providers for that turn; the error belongs to the visible
-   attempt and is surfaced through normal failed-turn recovery. Assistant
-   startup retry behavior must be route-aware: when another provider/model
-   attempt is already queued for the same user turn, non-final attempts must
-   fast-fail before visible output instead of spending hidden same-route retry
-   and backoff time. In that mode, provider-owned startup retries and the
-   agentic loop's retry of the same provider turn are disabled for the
-   non-final attempt, and the wait for the first visible provider event is
-   bounded so `provider_fallback` can reach the browser quickly. The final
-   queued attempt keeps normal provider retry behavior because there is no
-   later route to hand off to. The referenced OpenCode source at fetched
-   `origin/dev` commit `1025540fcc2a69609a0131a7168300205656d728` models retry
-   as explicit session activity through `RetryError` and `Retried` in
-   `packages/core/src/session/event.ts` and renders retry status in
-   `packages/ui/src/components/session-retry.tsx`; Pulse adapts that principle
-   by surfacing route changes through workflow/model rows and by yielding to
-   the next configured route quickly instead of hiding repeated startup
-   retries behind a static waiting label. Provider retry workflow states that
-   include `retry_after_ms` must render as live countdowns derived from the
-   workflow `started_at` timestamp while the turn is active, matching
-   OpenCode's visible retry wait behavior instead of freezing the first retry
-   label.
+   belongs to settings/model-list and explicit route-recovery flows, not
+   `/api/ai/chat` first-response startup. Once the selected provider route
+   starts, transient pre-output transport failures may retry the same route and
+   must surface a `provider_retry` workflow state with `attempt`,
+   `max_attempts`, and `retry_after_ms` before sleeping. If same-route retry is
+   exhausted before visible output, Pulse must emit a normal provider error and
+   leave recovery to the failed-turn actions; it must not emit
+   `provider_fallback`, `failed_provider`, `failed_model`, `next_provider`, or
+   `next_model` workflow metadata. Once visible output has streamed, Pulse must
+   also keep the failure on that visible attempt rather than silently changing
+   route mid-turn. Provider retry workflow states that include `retry_after_ms`
+   must render as live countdowns derived from the workflow `started_at`
+   timestamp while the turn is active, matching OpenCode's visible retry wait
+   behavior instead of freezing the first retry label.
    Assistant completion events must carry the effective model route that actually
-   completed the turn, and the drawer must update the in-flight transcript row
-   when `provider_fallback` names the next route so message labels, cost
-   context, retry decisions, and model-route recovery do not continue to point
-   at the failed provider. The referenced OpenCode source at fetched
-   `origin/dev` commit `4519a1da329c1a4fc384054e7203ba7d06928205` defines
-   `session.next.model.switched` in
-   `packages/core/src/session/event.ts`, appends it as a
-   `SessionMessage.ModelSwitched` transcript message in
-   `packages/core/src/session/message-updater.ts`, and renders it as
-   `ModelSwitchedMessage` in
-   `packages/opencode/src/cli/cmd/tui/feature-plugins/system/session-v2.tsx`.
-   Pulse adapts that by rendering model route changes as typed `model_switch`
-   stream rows on the active assistant turn instead of hiding the route in
-   transient status text or assistant prose. The initial `provider_start`
-   workflow state must carry the selected `provider` and concrete `model` route,
-   and the drawer must render that first route as a selected-model row (`Using
-   ...`) before visible content or tool activity when the event arrives.
-   Provider-fallback rows must preserve the failed route and next route
-   together, using the backend `failed_model` and `next_model` payloads, so the
-   transcript shows the actual recovery path rather than only the successful
-   replacement model. A successful provider fallback is also a session-model
-   event, not only transcript decoration: once the live assistant turn completes
-   with the fallback route as its effective model, the drawer must promote that
-   route into the active session selection and recent-model list so the next
-   prompt does not retry the just-failed direct provider before using the route
-   that worked. This promotion may only apply to fallback rows observed during
-   the live stream and must not rewrite historical loaded sessions or override
-   an explicit user model change made while the fallback turn was still running.
+   completed the selected turn. The drawer must render the initial
+   `provider_start` workflow state as a selected-model row (`Using ...`) before
+   visible content or tool activity when the event arrives. Later model-route
+   changes may still be rendered as typed `model_switch` transcript rows when
+   they come from explicit route recovery or restored historical data, but the UI
+   must label them neutrally as selected or switched routes rather than as
+   automatic provider fallback, and completing a streamed route-switch row must
+   not promote a fallback route into the active model selection without an
+   explicit user action.
    Interactive Assistant streams must establish the session ID and emit the
    `session` event once as soon as the HTTP SSE writer is ready, before finding
-   handoff recovery, model resolution, provider fallback planning,
+   handoff recovery, model resolution, selected-provider startup,
    handoff/context prefetch, recent-session injection, inventory summary reads,
    tool scoping, or provider startup. The chat service then persists/ensures
    that same session ID while suppressing duplicate session events. This keeps
@@ -1008,7 +961,7 @@ leave the transcript without exposing hidden provider/tool metadata.
    external providers or models. The live fixture must pace its status, tool,
    and content events enough for the browser to paint each state in sequence;
    unit tests may disable that pace, but the dev fixture must not collapse into
-   a single final completed row. This fixture must not run as provider fallback
+   a single final completed row. This fixture must not emit `provider_fallback`
    in non-mock mode and must not become Pulse-authored remediation or routing
    behavior. The referenced OpenCode source at fetched `origin/dev` commit
    `4519a1da329c1a4fc384054e7203ba7d06928205` publishes tool-call and

docs/release-control/v6/internal/subsystems/api-contracts.md

@@ -123,7 +123,13 @@ to `frontend-modern/src/api/generated/aiChatEvents.ts` by
 Provider retry progress must use typed fields (`attempt`, `max_attempts`,
 `retry_after_ms`) on that same event instead of frontend-only string parsing or
 provider-specific ad hoc events; the Assistant UI may format those fields, but
-must not invent retry progress that the stream contract did not carry.
+must not invent retry progress that the stream contract did not carry. Assistant
+chat stream payloads must not expose automatic provider fallback metadata:
+`workflow_state` may carry the selected `provider` and `model` plus same-route
+retry fields, but `provider_fallback`, `failed_provider`, `failed_model`,
+`next_provider`, and `next_model` are retired from the generated stream event
+contract. Cross-route recovery belongs to explicit failed-turn actions, not
+hidden `/api/ai/chat` stream mutation.
 Assistant local stream fixtures are part of the same frontend API contract:
 `frontend-modern/src/api/aiChatDevStreamFixture.ts` may short-circuit only
 explicit `/fixture ...` prompts in development or test mode, must emit the same
@@ -134,6 +140,10 @@ obsolete grouped-context wording in fixture answer content. The fixture payload
 contract proves the stream reducer and transcript renderer against the same
 chronological event order a live provider would produce; UI grouping or footer
 summaries are not part of the fixture contract.
+The provider-retry fixture must exercise selected-route retry by emitting
+`provider_retry` for the selected route and completing with that same model; it
+must not simulate an automatic switch to a configured gateway or alternate
+provider route.
 Queue verification fixtures must cover both the active hold turn and the queued
 drain turn so UX proof can exercise queued follow-up ordering and tool rows
 without consuming external model quota.

frontend-modern/src/api/__tests__/aiChat.test.ts

@@ -244,12 +244,7 @@ describe('AIChatAPI', () => {
   it('runs the send-hold dev stream fixture without opening a provider request', async () => {
     const onEvent = vi.fn();
 
-    await AIChatAPI.chat(
-      '/fixture send-hold',
-      undefined,
-      'openrouter:qwen/qwen3.7-plus',
-      onEvent,
-    );
+    await AIChatAPI.chat('/fixture send-hold', undefined, 'openrouter:qwen/qwen3.7-plus', onEvent);
 
     expect(apiFetchMock).not.toHaveBeenCalled();
     expect(onEvent.mock.calls.map(([event]) => event.type)).toEqual([
@@ -551,7 +546,6 @@ describe('AIChatAPI', () => {
       'workflow_state',
       'workflow_state',
       'workflow_state',
-      'workflow_state',
       'content',
       'done',
     ]);
@@ -559,28 +553,17 @@ describe('AIChatAPI', () => {
       type: 'workflow_state',
       data: {
         phase: 'provider_retry',
-        message: 'DeepSeek failed before output; retrying through OpenRouter.',
+        message: 'Provider connection failed before any output; retrying.',
         attempt: 2,
-        failed_model: 'deepseek:deepseek-chat',
         max_attempts: 3,
-        next_model: 'openrouter:deepseek/deepseek-chat',
         retry_after_ms: 3200,
       },
     });
-    expect(onEvent.mock.calls[4][0]).toMatchObject({
-      type: 'workflow_state',
-      data: {
-        phase: 'provider_start',
-        message: 'Retrying through OpenRouter with deepseek/deepseek-chat.',
-        provider: 'openrouter',
-        model: 'openrouter:deepseek/deepseek-chat',
-      },
-    });
-    expect(onEvent.mock.calls[6][0]).toMatchObject({
+    expect(onEvent.mock.calls[5][0]).toMatchObject({
       type: 'done',
       data: {
         session_id: 'dev-fixture-provider-retry',
-        model: 'openrouter:deepseek/deepseek-chat',
+        model: 'deepseek:deepseek-chat',
       },
     });
   });

frontend-modern/src/api/__tests__/aiChatEvents.test.ts

@@ -24,22 +24,12 @@ describe('AI chat stream event contract', () => {
     expect(aiChatEventsSource).toContain("type: 'session'");
   });
 
