Add local and OGAD remote video generation to Mobile

.gitignore

@@ -98,3 +98,7 @@ docs/wire-captures/
 
 # Scratch renders derived from scripts/e2e/fixtures (regenerable, not versioned)
 /tmp/
+
+# Generated native video runtime
+.video-build/
+native/video/*.xcframework/

App.tsx

@@ -1,3 +1,5 @@
+import { videoGenerationService } from './src/services/videoGenerationService';
+import { videoProvider } from './src/services/modelDownloadService/providers/videoProvider';
 /**
  * Off Grid - On-Device AI Chat Application
  * Private AI assistant that runs entirely on your device
@@ -121,7 +123,7 @@
   );
 }
 
 function App() {
   useEffect(() => () => {
     stopStartupProbe?.();
     stopStartupProbe = null;
@@ -235,6 +237,8 @@
       // becomes the SINGLE owner only once the Download Manager consumes the service
       // and the old recovery paths are folded into the providers.
       registerCoreDownloadProviders();
+      await videoProvider.reconcile?.().catch(error => logger.error('[Video] Download recovery failed:', error));
+      await videoGenerationService.restore().catch(error => logger.error('[Video] Job recovery failed:', error));
 
       // Re-surface QUEUED downloads that never started before an app kill. A queued item (waiting for
       // one of the 3 concurrency slots) has no native row, so hydrateDownloadStore can't recover it —

__tests__/hardening/batch7-ragProjectScoping.test.ts

@@ -55,6 +55,7 @@ jest.mock('../../src/services/rag/database', () => ({
 // Embedding native boundary: deterministic, direction-preserving unit vectors.
 jest.mock('../../src/services/rag/embedding', () => ({
   embeddingService: {
+    runExclusive: <T,>(operation: () => Promise<T>) => operation(),
     isLoaded: jest.fn(() => true),
     load: jest.fn(() => Promise.resolve()),
     embed: jest.fn((text: string) => {

__tests__/hardening/batch9-kb-roundtrip.test.ts

@@ -146,6 +146,7 @@ function mockDeterministicVector(text: string): number[] {
 }
 jest.mock('../../src/services/rag/embedding', () => ({
   embeddingService: {
+    runExclusive: <T,>(operation: () => Promise<T>) => operation(),
     load: jest.fn(() => Promise.resolve()),
     embed: jest.fn((text: string) => Promise.resolve(mockDeterministicVector(text))),
     embedBatch: jest.fn((texts: string[]) => Promise.resolve(texts.map(mockDeterministicVector))),
@@ -199,7 +200,15 @@ describe('BATCH 9 — KB add → indexed → searchable round-trip (real sqlite
     expect(mockMemDb._rows('rag_documents')[0]).toMatchObject({ id: docId, project_id: PROJECT, name: 'solar.txt', size: 4200, enabled: 1 });
     expect(mockMemDb._rows('rag_chunks').length).toBeGreaterThan(0);
     expect(mockMemDb._rows('rag_embeddings').length).toBe(mockMemDb._rows('rag_chunks').length);
-    expect(stages).toEqual(['extracting', 'chunking', 'indexing', 'embedding', 'done']);
+    // Embedding announces itself, then reports once per chunk.
+    const chunkCount = mockMemDb._rows('rag_chunks').length;
+