-  it('exposes provider fallback metadata on workflow state events', () => {
-    const workflow: WorkflowStateData = {
-      phase: 'provider_fallback',
-      message: 'OpenRouter did not start a response; trying DeepSeek.',
-      state: 'provider_fallback',
-      failed_provider: 'openrouter',
-      failed_model: 'openrouter:qwen/qwen3.7-plus',
-      next_provider: 'deepseek',
-      next_model: 'deepseek:deepseek-v4-pro',
-    };
-    const event: AIChatStreamEvent = { type: 'workflow_state', data: workflow };
-
-    expect(event.data.failed_model).toBe('openrouter:qwen/qwen3.7-plus');
-    expect(event.data.next_model).toBe('deepseek:deepseek-v4-pro');
-    expect(aiChatEventsSource).toContain('failed_model?: string');
-    expect(aiChatEventsSource).toContain('next_model?: string');
+  it('does not expose automatic provider fallback metadata on workflow state events', () => {
+    expect(aiChatEventsSource).not.toContain('provider_fallback');
+    expect(aiChatEventsSource).not.toContain('failed_provider?: string');
+    expect(aiChatEventsSource).not.toContain('failed_model?: string');
+    expect(aiChatEventsSource).not.toContain('next_provider?: string');
+    expect(aiChatEventsSource).not.toContain('next_model?: string');
   });
 
   it('exposes selected provider and model metadata on workflow state events', () => {

frontend-modern/src/api/aiChatDevStreamFixture.ts

@@ -620,36 +620,25 @@ const buildProviderRetryFixtureEvents = (): AIChatStreamEvent[] => [
     type: 'workflow_state',
     data: {
       phase: 'provider_retry',
-      message: 'DeepSeek failed before output; retrying through OpenRouter.',
+      message: 'Provider connection failed before any output; retrying.',
       provider: 'deepseek',
       model: 'deepseek:deepseek-chat',
-      failed_model: 'deepseek:deepseek-chat',
-      next_model: 'openrouter:deepseek/deepseek-chat',
       attempt: 2,
       max_attempts: 3,
       retry_after_ms: 3200,
     },
   },
-  {
-    type: 'workflow_state',
-    data: {
-      phase: 'provider_start',
-      message: 'Retrying through OpenRouter with deepseek/deepseek-chat.',
-      provider: 'openrouter',
-      model: 'openrouter:deepseek/deepseek-chat',
-    },
-  },
   {
     type: 'content',
     data: {
-      text: 'The provider retry fixture switched to OpenRouter after the direct provider failed before output.',
+      text: 'The provider retry fixture retried the selected route after a transient startup failure.',
     },
   },
   {
     type: 'done',
     data: {
       session_id: 'dev-fixture-provider-retry',
-      model: 'openrouter:deepseek/deepseek-chat',
+      model: 'deepseek:deepseek-chat',
       input_tokens: 73,
       output_tokens: 29,
     },

frontend-modern/src/api/generated/aiChatEvents.ts

@@ -133,10 +133,6 @@ export interface WorkflowStateData {
   attempt?: number;
   max_attempts?: number;
   retry_after_ms?: number;
-  failed_provider?: string;
-  failed_model?: string;
-  next_provider?: string;
-  next_model?: string;
 }
 
 export type AIChatStreamEvent =

frontend-modern/src/components/AI/Chat/MessageItem.tsx

@@ -350,7 +350,7 @@ export const MessageItem: Component<MessageItemProps> = (props) => {
     if (!model) return '';
     return props.getModelRouteLabel?.(model) || formatAIModelRouteLabel(model);
   };
-  const isProviderFallbackEvent = (event: StreamDisplayEvent) => {
+  const hasPreviousModelRoute = (event: StreamDisplayEvent) => {
     const model = event.model?.trim();
     const failed = event.failedModel?.trim();
     return !!model && !!failed && failed !== model;
@@ -358,7 +358,7 @@ export const MessageItem: Component<MessageItemProps> = (props) => {
   const isSelectedModelEvent = (event: StreamDisplayEvent) =>
     event.modelEvent === 'selected' && !!event.model?.trim();
   const modelSwitchTitle = (event: StreamDisplayEvent) => {
-    if (!isProviderFallbackEvent(event)) return modelRouteLabel(event.model);
+    if (!hasPreviousModelRoute(event)) return modelRouteLabel(event.model);
     return `${modelRouteLabel(event.failedModel)} -> ${modelRouteLabel(event.model)}`;
   };
   const messageModelLabel = () => modelRouteLabel(props.message.model);
@@ -673,11 +673,9 @@ export const MessageItem: Component<MessageItemProps> = (props) => {
                                 class="my-2 inline-flex max-w-full items-center gap-2 rounded-md border border-border-subtle bg-surface-alt px-2.5 py-1.5 text-xs text-muted"
                                 role="status"
                                 aria-label={
-                                  isProviderFallbackEvent(event)
-                                    ? 'Assistant provider fallback route changed'
-                                    : isSelectedModelEvent(event)
-                                      ? 'Assistant model route selected'
-                                      : 'Assistant model route changed'
+                                  isSelectedModelEvent(event)
+                                    ? 'Assistant model route selected'
+                                    : 'Assistant model route changed'
                                 }
                                 title={modelSwitchTitle(event)}
                               >
@@ -686,7 +684,7 @@ export const MessageItem: Component<MessageItemProps> = (props) => {
                                   aria-hidden="true"
                                 />
                                 <Show
-                                  when={isProviderFallbackEvent(event)}
+                                  when={hasPreviousModelRoute(event)}
                                   fallback={
                                     <>
                                       <Show when={isSelectedModelEvent(event)}>
@@ -701,7 +699,7 @@ export const MessageItem: Component<MessageItemProps> = (props) => {
                                     </>
                                   }
                                 >
-                                  <span class="shrink-0">Provider fallback</span>
+                                  <span class="shrink-0">Switched from</span>
                                   <span class="min-w-0 truncate font-medium text-base-content">
                                     {modelRouteLabel(event.failedModel)}
                                   </span>

frontend-modern/src/components/AI/Chat/__tests__/AIChat.test.tsx

@@ -3018,7 +3018,10 @@ describe('AIChat', () => {
         expect(mockChat.loadSession).toHaveBeenCalledWith('source-session');
       });
       expect(mockAIChatAPI.listSessions).toHaveBeenCalledWith({ limit: 30 });
-      expect(mockNotificationStore.info).toHaveBeenCalledWith('Compacting Assistant session...', 2000);
+      expect(mockNotificationStore.info).toHaveBeenCalledWith(
+        'Compacting Assistant session...',
+        2000,
+      );
       await waitFor(() => {
         expect(mockNotificationStore.success).toHaveBeenCalledWith(
           'Compacted 8 older messages into a session summary.',
@@ -4552,7 +4555,7 @@ describe('AIChat', () => {
       });
     });
 
-    it('adopts a successful provider fallback route for the active session', async () => {
+    it('does not adopt a completed route-switch row for the active session automatically', async () => {
       const [messages, setMessages] = createSignal<ChatMessage[]>([]);
       mockChat.messages.mockImplementation(() => messages());
       mockChat.sessionId.mockReturnValue('session-1');
@@ -4569,7 +4572,7 @@ describe('AIChat', () => {
 
       setMessages([
         {
-          id: 'assistant-fallback',
+          id: 'assistant-route-switch',
           role: 'assistant',
           content: '',
           timestamp: new Date('2026-06-06T12:00:00Z'),
@@ -4589,9 +4592,9 @@ describe('AIChat', () => {
 
       setMessages([
         {
-          id: 'assistant-fallback',
+          id: 'assistant-route-switch',
           role: 'assistant',
-          content: 'DeepSeek is reachable through OpenRouter.',
+          content: 'The response completed on another route.',
           timestamp: new Date('2026-06-06T12:00:00Z'),
           completedAt: new Date('2026-06-06T12:00:03Z'),
           model: 'openrouter:deepseek/deepseek-v4-pro',
@@ -4604,34 +4607,27 @@ describe('AIChat', () => {
             },
             {
               type: 'content',
-              content: 'DeepSeek is reachable through OpenRouter.',
+              content: 'The response completed on another route.',
             },
           ],
         },
       ]);
 
-      await waitFor(() => {
-        expect(mockChat.setModel).toHaveBeenCalledWith('openrouter:deepseek/deepseek-v4-pro');
-      });
       expect(
-        screen.getByRole('status', { name: 'Assistant fallback route adopted' }),
-      ).toHaveTextContent('after fallback from');
-      expect(
-        screen.getByRole('button', { name: 'Dismiss fallback route notice' }),
-      ).toBeInTheDocument();
-      expect(mockNotificationStore.success).toHaveBeenCalledWith(
+        screen.queryByRole('status', { name: 'Assistant fallback route adopted' }),
+      ).not.toBeInTheDocument();
+      expect(mockChat.setModel).not.toHaveBeenCalledWith('openrouter:deepseek/deepseek-v4-pro');
+      expect(mockNotificationStore.success).not.toHaveBeenCalledWith(
         expect.stringContaining('Assistant model route switched to'),
-        2500,
+        expect.any(Number),
       );
       expect(localStorage.getItem('pulse:ai_chat_models_by_session')).toBe(
-        JSON.stringify({ 'session-1': 'openrouter:deepseek/deepseek-v4-pro' }),
-      );
-      expect(localStorage.getItem('pulse:ai_chat_recent_models')).toBe(
-        JSON.stringify(['openrouter:deepseek/deepseek-v4-pro']),
+        JSON.stringify({ 'session-1': 'deepseek:deepseek-v4-pro' }),
       );
+      expect(localStorage.getItem('pulse:ai_chat_recent_models')).toBeNull();
     });
 
-    it('does not override a user-selected route when a fallback turn completes', async () => {
+    it('does not override a user-selected route when a route-switch row completes', async () => {
       const [messages, setMessages] = createSignal<ChatMessage[]>([]);
       const [selectedModel, setSelectedModel] = createSignal('deepseek:deepseek-v4-pro');
       mockChat.messages.mockImplementation(() => messages());
@@ -4652,7 +4648,7 @@ describe('AIChat', () => {
 
       setMessages([
         {
-          id: 'assistant-fallback',
+          id: 'assistant-route-switch',
           role: 'assistant',
           content: '',
           timestamp: new Date('2026-06-06T12:00:00Z'),
@@ -4672,9 +4668,9 @@ describe('AIChat', () => {
 
       setMessages([
         {
-          id: 'assistant-fallback',
+          id: 'assistant-route-switch',
           role: 'assistant',
-          content: 'Fallback answer.',
+          content: 'Route switch answer.',
           timestamp: new Date('2026-06-06T12:00:00Z'),
           completedAt: new Date('2026-06-06T12:00:03Z'),
           model: 'openrouter:deepseek/deepseek-v4-pro',
@@ -4687,7 +4683,7 @@ describe('AIChat', () => {
             },
             {
               type: 'content',
-              content: 'Fallback answer.',
+              content: 'Route switch answer.',
             },
           ],
         },
@@ -5306,9 +5302,7 @@ describe('AIChat', () => {
 
       const status = screen.getByLabelText('Assistant active turn status');
       expect(status).toHaveTextContent('Preparing Pulse context.');
-      expect(status).not.toHaveTextContent(
-        'OpenRouter is starting the response.',
-      );
+      expect(status).not.toHaveTextContent('OpenRouter is starting the response.');
 
       await vi.advanceTimersByTimeAsync(650);
       expect(status).toHaveTextContent('Reading current Pulse inventory.');
@@ -5381,15 +5375,11 @@ describe('AIChat', () => {
       expect(status).not.toHaveTextContent('OpenRouter is starting the response.');
 
       await vi.advanceTimersByTimeAsync(650);
-      expect(status).toHaveTextContent(
-        'Reading current Pulse inventory. · 1 follow-up queued',
-      );
+      expect(status).toHaveTextContent('Reading current Pulse inventory. · 1 follow-up queued');
       expect(status).not.toHaveTextContent('pulse_query');
 
       await vi.advanceTimersByTimeAsync(650);
-      expect(status).toHaveTextContent(
-        'OpenRouter is starting the response. · 1 follow-up queued',
-      );
+      expect(status).toHaveTextContent('OpenRouter is starting the response. · 1 follow-up queued');
       expect(status).not.toHaveTextContent('Preparing Pulse context.');
     });
 
@@ -5571,7 +5561,9 @@ describe('AIChat', () => {
 
       await waitFor(() => {
         const activityDock = screen.getByTestId('assistant-activity-dock');
-        expect(activityDock).toContainElement(screen.getByLabelText('Assistant active turn status'));
+        expect(activityDock).toContainElement(
+          screen.getByLabelText('Assistant active turn status'),
+        );
         expect(activityDock).toContainElement(
           screen.getByRole('status', { name: 'Assistant autonomous control warning' }),
         );

frontend-modern/src/components/AI/Chat/__tests__/MessageItem.test.tsx

@@ -1023,7 +1023,7 @@ describe('MessageItem', () => {
       expect(screen.queryByText(/Hidden reasoning/i)).not.toBeInTheDocument();
     });
 
-    it('renders provider fallback model switches as typed transcript status', () => {
+    it('renders model switches as typed transcript status', () => {
       const events: StreamDisplayEvent[] = [
         { type: 'model_switch', model: 'openrouter:deepseek/deepseek-v4-pro' },
       ];
@@ -1112,7 +1112,7 @@ describe('MessageItem', () => {
       expect(screen.getByText(/Preparing Pulse context\./)).toBeInTheDocument();
     });
 
-    it('renders provider fallback model switches with failed and next routes', () => {
+    it('renders model route switches with previous and next routes', () => {
       const events: StreamDisplayEvent[] = [
         {
           type: 'model_switch',
@@ -1136,9 +1136,9 @@ describe('MessageItem', () => {
       ));
 
       const status = screen.getByRole('status', {
-        name: 'Assistant provider fallback route changed',
+        name: 'Assistant model route changed',
       });
-      expect(status).toHaveTextContent('Provider fallback');
+      expect(status).toHaveTextContent('Switched from');
       expect(status).toHaveTextContent('OpenAI GPT-4o Mini via OpenRouter');
       expect(status).toHaveTextContent('Gemini 3.1 Flash Lite');
       expect(status).toHaveAttribute(

frontend-modern/src/components/AI/Chat/__tests__/activeTurnStatus.test.ts

@@ -36,11 +36,7 @@ describe('getAssistantActiveTurnStatus', () => {
 
   it('tracks startup timing from the submitted user turn', () => {
     expect(
-      getAssistantActiveTurnStatus(
-        [userMessage({ timestamp: new Date(1_000) })],
-        true,
-        4_000,
-      ),
+      getAssistantActiveTurnStatus([userMessage({ timestamp: new Date(1_000) })], true, 4_000),
     ).toEqual({
       type: 'thinking',
       text: 'Sending prompt',
@@ -183,8 +179,7 @@ describe('getAssistantActiveTurnStatus', () => {
       ),
     ).toEqual({
       type: 'tool',
-      text:
-        'Running $ curl -H "Authorization: Bearer [redacted-secret]" --password [redacted-secret] https://example.local',
+      text: 'Running $ curl -H "Authorization: Bearer [redacted-secret]" --password [redacted-secret] https://example.local',
     });
   });
 
@@ -652,7 +647,7 @@ describe('getAssistantActiveTurnStatus', () => {
     });
   });
 
-  it('shows failed and next model routes for provider fallback status', () => {
+  it('shows previous and next model routes for explicit route switches', () => {
     expect(
       getAssistantActiveTurnStatus(
         [
@@ -672,7 +667,7 @@ describe('getAssistantActiveTurnStatus', () => {
       ),
     ).toEqual({
       type: 'thinking',
-      text: 'Provider fallback: OpenAI: GPT 4o Mini via OpenRouter -> DeepSeek: DeepSeek V4 Pro via OpenRouter',
+      text: 'Switched from OpenAI: GPT 4o Mini via OpenRouter to DeepSeek: DeepSeek V4 Pro via OpenRouter',
       startedAt: 2_000,
     });
   });

frontend-modern/src/components/AI/Chat/__tests__/useChat.test.ts

@@ -1279,7 +1279,7 @@ describe('useChat', () => {
       dispose();
     });
 
-    it('updates the in-flight message model when provider fallback starts', async () => {
+    it('ignores obsolete provider fallback workflow metadata', async () => {
       const { getFireEvent } = setupWithEventCapture();
       const { value: chat, dispose } = withRoot(() =>
         useChat({ sessionId: 's', model: 'openrouter:openai/gpt-4o-mini' }),
@@ -1299,21 +1299,14 @@ describe('useChat', () => {
       });
 
       const assistant = chat.messages().find((m) => m.role === 'assistant')!;
-      expect(assistant.model).toBe('gemini:gemini-3.1-flash-lite');
-      expect(assistant.streamEvents).toContainEqual(
-        expect.objectContaining({
-          type: 'model_switch',
-          model: 'gemini:gemini-3.1-flash-lite',
-          failedModel: 'openrouter:openai/gpt-4o-mini',
-          startedAt: expect.any(Number),
-          updatedAt: expect.any(Number),
-        }),
-      );
-      expect(assistant.workflowStatus).toEqual(
-        expect.objectContaining({
-          phase: 'provider_fallback',
-          message: 'OpenRouter did not start a response; trying Gemini.',
-        }),
+      expect(assistant.model).toBe('openrouter:openai/gpt-4o-mini');
+      expect(assistant.streamEvents).not.toEqual(
+        expect.arrayContaining([
+          expect.objectContaining({
+            type: 'model_switch',
+            model: 'gemini:gemini-3.1-flash-lite',
+          }),
+        ]),
       );
       dispose();
     });
@@ -1369,7 +1362,7 @@ describe('useChat', () => {
       dispose();
     });
 
-    it('does not infer a failed route when workflow only carries a next model', async () => {
+    it('does not infer a model switch from legacy next-model workflow metadata', async () => {
       const { getFireEvent } = setupWithEventCapture();
       const { value: chat, dispose } = withRoot(() =>
         useChat({ sessionId: 's', model: 'openrouter:openai/gpt-4o-mini' }),
@@ -1388,13 +1381,14 @@ describe('useChat', () => {
       });
 
       const assistant = chat.messages().find((m) => m.role === 'assistant')!;
-      expect(assistant.model).toBe('gemini:gemini-3.1-flash-lite');
-      expect(assistant.streamEvents).toContainEqual(
-        expect.objectContaining({
-          type: 'model_switch',
-          model: 'gemini:gemini-3.1-flash-lite',
-          failedModel: undefined,
-        }),
+      expect(assistant.model).toBe('openrouter:openai/gpt-4o-mini');
+      expect(assistant.streamEvents).not.toEqual(
+        expect.arrayContaining([
+          expect.objectContaining({
+            type: 'model_switch',
+            model: 'gemini:gemini-3.1-flash-lite',
+          }),
+        ]),
       );
       dispose();
     });
@@ -2146,7 +2140,8 @@ describe('useChat', () => {
 
       await chat.sendMessage('hi');
       const fire = getFireEvent();
-      const input = '{"action":"exec","command":"ls /dev | wc -l","target_host":"current_resource"}';
+      const input =
+        '{"action":"exec","command":"ls /dev | wc -l","target_host":"current_resource"}';
 
       fire({
         type: 'tool_start',

frontend-modern/src/components/AI/Chat/activeTurnStatus.ts

@@ -522,7 +522,7 @@ const modelSwitchStatusText = (event: StreamDisplayEvent): string => {
   const failed = event.failedModel?.trim();
   if (event.modelEvent === 'selected') return `Using ${next}`;
   if (!failed || failed === model) return `Switched to ${next}`;
-  return `Provider fallback: ${formatAIModelRouteLabel(failed)} -> ${next}`;
+  return `Switched from ${formatAIModelRouteLabel(failed)} to ${next}`;
 };
 
 const hasVisibleAssistantOutput = (message: ChatMessage): boolean => {
@@ -664,5 +664,8 @@ export const getAssistantActiveTurnStatus = (
     );
   }
 
-  return withAssistantQueuedFollowUpStatus(initialRequestStatus(messages, assistantMessage), messages);
+  return withAssistantQueuedFollowUpStatus(
+    initialRequestStatus(messages, assistantMessage),
+    messages,
+  );
 };

frontend-modern/src/components/AI/Chat/hooks/useChat.ts

@@ -496,10 +496,8 @@ export function useChat(options: UseChatOptions = {}) {
   };
 
   const streamModelEventKind = (event: StreamDisplayEvent): StreamDisplayEvent['modelEvent'] => {
-    if (event.modelEvent) return event.modelEvent;
-    const model = event.model?.trim();
-    const failed = event.failedModel?.trim();
-    return model && failed && model !== failed ? 'fallback' : 'switch';
+    if (event.modelEvent && event.modelEvent !== 'fallback') return event.modelEvent;
+    return 'switch';
   };
 
   const isDurableAssistantStreamBoundary = (event: StreamDisplayEvent): boolean =>
@@ -924,20 +922,6 @@ export function useChat(options: UseChatOptions = {}) {
 
   const extractWorkflowModel = (data: unknown): string => extractCompletedModel(data);
 
-  const extractWorkflowNextModel = (data: unknown): string => {
-    if (!data || typeof data !== 'object') return '';
-    const record = data as Record<string, unknown>;
-    const nextModel = record.next_model ?? record.nextModel;
-    return typeof nextModel === 'string' ? nextModel.trim() : '';
-  };
-
-  const extractWorkflowFailedModel = (data: unknown): string => {
-    if (!data || typeof data !== 'object') return '';
-    const record = data as Record<string, unknown>;
-    const failedModel = record.failed_model ?? record.failedModel;
-    return typeof failedModel === 'string' ? failedModel.trim() : '';
-  };
-
   const extractErrorMessage = (data: unknown): string => {
     if (typeof data === 'string') return data;
     if (data && typeof data === 'object') {
@@ -954,6 +938,7 @@ export function useChat(options: UseChatOptions = {}) {
     if (!message) return null;
 
     const phase = typeof record.phase === 'string' ? record.phase.trim() : '';
+    if (phase === 'provider_fallback') return null;
     const state = typeof record.state === 'string' ? record.state.trim() : '';
     const tool = typeof record.tool === 'string' ? record.tool.trim() : '';
     const attempt = positiveNumber(record.attempt);
@@ -1182,25 +1167,14 @@ export function useChat(options: UseChatOptions = {}) {
 
     if (event.type === 'workflow_state') {
       const workflowStatus = extractWorkflowStatus(event.data);
-      const nextModel = extractWorkflowNextModel(event.data);
-      const failedModel = extractWorkflowFailedModel(event.data);
       const startedModel =
         workflowStatus?.phase === 'provider_start' ? extractWorkflowModel(event.data) : '';
-      const routeEvent = nextModel
+      const routeEvent = startedModel
         ? {
-            model: nextModel,
-            failedModel,
-            modelEvent: (failedModel ? 'fallback' : 'switch') as NonNullable<
-              StreamDisplayEvent['modelEvent']
-            >,
+            model: startedModel,
+            modelEvent: 'selected' as const,
           }
-        : startedModel
-          ? {
-              model: startedModel,
-              failedModel: '',
-              modelEvent: 'selected' as const,
-            }
-          : null;
+        : null;
       if (!workflowStatus && !routeEvent) return;
 
       if (routeEvent) {
@@ -1208,7 +1182,6 @@ export function useChat(options: UseChatOptions = {}) {
           prev.map((msg) => {
             if (msg.id !== assistantId) return msg;
             return withModelRouteEvent(msg, routeEvent.model, {
-              failedModel: routeEvent.failedModel || undefined,
               modelEvent: routeEvent.modelEvent,
             });
           }),
@@ -1480,7 +1453,11 @@ export function useChat(options: UseChatOptions = {}) {
                   ) {
                     const startedAt = evt.pendingTool?.startedAt || evt.startedAt;
                     const settleUntil = newToolCall.success
-                      ? getAssistantFastToolCompletionSettleUntil(startedAt, completedAt, completedAt)
+                      ? getAssistantFastToolCompletionSettleUntil(
+                          startedAt,
+                          completedAt,
+                          completedAt,
+                        )
                       : undefined;
                     updatedEvents[i] = {
                       type: 'tool',
@@ -1505,8 +1482,8 @@ export function useChat(options: UseChatOptions = {}) {
               }
 
               // Also remove from pendingApprovals if present
-              const updatedApprovals = (msg.pendingApprovals || []).filter((a) =>
-                !matchesCompletedTool(a.toolId, a.toolName),
+              const updatedApprovals = (msg.pendingApprovals || []).filter(
+                (a) => !matchesCompletedTool(a.toolId, a.toolName),
               );
 
               return {

frontend-modern/src/components/AI/Chat/index.tsx

@@ -268,17 +268,6 @@ interface AIChatProps {
 
 let stashedComposerDraft: ComposerDraftStash | null = null;
 
-interface AssistantFallbackRouteAdoptionCandidate {
-  route: string;
-  failedModel: string;
-}
-
-interface AssistantFallbackRouteNotice extends AssistantFallbackRouteAdoptionCandidate {
-  failedModelLabel: string;
-  messageId: string;
-  routeLabel: string;
-}
-
 interface AssistantProviderReadinessRouteNotice {
   failedProviderLabel: string;
   failedRouteLabel: string;
@@ -307,34 +296,6 @@ const hasProviderRouteFailureEvidence = (message: ChatMessage): boolean => {
   );
 };
 
-const getAssistantFallbackRouteAdoptionCandidate = (
-  message: ChatMessage,
-): AssistantFallbackRouteAdoptionCandidate | null => {
-  if (message.role !== 'assistant' || message.error) return null;
-
-  const events = [...(message.streamEvents || [])].reverse();
-  const modelSwitch = events.find(
-    (event) =>
-      event.type === 'model_switch' &&
-      typeof event.model === 'string' &&
-      event.model.trim() &&
-      typeof event.failedModel === 'string' &&
-      event.failedModel.trim(),
-  );
-  if (!modelSwitch) return null;
-
-  const route = modelSwitch.model?.trim() || '';
-  const failedModel = modelSwitch.failedModel?.trim() || '';
-  if (!route || !failedModel) return null;
-
-  const completedModel = (message.model || '').trim();
-  if (message.isStreaming === false && completedModel && completedModel !== route) {
-    return null;
-  }
-
-  return { route, failedModel };
-};
-
 const compactText = (items: Array<string | undefined>): string[] =>
   items.filter((item): item is string => typeof item === 'string' && item.trim().length > 0);
 
@@ -740,10 +701,6 @@ export const AIChat: Component<AIChatProps> = (props) => {
   const [modelSelectorOpenRequest, setModelSelectorOpenRequest] = createSignal(0);
   const [defaultModel, setDefaultModel] = createSignal('');
   const [chatOverrideModel, setChatOverrideModel] = createSignal('');
-  const pendingFallbackRouteAdoptions = new Map<string, AssistantFallbackRouteAdoptionCandidate>();
-  const adoptedFallbackRouteMessageIds = new Set<string>();
-  const [fallbackRouteNotice, setFallbackRouteNotice] =
-    createSignal<AssistantFallbackRouteNotice | null>(null);
   const [providerReadinessRouteNotice, setProviderReadinessRouteNotice] =
     createSignal<AssistantProviderReadinessRouteNotice | null>(null);
   const [providerReadiness, setProviderReadiness] = createSignal<ChatProviderReadinessState>({
@@ -2137,7 +2094,6 @@ export const AIChat: Component<AIChatProps> = (props) => {
     if (options.rememberRecent !== false) {
       rememberRecentModel(modelId);
     }
-    setFallbackRouteNotice(null);
     setProviderReadinessRouteNotice(null);
   };
 
@@ -2272,59 +2228,6 @@ export const AIChat: Component<AIChatProps> = (props) => {
     return alternative;
   };
 
-  createEffect(() => {
-    const messages = chat.messages();
-    const visibleMessageIds = new Set(messages.map((message) => message.id));
-    for (const messageId of [...pendingFallbackRouteAdoptions.keys()]) {
-      if (!visibleMessageIds.has(messageId)) {
-        pendingFallbackRouteAdoptions.delete(messageId);
-      }
-    }
-    for (const messageId of [...adoptedFallbackRouteMessageIds]) {
-      if (!visibleMessageIds.has(messageId)) {
-        adoptedFallbackRouteMessageIds.delete(messageId);
-      }
-    }
-
-    for (const message of messages) {
-      const candidate = getAssistantFallbackRouteAdoptionCandidate(message);
-      if (!candidate) continue;
-
-      if (message.isStreaming !== false) {
-        pendingFallbackRouteAdoptions.set(message.id, candidate);
-        continue;
-      }
-
-      const pending = pendingFallbackRouteAdoptions.get(message.id);
-      if (
-        !pending ||
-        pending.route !== candidate.route ||
-        adoptedFallbackRouteMessageIds.has(message.id)
-      ) {
-        continue;
-      }
-
-      const currentRoute = selectedChatModel().trim();
-      if (currentRoute && currentRoute !== pending.failedModel) {
-        pendingFallbackRouteAdoptions.delete(message.id);
-        continue;
-      }
-
-      adoptedFallbackRouteMessageIds.add(message.id);
-      pendingFallbackRouteAdoptions.delete(message.id);
-      selectModel(candidate.route);
-      const routeLabel = formatChatMessageModelRoute(candidate.route);
-      const failedModelLabel = formatChatMessageModelRoute(pending.failedModel);
-      setFallbackRouteNotice({
-        ...candidate,
-        failedModelLabel,
-        messageId: message.id,
-        routeLabel,
-      });
-      notificationStore.success(`Assistant model route switched to ${routeLabel}`, 2500);
-    }
-  });
-
   createEffect(() => {
     const sessionId = chat.sessionId();
     const storedModel = getStoredModel(sessionId);
@@ -3108,7 +3011,6 @@ export const AIChat: Component<AIChatProps> = (props) => {
     const submittedInput = readComposerInputForSubmit();
     const prompt = submittedInput.trim();
     if (!prompt) return;
-    setFallbackRouteNotice(null);
     setProviderReadinessRouteNotice(null);
     composerSubmitDispatchLocked = true;
     queueMicrotask(() => {
@@ -3438,7 +3340,6 @@ export const AIChat: Component<AIChatProps> = (props) => {
     resetPromptHistoryNavigation();
     setEditingQueuedFollowUp(null);
     setRestoredPromptDraft(null);
-    setFallbackRouteNotice(null);
     setProviderReadinessRouteNotice(null);
     setRedoLastTurnAvailable(false);
     aiChatStore.clearContext?.();
@@ -3533,7 +3434,8 @@ export const AIChat: Component<AIChatProps> = (props) => {
       focusComposer();
     } catch (error) {
       logger.error('[AIChat] Failed to compact Assistant session:', error);
-      const message = error instanceof Error ? error.message : AI_CHAT_COMPACT_SESSION_ERROR_MESSAGE;
+      const message =
+        error instanceof Error ? error.message : AI_CHAT_COMPACT_SESSION_ERROR_MESSAGE;
       notificationStore.error(message);
     } finally {
       setCompactingSession(false);
@@ -3703,7 +3605,6 @@ export const AIChat: Component<AIChatProps> = (props) => {
         resetPromptHistoryNavigation();
         setEditingQueuedFollowUp(null);
         setRestoredPromptDraft(null);
-        setFallbackRouteNotice(null);
         setProviderReadinessRouteNotice(null);
         setRedoLastTurnAvailable(false);
       }
@@ -4482,7 +4383,6 @@ export const AIChat: Component<AIChatProps> = (props) => {
               when={
                 currentStatus() ||
                 autonomousWarningVisible() ||
-                fallbackRouteNotice() ||
                 providerReadinessRouteNotice() ||
                 chat.queuedFollowUpCount() > 0
               }
@@ -4492,9 +4392,7 @@ export const AIChat: Component<AIChatProps> = (props) => {
                 data-testid="assistant-activity-dock"
               >
                 <Show when={currentStatus()}>
-                  <div
-                    class="flex min-h-8 min-w-0 items-center gap-2 px-2.5 py-1.5 text-xs"
-                  >
+                  <div class="flex min-h-8 min-w-0 items-center gap-2 px-2.5 py-1.5 text-xs">
                     <div
                       class="flex min-w-0 flex-1 items-center gap-2"
                       role="status"
@@ -4509,9 +4407,7 @@ export const AIChat: Component<AIChatProps> = (props) => {
                         }`}
                         aria-hidden="true"
                       />
-                      <span class="min-w-0 flex-1 truncate font-medium">
-                        {currentStatusText()}
-                      </span>
+                      <span class="min-w-0 flex-1 truncate font-medium">{currentStatusText()}</span>
                       <span class="flex shrink-0 gap-0.5" aria-hidden="true">
                         <span
                           class="h-1 w-1 rounded-full bg-blue-400 animate-bounce"
@@ -4590,45 +4486,11 @@ export const AIChat: Component<AIChatProps> = (props) => {
                     </button>
                   </div>
                 </Show>
-                <Show when={fallbackRouteNotice()}>
-                  {(notice) => (
-                    <div
-                      class={`flex min-h-8 min-w-0 items-center gap-2 px-2.5 py-1.5 text-xs ${
-                        currentStatus() || autonomousWarningVisible()
-                          ? 'border-t border-border/70'
-                          : ''
-                      }`}
-                      role="status"
-                      aria-label="Assistant fallback route adopted"
-                      aria-live="polite"
-                    >
-                      <CheckIcon
-                        class="h-3.5 w-3.5 shrink-0 text-emerald-600 dark:text-emerald-300"
-                        aria-hidden="true"
-                      />
-                      <span class="min-w-0 flex-1 truncate font-medium">
-                        Using {notice().routeLabel} after fallback from {notice().failedModelLabel}
-                      </span>
-                      <button
-                        type="button"
-                        onClick={() => {
-                          setFallbackRouteNotice(null);
-                          focusComposer();
-                        }}
-                        class="inline-flex h-6 w-6 shrink-0 items-center justify-center rounded-md text-blue-700 transition-colors hover:bg-blue-100 hover:text-blue-900 dark:text-blue-200 dark:hover:bg-blue-900/50"
-                        title="Dismiss fallback route notice"
-                        aria-label="Dismiss fallback route notice"
-                      >
-                        <XIcon class="h-3.5 w-3.5" aria-hidden="true" />
-                      </button>
-                    </div>
-                  )}
-                </Show>
                 <Show when={providerReadinessRouteNotice()}>
                   {(notice) => (
                     <div
                       class={`flex min-h-8 min-w-0 items-center gap-2 px-2.5 py-1.5 text-xs ${
-                        currentStatus() || autonomousWarningVisible() || fallbackRouteNotice()
+                        currentStatus() || autonomousWarningVisible()
                           ? 'border-t border-border/70'
                           : ''
                       }`}
@@ -4667,7 +4529,6 @@ export const AIChat: Component<AIChatProps> = (props) => {
                     class={`px-2.5 py-1.5 ${
                       currentStatus() ||
                       autonomousWarningVisible() ||
-                      fallbackRouteNotice() ||
                       providerReadinessRouteNotice()
                         ? 'border-t border-border/70'
                         : ''

internal/ai/chat/agentic.go

@@ -7,7 +7,6 @@ import (
 	"runtime/debug"
 	"strings"
 	"sync"
-	"sync/atomic"
 	"time"
 
 	"github.com/google/uuid"
@@ -28,8 +27,6 @@ type parallelToolResult struct {
 	Err    error
 }
 
-const providerFallbackStartupTimeout = 4 * time.Second
-
 func isRetryableProviderStreamError(err error) bool {
 	if err == nil {
 		return false
@@ -69,15 +66,6 @@ func isRetryableProviderStreamError(err error) bool {
 	return false
 }
 
-func providerStreamEventShowsStartup(event providers.StreamEvent) bool {
-	switch strings.ToLower(strings.TrimSpace(event.Type)) {
-	case "content", "done", "thinking", "tool_progress", "tool_start":
-		return true
-	default:
-		return false
-	}
-}
-
 type workflowStateOption func(*WorkflowStateData)
 
 func withWorkflowRetry(nextAttempt, maxAttempts int, retryAfter time.Duration) workflowStateOption {
@@ -565,14 +553,10 @@ type AgenticLoop struct {
 	requestSanitizer func(providers.ChatRequest) providers.ChatRequest
 
 	// When true, provider terminal errors are returned to the caller without
-	// emitting a stream error event. The chat service uses this to try another
-	// configured provider before the browser sees a failed first attempt.
+	// emitting a stream error event. The chat service uses this to centralize
+	// terminal provider error presentation at the stream boundary.
 	suppressProviderErrorEvents bool
 
-	// When true, this loop is running as one attempt in a service-owned provider
-	// fallback chain, so hidden startup retries should yield to the next route.
-	fastFailProviderStartup bool
-
 	// Query-only count/overview turns should not let a provider accidentally
 	// expand a topology request into the full infrastructure tree.
 	preferSummaryOnlyQueries bool
@@ -619,12 +603,6 @@ func (a *AgenticLoop) SetSuppressProviderErrorEvents(suppress bool) {
 	a.suppressProviderErrorEvents = suppress
 }
 
-func (a *AgenticLoop) SetFastFailProviderStartup(enabled bool) {
-	a.mu.Lock()
-	defer a.mu.Unlock()
-	a.fastFailProviderStartup = enabled
-}
-
 func (a *AgenticLoop) SetPreferSummaryOnlyQueries(prefer bool) {
 	a.mu.Lock()
 	defer a.mu.Unlock()
@@ -722,7 +700,6 @@ func (a *AgenticLoop) executeWithTools(ctx context.Context, sessionID string, me
 		modelName := a.modelName
 		requestSanitizer := a.requestSanitizer
 		preferSummaryOnlyQueries := a.preferSummaryOnlyQueries
-		fastFailProviderStartup := a.fastFailProviderStartup
 		a.mu.Unlock()
 
 		// Record telemetry for loop iteration
@@ -755,9 +732,6 @@ func (a *AgenticLoop) executeWithTools(ctx context.Context, sessionID string, me
 			Tools:       tools,
 			ExecutionID: a.executionID,
 		}
-		if fastFailProviderStartup {
-			req.ProviderStartupMode = providers.ProviderStartupFastFailBeforeVisibleOutput
-		}
 
 		// Tool selection is model-owned. Pulse normally exposes the governed tool
 		// manifest unchanged. When a run must stop for safety or budget reasons,
@@ -913,36 +887,13 @@ func (a *AgenticLoop) executeWithTools(ctx context.Context, sessionID string, me
 		}
 
 		maxProviderAttempts := 2
-		if fastFailProviderStartup {
-			maxProviderAttempts = 1
-		}
 		err := error(nil)
 		for attempt := 1; attempt <= maxProviderAttempts; attempt++ {
 			attemptSawDone := false
 			attemptEmittedVisibleEvents := false
 			var attemptErrorMessages []string
-			providerCtx := ctx
-			var cancelProviderStartup context.CancelFunc
-			var startupTimedOut atomic.Bool
-			var startupTimer *time.Timer
-			stopStartupTimer := func() {}
-			if fastFailProviderStartup {
-				providerCtx, cancelProviderStartup = context.WithCancel(ctx)
-				startupTimer = time.AfterFunc(providerFallbackStartupTimeout, func() {
-					startupTimedOut.Store(true)
-					cancelProviderStartup()
-				})
-				stopStartupTimer = func() {
-					if startupTimer != nil {
-						startupTimer.Stop()
-					}
-				}
-			}
 
-			err = a.provider.ChatStream(providerCtx, req, func(event providers.StreamEvent) {
-				if providerStreamEventShowsStartup(event) {
-					stopStartupTimer()
-				}
+			err = a.provider.ChatStream(ctx, req, func(event providers.StreamEvent) {
 				switch event.Type {
 				case "content":
 					if data, ok := event.Data.(providers.ContentEvent); ok {
@@ -1070,14 +1021,6 @@ func (a *AgenticLoop) executeWithTools(ctx context.Context, sessionID string, me
 					}
 				}
 			})
-			stopStartupTimer()
-			if cancelProviderStartup != nil {
-				cancelProviderStartup()
-			}
-			if startupTimedOut.Load() && err != nil {
-				err = fmt.Errorf("provider startup timed out after %s: %w", providerFallbackStartupTimeout, err)
-			}
-
 			effectiveErr := err
 			if effectiveErr == nil && len(attemptErrorMessages) > 0 {
 				effectiveErr = fmt.Errorf("stream error: %s", attemptErrorMessages[0])

internal/ai/chat/service.go

@@ -805,31 +805,21 @@ func (s *Service) ExecuteStream(ctx context.Context, req ExecuteRequest, callbac
 		streamCallback(event)
 	}
 
-	attempts, initialProviderErr := s.chatProviderAttempts(ctx, cfgSnapshot, selectedModel, configuredModel, configuredProvider)
-	if len(attempts) == 0 {
-		if initialProviderErr != nil {
-			return initialProviderErr
-		}
-		return fmt.Errorf("provider not initialized")
-	}
-	initialFallbackModel := ""
-	if initialProviderErr != nil {
-		initialFallbackModel = attempts[0].Model
-	}
-	if initialFallbackModel != "" {
-		emitChatProviderFallback(streamCallback, selectedModel, initialFallbackModel)
+	attempt, providerErr := s.chatProviderAttempt(selectedModel, configuredModel, configuredProvider)
+	if providerErr != nil {
+		return providerErr
 	}
 
-	runAttempt := func(attempt chatProviderAttempt, hasFallback bool) ([]Message, *AgenticLoop, bool, error) {
+	runAttempt := func(attempt chatProviderAttempt) ([]Message, *AgenticLoop, error) {
 		attemptProvider := attempt.Provider
 		if attemptProvider == nil {
 			if strings.TrimSpace(attempt.Model) == "" {
-				return nil, nil, false, fmt.Errorf("no chat model configured")
+				return nil, nil, fmt.Errorf("no chat model configured")
 			}
 			var providerErr error
 			attemptProvider, providerErr = s.createProviderForModel(attempt.Model)
 			if providerErr != nil {
-				return nil, nil, false, fmt.Errorf("failed to create provider for model %q: %w", attempt.Model, providerErr)
+				return nil, nil, fmt.Errorf("failed to create provider for model %q: %w", attempt.Model, providerErr)
 			}
 		}
 
@@ -846,14 +836,13 @@ func (s *Service) ExecuteStream(ctx context.Context, req ExecuteRequest, callbac
 		}
 
 		// Create a per-attempt AgenticLoop to ensure complete isolation between
-		// concurrent sessions and provider fallback attempts. This prevents race
+		// concurrent sessions and chat attempts. This prevents race
 		// conditions where ExecuteStream calls overwrite each other's FSM,
 		// knowledge accumulator, autonomous mode, budget checker, and provider info.
 		systemPrompt := s.buildSystemPromptForOfferedTools(filteredTools)
 		loop := NewAgenticLoop(attemptProvider, executor, systemPrompt)
 		loop.SetOrgID(s.orgID)
 		loop.SetAutonomousMode(autonomousMode)
-		loop.SetFastFailProviderStartup(hasFallback)
 		loop.SetPreferSummaryOnlyQueries(assistantToolScope == assistantTurnToolScopeQueryOnly)
 		sanitizerOptions := []modelboundary.RequestSanitizerOption{}
 		if modelBoundaryAllowedInventoryContext != "" {
@@ -891,7 +880,6 @@ func (s *Service) ExecuteStream(ctx context.Context, req ExecuteRequest, callbac
 			Str("model", attempt.Model).
 			Msg("[ChatService] Set session FSM on agentic loop")
 
-		attemptVisible := false
 		attemptCallback := func(event StreamEvent) {
 			if event.Type == "question" {
 				var data QuestionData
@@ -899,9 +887,6 @@ func (s *Service) ExecuteStream(ctx context.Context, req ExecuteRequest, callbac
 					s.registerQuestionLoop(data.QuestionID, loop)
 				}
 			}
-			if chatStreamEventBlocksProviderFallback(event) {
-				attemptVisible = true
-			}
 			wrappedCallback(event)
 		}
 
@@ -918,36 +903,17 @@ func (s *Service) ExecuteStream(ctx context.Context, req ExecuteRequest, callbac
 			Err(err).
 			Msg("[ChatService] Agentic loop returned")
 
-		return resultMessages, loop, attemptVisible, err
+		return resultMessages, loop, err
 	}
 
 	var resultMessages []Message
 	var loop *AgenticLoop
 	var streamErr error
-	lastAttemptModel := selectedModel
-	for attemptIndex, attempt := range attempts {
-		var attemptVisible bool
-		lastAttemptModel = attempt.Model
-		resultMessages, loop, attemptVisible, streamErr = runAttempt(attempt, attemptIndex < len(attempts)-1)
-		if streamErr == nil {
-			selectedModel = attempt.Model
-			break
-		}
-		if attemptVisible || attemptIndex == len(attempts)-1 || ctx.Err() != nil {
-			emitChatProviderError(streamCallback, streamErr)
-			break
-		}
-		nextAttempt := attempts[attemptIndex+1]
-		log.Warn().
-			Err(streamErr).
-			Str("failed_model", attempt.Model).
-			Str("fallback_model", nextAttempt.Model).
-			Str("session_id", session.ID).
-			Msg("[ChatService] Provider failed before visible output; trying configured fallback provider")
-		emitChatProviderFallback(streamCallback, attempt.Model, nextAttempt.Model)
-	}
+	lastAttemptModel := attempt.Model
+	resultMessages, loop, streamErr = runAttempt(attempt)
 
 	if streamErr != nil {
+		emitChatProviderError(streamCallback, streamErr)
 		// Still save any messages we got
 		for _, msg := range resultMessages {
 			if msg.Role == "assistant" && strings.TrimSpace(msg.Model) == "" {
@@ -3041,126 +3007,21 @@ type chatProviderAttempt struct {
 	Provider providers.StreamingProvider
 }
 
-func (s *Service) chatProviderAttempts(_ context.Context, cfg *config.AIConfig, primaryModel, configuredModel string, configuredProvider providers.StreamingProvider) ([]chatProviderAttempt, error) {
-	var attempts []chatProviderAttempt
-	var primaryErr error
-	seen := make(map[string]struct{})
-	seenProviders := make(map[string]struct{})
-	addAttempt := func(model string, provider providers.StreamingProvider) {
-		model = strings.TrimSpace(model)
-		if model == "" && provider == nil {
-			return
-		}
-		key := model
-		if key == "" {
-			key = fmt.Sprintf("__provider_%d", len(attempts))
-		}
-		if _, ok := seen[key]; ok {
-			return
-		}
-		seen[key] = struct{}{}
-		if attemptProvider, _ := config.ParseModelString(model); attemptProvider != "" {
-			seenProviders[attemptProvider] = struct{}{}
-		}
-		attempts = append(attempts, chatProviderAttempt{Model: model, Provider: provider})
-	}
-
+func (s *Service) chatProviderAttempt(primaryModel, configuredModel string, configuredProvider providers.StreamingProvider) (chatProviderAttempt, error) {
 	primaryModel = strings.TrimSpace(primaryModel)
 	configuredModel = strings.TrimSpace(configuredModel)
 	if primaryModel == "" {
 		if configuredProvider != nil {
-			addAttempt("", configuredProvider)
-		} else {
-			primaryErr = fmt.Errorf("no chat model configured")
+			return chatProviderAttempt{Provider: configuredProvider}, nil
 		}
-	} else {
-		provider := configuredProvider
-		if primaryModel != configuredModel {
-			provider = nil
-		}
-		addAttempt(primaryModel, provider)
+		return chatProviderAttempt{}, fmt.Errorf("no chat model configured")
 	}
 
-	if cfg == nil {
-		return attempts, primaryErr
+	provider := configuredProvider
+	if primaryModel != configuredModel {
+		provider = nil
 	}
-
-	primaryProvider, _ := config.ParseModelString(primaryModel)
-	for _, gatewayModel := range modelresolution.GatewayEquivalentChatModels(cfg, primaryModel) {
-		addAttempt(gatewayModel, nil)
-	}
-	for _, providerName := range cfg.GetConfiguredProviders() {
-		if providerName == primaryProvider {
-			continue
-		}
-		if _, ok := seenProviders[providerName]; ok {
-			continue
-		}
-		model := chatFallbackModelForProvider(cfg, providerName)
-		if strings.TrimSpace(model) == "" {
-			log.Debug().
-				Str("provider", providerName).
-				Msg("[ChatService] Skipping provider fallback candidate with no stable chat model")
-			continue
-		}
-		addAttempt(model, nil)
-	}
-
-	return attempts, primaryErr
-}
-
-func chatFallbackModelForProvider(cfg *config.AIConfig, provider string) string {
-	if cfg == nil {
-		return ""
-	}
-	if preferred := strings.TrimSpace(cfg.GetPreferredModelForProvider(provider)); preferred != "" &&
-		modelresolution.IsModelUsableForChatWithConfig(cfg, preferred) {
-		return preferred
-	}
-	if fallback := strings.TrimSpace(config.DefaultModelForProvider(provider)); fallback != "" &&
-		modelresolution.IsModelUsableForChatWithConfig(cfg, fallback) {
-		return fallback
-	}
-	return ""
-}
-
-func chatStreamEventBlocksProviderFallback(event StreamEvent) bool {
-	switch strings.ToLower(strings.TrimSpace(event.Type)) {
-	case "approval_needed", "content", "question", "tool_end", "tool_progress", "tool_start":
-		return true
-	default:
-		return false
-	}
-}
-
-func emitChatProviderFallback(callback StreamCallback, failedModel, nextModel string) {
-	if callback == nil {
-		return
-	}
-	failedProvider := ""
-	if strings.TrimSpace(failedModel) != "" {
-		failedProvider, _ = config.ParseModelString(strings.TrimSpace(failedModel))
-	}
-	nextProvider := ""
-	if strings.TrimSpace(nextModel) != "" {
-		nextProvider, _ = config.ParseModelString(strings.TrimSpace(nextModel))
-	}
-	if strings.TrimSpace(failedProvider) == "" {
-		failedProvider = "selected provider"
-	}
-	if strings.TrimSpace(nextProvider) == "" {
-		nextProvider = "another configured provider"
-	}
-	data, _ := json.Marshal(WorkflowStateData{
-		Phase:          "provider_fallback",
-		Message:        fmt.Sprintf("%s did not start a response; trying %s.", providerLabel(failedProvider), providerLabel(nextProvider)),
-		State:          "provider_fallback",
-		FailedProvider: failedProvider,
-		FailedModel:    strings.TrimSpace(failedModel),
-		NextProvider:   nextProvider,
-		NextModel:      strings.TrimSpace(nextModel),
-	})
-	callback(StreamEvent{Type: "workflow_state", Data: data})
+	return chatProviderAttempt{Model: primaryModel, Provider: provider}, nil
 }
 
 func emitChatProviderError(callback StreamCallback, err error) {

internal/ai/chat/service_execute_additional_test.go

@@ -809,35 +809,19 @@ func TestService_ExecuteStream_SuppressesSessionEventWhenCallerAlreadySentIt(t *
 	}
 }
 
-func TestService_ExecuteStream_FallsBackWhenPrimaryProviderFailsBeforeVisibleOutput(t *testing.T) {
+func TestService_ExecuteStream_DoesNotFallbackWhenSelectedProviderFailsBeforeVisibleOutput(t *testing.T) {
 	tmpDir := t.TempDir()
 	store, err := NewSessionStore(tmpDir)
 	if err != nil {
 		t.Fatalf("failed to create session store: %v", err)
 	}
 
-	var primaryStartupMode providers.ProviderStartupMode
-	var fallbackStartupMode providers.ProviderStartupMode
+	var fallbackCalled atomic.Bool
 	primary := &stubServiceProvider{
 		streamFn: func(ctx context.Context, req providers.ChatRequest, callback providers.StreamCallback) error {
-			primaryStartupMode = req.ProviderStartupMode
 			return errors.New("API error (401): unauthorized")
 		},
 	}
-	fallback := &stubServiceProvider{
-		streamFn: func(ctx context.Context, req providers.ChatRequest, callback providers.StreamCallback) error {
-			fallbackStartupMode = req.ProviderStartupMode
-			callback(providers.StreamEvent{
-				Type: "content",
-				Data: providers.ContentEvent{Text: "PULSE_OK"},
-			})
-			callback(providers.StreamEvent{
-				Type: "done",
-				Data: providers.DoneEvent{InputTokens: 1, OutputTokens: 2},
-			})
-			return nil
-		},
-	}
 
 	service := NewService(Config{
 		AIConfig: &config.AIConfig{
@@ -851,9 +835,7 @@ func TestService_ExecuteStream_FallsBackWhenPrimaryProviderFailsBeforeVisibleOut
 	service.sessions = store
 	service.provider = primary
 	service.providerFactory = func(model string) (providers.StreamingProvider, error) {
-		if model == "gemini:gemini-test" {
-			return fallback, nil
-		}
+		fallbackCalled.Store(true)
 		return nil, errors.New("unexpected model " + model)
 	}
 	service.started = true
@@ -863,7 +845,7 @@ func TestService_ExecuteStream_FallsBackWhenPrimaryProviderFailsBeforeVisibleOut
 	var fallbackEvents int
 	var doneModel string
 	err = service.ExecuteStream(context.Background(), ExecuteRequest{
-		SessionID: "fallback-before-visible-output",
+		SessionID: "no-fallback-before-visible-output",
 		Prompt:    "reply",
 	}, func(event StreamEvent) {
 		switch event.Type {
@@ -891,29 +873,26 @@ func TestService_ExecuteStream_FallsBackWhenPrimaryProviderFailsBeforeVisibleOut
 			doneModel = data.Model
 		}
 	})
-	if err != nil {
-		t.Fatalf("ExecuteStream failed: %v", err)
+	if err == nil {
+		t.Fatal("expected selected provider error")
 	}
-	if got := content.String(); got != "PULSE_OK" {
-		t.Fatalf("content = %q, want fallback response", got)
+	if got := content.String(); got != "" {
+		t.Fatalf("content = %q, want no hidden alternate-route response", got)
 	}
-	if errorEvents != 0 {
-		t.Fatalf("expected hidden primary failure without client error event, got %d", errorEvents)
+	if errorEvents != 1 {
+		t.Fatalf("error events = %d, want selected provider failure event", errorEvents)
 	}
-	if fallbackEvents != 1 {
-		t.Fatalf("fallback workflow events = %d, want 1", fallbackEvents)
+	if fallbackEvents != 0 {
+		t.Fatalf("fallback workflow events = %d, want 0", fallbackEvents)
 	}
-	if doneModel != "gemini:gemini-test" {
-		t.Fatalf("done model = %q, want fallback model", doneModel)
+	if doneModel != "" {
+		t.Fatalf("done model = %q, want no done event", doneModel)
 	}
-	if primaryStartupMode != providers.ProviderStartupFastFailBeforeVisibleOutput {
-		t.Fatalf("primary startup mode = %q, want fast-fail before fallback", primaryStartupMode)
-	}
-	if fallbackStartupMode != providers.ProviderStartupDefault {
-		t.Fatalf("fallback startup mode = %q, want provider default", fallbackStartupMode)
+	if fallbackCalled.Load() {
+		t.Fatal("providerFactory should not be called for a hidden alternate route")
 	}
 
-	messages, err := store.GetMessages("fallback-before-visible-output")
+	messages, err := store.GetMessages("no-fallback-before-visible-output")
 	if err != nil {
 		t.Fatalf("GetMessages failed: %v", err)
 	}
@@ -924,7 +903,7 @@ func TestService_ExecuteStream_FallsBackWhenPrimaryProviderFailsBeforeVisibleOut
 	}
 }
 
-func TestService_ExecuteStream_FallsBackToSameModelGatewayRouteBeforeDefaults(t *testing.T) {
+func TestService_ExecuteStream_DoesNotFallbackToSameModelGatewayRoute(t *testing.T) {
 	tmpDir := t.TempDir()
 	store, err := NewSessionStore(tmpDir)
 	if err != nil {
@@ -932,25 +911,13 @@ func TestService_ExecuteStream_FallsBackToSameModelGatewayRouteBeforeDefaults(t
 	}
 
 	var primaryCalls atomic.Int32
+	var providerFactoryCalls atomic.Int32
 	primary := &stubServiceProvider{
 		streamFn: func(ctx context.Context, req providers.ChatRequest, callback providers.StreamCallback) error {
 			primaryCalls.Add(1)
 			return errors.New("dial tcp: i/o timeout")
 		},
 	}
-	fallback := &stubServiceProvider{
-		streamFn: func(ctx context.Context, req providers.ChatRequest, callback providers.StreamCallback) error {
-			callback(providers.StreamEvent{
-				Type: "content",
-				Data: providers.ContentEvent{Text: "GATEWAY_OK"},
-			})
-			callback(providers.StreamEvent{
-				Type: "done",
-				Data: providers.DoneEvent{InputTokens: 1, OutputTokens: 2},
-			})
-			return nil
-		},
-	}
 
 	service := NewService(Config{
 		AIConfig: &config.AIConfig{
@@ -963,18 +930,16 @@ func TestService_ExecuteStream_FallsBackToSameModelGatewayRouteBeforeDefaults(t
 	service.sessions = store
 	service.provider = primary
 	service.providerFactory = func(model string) (providers.StreamingProvider, error) {
-		if model == "openrouter:deepseek/deepseek-v4-pro" {
-			return fallback, nil
-		}
+		providerFactoryCalls.Add(1)
 		return nil, errors.New("unexpected model " + model)
 	}
 	service.started = true
 
 	var content strings.Builder
-	var fallbackNextModel string
+	var fallbackEvents int
 	var doneModel string
 	err = service.ExecuteStream(context.Background(), ExecuteRequest{
-		SessionID: "fallback-to-same-model-gateway",
+		SessionID: "no-fallback-to-same-model-gateway",
 		Prompt:    "reply",
 	}, func(event StreamEvent) {
 		switch event.Type {
@@ -990,7 +955,7 @@ func TestService_ExecuteStream_FallsBackToSameModelGatewayRouteBeforeDefaults(t
 				t.Fatalf("unmarshal workflow state: %v", err)
 			}
 			if data.Phase == "provider_fallback" {
-				fallbackNextModel = data.NextModel
+				fallbackEvents++
 			}
 		case "done":
 			var data DoneData
@@ -1000,106 +965,61 @@ func TestService_ExecuteStream_FallsBackToSameModelGatewayRouteBeforeDefaults(t
 			doneModel = data.Model
 		}
 	})
-	if err != nil {
-		t.Fatalf("ExecuteStream failed: %v", err)
+	if err == nil {
+		t.Fatal("expected selected provider error")
 	}
-	if got := content.String(); got != "GATEWAY_OK" {
-		t.Fatalf("content = %q, want gateway fallback response", got)
+	if got := content.String(); got != "" {
+		t.Fatalf("content = %q, want no gateway fallback response", got)
 	}
-	if fallbackNextModel != "openrouter:deepseek/deepseek-v4-pro" {
-		t.Fatalf("fallback next model = %q, want same-model gateway route", fallbackNextModel)
+	if fallbackEvents != 0 {
+		t.Fatalf("fallback workflow events = %d, want 0", fallbackEvents)
 	}
-	if doneModel != "openrouter:deepseek/deepseek-v4-pro" {
-		t.Fatalf("done model = %q, want same-model gateway route", doneModel)
+	if doneModel != "" {
+		t.Fatalf("done model = %q, want no done event", doneModel)
 	}
-	if got := primaryCalls.Load(); got != 1 {
-		t.Fatalf("primary provider calls = %d, want one fast-fail call before fallback", got)
+	if got := primaryCalls.Load(); got != 2 {
+		t.Fatalf("primary provider calls = %d, want two selected-route retry calls", got)
+	}
+	if got := providerFactoryCalls.Load(); got != 0 {
+		t.Fatalf("provider factory calls = %d, want no automatic gateway provider", got)
 	}
 }
 
-func TestService_ChatProviderAttemptsUsesStableFallbackModelsWithoutLiveCatalog(t *testing.T) {
+func TestService_ChatProviderAttemptUsesSelectedRouteOnly(t *testing.T) {
 	service := &Service{}
-	cfg := &config.AIConfig{
-		ChatModel:        "openrouter:qwen/qwen3.7-plus",
-		DiscoveryModel:   "gemini:gemini-3.1-flash-lite",
-		OpenRouterAPIKey: "sk-or-test",
-		GeminiAPIKey:     "gemini-test",
-		DeepSeekAPIKey:   "deepseek-test",
-	}
 
-	attempts, err := service.chatProviderAttempts(
-		context.Background(),
-		cfg,
+	attempt, err := service.chatProviderAttempt(
 		"openrouter:qwen/qwen3.7-plus",
 		"openrouter:qwen/qwen3.7-plus",
 		&stubServiceProvider{},
 	)
 	if err != nil {
-		t.Fatalf("chatProviderAttempts failed: %v", err)
+		t.Fatalf("chatProviderAttempt failed: %v", err)
 	}
-
-	got := make([]string, 0, len(attempts))
-	for _, attempt := range attempts {
-		got = append(got, attempt.Model)
+	if attempt.Model != "openrouter:qwen/qwen3.7-plus" {
+		t.Fatalf("attempt model = %q, want selected route", attempt.Model)
 	}
-	want := []string{
-		"openrouter:qwen/qwen3.7-plus",
-		"deepseek:deepseek-v4-flash",
-		"gemini:gemini-3.1-flash-lite",
-	}
-	if !reflect.DeepEqual(got, want) {
-		t.Fatalf("attempt models = %#v, want %#v", got, want)
-	}
-	if attempts[0].Provider == nil {
-		t.Fatal("primary attempt should reuse the configured provider")
-	}
-	for i, attempt := range attempts[1:] {
-		if attempt.Provider != nil {
-			t.Fatalf("fallback attempt %d should be lazy provider creation", i+1)
-		}
+	if attempt.Provider == nil {
+		t.Fatal("selected configured route should reuse the configured provider")
 	}
 }
 
-func TestService_ChatProviderAttemptsPrefersSameModelGatewayFallbackBeforeProviderDefaults(t *testing.T) {
+func TestService_ChatProviderAttemptDoesNotAddGatewayEquivalentRoute(t *testing.T) {
 	service := &Service{}
-	cfg := &config.AIConfig{
-		ChatModel:        "deepseek:deepseek-v4-pro",
-		DiscoveryModel:   "gemini:gemini-3.1-flash-lite",
-		OpenRouterAPIKey: "sk-or-test",
-		DeepSeekAPIKey:   "deepseek-test",
-		GeminiAPIKey:     "gemini-test",
-	}
 
-	attempts, err := service.chatProviderAttempts(
-		context.Background(),
-		cfg,
+	attempt, err := service.chatProviderAttempt(
 		"deepseek:deepseek-v4-pro",
 		"deepseek:deepseek-v4-pro",
 		&stubServiceProvider{},
 	)
 	if err != nil {
-		t.Fatalf("chatProviderAttempts failed: %v", err)
+		t.Fatalf("chatProviderAttempt failed: %v", err)
 	}
-
-	got := make([]string, 0, len(attempts))
-	for _, attempt := range attempts {
-		got = append(got, attempt.Model)
+	if attempt.Model != "deepseek:deepseek-v4-pro" {
+		t.Fatalf("attempt model = %q, want selected route only", attempt.Model)
 	}
-	want := []string{
-		"deepseek:deepseek-v4-pro",
-		"openrouter:deepseek/deepseek-v4-pro",
-		"gemini:gemini-3.1-flash-lite",
-	}
-	if !reflect.DeepEqual(got, want) {
-		t.Fatalf("attempt models = %#v, want %#v", got, want)
-	}
-	if attempts[0].Provider == nil {
-		t.Fatal("primary attempt should reuse the configured provider")
-	}
-	for i, attempt := range attempts[1:] {
-		if attempt.Provider != nil {
-			t.Fatalf("fallback attempt %d should be lazy provider creation", i+1)
-		}
+	if attempt.Provider == nil {
+		t.Fatal("selected configured route should reuse the configured provider")
 	}
 }
 

internal/ai/chat/types.go

@@ -272,19 +272,15 @@ type ThinkingData struct {
 
 // WorkflowStateData is the data for "workflow_state" events.
 type WorkflowStateData struct {
-	Phase          string `json:"phase"`                     // investigate | clarify | plan | approve | execute | verify | complete
-	Message        string `json:"message"`                   // Human-readable status text for the UI
-	State          string `json:"state,omitempty"`           // Backend workflow state, when available
-	Tool           string `json:"tool,omitempty"`            // Tool associated with this transition
-	Provider       string `json:"provider,omitempty"`        // Provider selected for this workflow step
-	Model          string `json:"model,omitempty"`           // Model route selected for this workflow step
-	Attempt        int    `json:"attempt,omitempty"`         // One-based attempt number for transient provider retries
-	MaxAttempts    int    `json:"max_attempts,omitempty"`    // Maximum attempts available for the current provider request
-	RetryAfterMS   int64  `json:"retry_after_ms,omitempty"`  // Milliseconds until the next provider attempt starts
-	FailedProvider string `json:"failed_provider,omitempty"` // Provider that failed before visible output
-	FailedModel    string `json:"failed_model,omitempty"`    // Model that failed before visible output
-	NextProvider   string `json:"next_provider,omitempty"`   // Provider selected for the fallback attempt
-	NextModel      string `json:"next_model,omitempty"`      // Model selected for the fallback attempt
+	Phase        string `json:"phase"`                    // investigate | clarify | plan | approve | execute | verify | complete
+	Message      string `json:"message"`                  // Human-readable status text for the UI
+	State        string `json:"state,omitempty"`          // Backend workflow state, when available
+	Tool         string `json:"tool,omitempty"`           // Tool associated with this transition
+	Provider     string `json:"provider,omitempty"`       // Provider selected for this workflow step
+	Model        string `json:"model,omitempty"`          // Model route selected for this workflow step
+	Attempt      int    `json:"attempt,omitempty"`        // One-based attempt number for transient provider retries
+	MaxAttempts  int    `json:"max_attempts,omitempty"`   // Maximum attempts available for the current provider request
+	RetryAfterMS int64  `json:"retry_after_ms,omitempty"` // Milliseconds until the next provider attempt starts
 }
 
 // SessionData is the data for "session" events emitted once the backend has

internal/ai/providers/gemini.go

@@ -426,10 +426,6 @@ func (c *GeminiClient) Chat(ctx context.Context, req ChatRequest) (*ChatResponse
 	var lastErr error
 	maxRetries := geminiMaxRetries
 
-	if req.FastFailProviderStartup() {
-		maxRetries = 0
-	}
-
 	for attempt := 0; attempt <= maxRetries; attempt++ {
 		if attempt > 0 {
 			// Exponential backoff: 2s, 4s, 8s
@@ -833,10 +829,6 @@ func (c *GeminiClient) ChatStream(ctx context.Context, req ChatRequest, callback
 	var lastErr error
 	maxRetries := geminiMaxRetries
 
-	if req.FastFailProviderStartup() {
-		maxRetries = 0
-	}
-
 	for attempt := 0; attempt <= maxRetries; attempt++ {
 		// Bail out early if the parent context is already cancelled
 		if ctx.Err() != nil {

internal/ai/providers/openai.go

@@ -440,9 +440,6 @@ func (c *OpenAIClient) Chat(ctx context.Context, req ChatRequest) (*ChatResponse
 	var respBody []byte
 	var lastErr error
 	maxRetries := openaiMaxRetries
-	if req.FastFailProviderStartup() {
-		maxRetries = 0
-	}
 
 	for attempt := 0; attempt <= maxRetries; attempt++ {
 		if attempt > 0 {
@@ -912,9 +909,6 @@ func (c *OpenAIClient) ChatStream(ctx context.Context, req ChatRequest, callback
 		streamClient = c.client
 	}
 	maxRetries := openaiStreamMaxRetries
-	if req.FastFailProviderStartup() {
-		maxRetries = 0
-	}
 
 	for attempt := 0; attempt <= maxRetries; attempt++ {
 		if attempt > 0 {

internal/ai/providers/openai_test.go

@@ -181,26 +181,6 @@ func TestOpenAIClient_ChatStream_RetriesTransientStartupError(t *testing.T) {
 	assert.Equal(t, "ok", content)
 }
 
-func TestOpenAIClient_ChatStream_FastFailSkipsTransientStartupRetry(t *testing.T) {
-	client := NewOpenAIClient("sk-test", "gpt-4", "https://example.invalid/v1", 0)
-	attempts := 0
-	client.streamClient = &http.Client{
-		Transport: roundTripFunc(func(r *http.Request) (*http.Response, error) {
-			attempts++
-			return nil, io.ErrUnexpectedEOF
-		}),
-	}
-
-	err := client.ChatStream(context.Background(), ChatRequest{
-		Messages:            []Message{{Role: "user", Content: "Hi"}},
-		ProviderStartupMode: ProviderStartupFastFailBeforeVisibleOutput,
-	}, func(StreamEvent) {})
-
-	require.Error(t, err)
-	assert.Equal(t, 1, attempts)
-	assert.Contains(t, err.Error(), "request failed after 0 stream retries")
-}
-
 func TestNewOpenAIClient_BoundsStreamResponseHeaderTimeout(t *testing.T) {
 	client := NewOpenAIClient("sk-test", "gpt-4", "https://api.openai.com/v1", 0)
 	transport, ok := client.streamClient.Transport.(*http.Transport)

internal/ai/providers/provider.go

@@ -89,28 +89,16 @@ type ToolChoice struct {
 	Type ToolChoiceType `json:"type"`
 }
 
-type ProviderStartupMode string
-
-const (
-	// ProviderStartupDefault keeps provider-owned startup retries enabled.
-	ProviderStartupDefault ProviderStartupMode = ""
-	// ProviderStartupFastFailBeforeVisibleOutput is used when the caller has a
-	// separate fallback route ready and wants startup failures to return quickly
-	// before any user-visible provider output is emitted.
-	ProviderStartupFastFailBeforeVisibleOutput ProviderStartupMode = "fast_fail_before_visible_output"
-)
-
 // ChatRequest represents a request to the AI provider
 type ChatRequest struct {
-	Messages            []Message           `json:"messages"`
-	Model               string              `json:"model"`
-	ExecutionID         string              `json:"execution_id,omitempty"` // Stable higher-level run ID shared across related provider turns
-	MaxTokens           int                 `json:"max_tokens,omitempty"`
-	Temperature         float64             `json:"temperature,omitempty"`
-	System              string              `json:"system,omitempty"`      // System prompt (Anthropic style)
-	Tools               []Tool              `json:"tools,omitempty"`       // Available tools
-	ToolChoice          *ToolChoice         `json:"tool_choice,omitempty"` // nil = model-owned automatic selection; none = text-only safety brake
-	ProviderStartupMode ProviderStartupMode `json:"-"`
+	Messages    []Message   `json:"messages"`
+	Model       string      `json:"model"`
+	ExecutionID string      `json:"execution_id,omitempty"` // Stable higher-level run ID shared across related provider turns
+	MaxTokens   int         `json:"max_tokens,omitempty"`
+	Temperature float64     `json:"temperature,omitempty"`
+	System      string      `json:"system,omitempty"`      // System prompt (Anthropic style)
+	Tools       []Tool      `json:"tools,omitempty"`       // Available tools
+	ToolChoice  *ToolChoice `json:"tool_choice,omitempty"` // nil = model-owned automatic selection; none = text-only safety brake
 }
 
 func (r ChatRequest) NormalizeCollections() ChatRequest {
@@ -129,10 +117,6 @@ func (r ChatRequest) NormalizeCollections() ChatRequest {
 	return r
 }
 
-func (r ChatRequest) FastFailProviderStartup() bool {
-	return r.ProviderStartupMode == ProviderStartupFastFailBeforeVisibleOutput
-}
-
 // ChatResponse represents a response from the AI provider
 type ChatResponse struct {
 	Content          string     `json:"content"`

internal/api/contract_test.go

@@ -10605,17 +10605,15 @@ func TestContract_ChatStreamEventJSONSnapshots(t *testing.T) {
 			want: `{"type":"content","data":{"text":"hello"}}`,
 		},
 		{
-			name: "workflow_state",
+			name: "workflow_state_provider_start",
 			event: mustStreamEvent(t, "workflow_state", chat.WorkflowStateData{
-				Phase:          "provider_fallback",
-				Message:        "OpenRouter did not start a response; trying DeepSeek.",
-				State:          "provider_fallback",
-				FailedProvider: "openrouter",
-				FailedModel:    "openrouter:qwen/qwen3.7-plus",
-				NextProvider:   "deepseek",
-				NextModel:      "deepseek:deepseek-v4-pro",
+				Phase:    "provider_start",
+				Message:  "OpenRouter is starting the response.",
+				State:    "investigating",
+				Provider: "openrouter",
+				Model:    "openrouter:qwen/qwen3.7-plus",
 			}),
-			want: `{"type":"workflow_state","data":{"phase":"provider_fallback","message":"OpenRouter did not start a response; trying DeepSeek.","state":"provider_fallback","failed_provider":"openrouter","failed_model":"openrouter:qwen/qwen3.7-plus","next_provider":"deepseek","next_model":"deepseek:deepseek-v4-pro"}}`,
+			want: `{"type":"workflow_state","data":{"phase":"provider_start","message":"OpenRouter is starting the response.","state":"investigating","provider":"openrouter","model":"openrouter:qwen/qwen3.7-plus"}}`,
 		},
 		{
 			name: "workflow_state_provider_retry",

scripts/release_control/subsystem_lookup_test.py

@@ -2860,7 +2860,7 @@ class SubsystemLookupTest(unittest.TestCase):
                 {
                     "heading": "## Shared Boundaries",
                     "path": "internal/api/access_control_handlers.go",
-                    "line": 417,
+                    "line": 427,
                     "heading_line": 117,
                 }
             ],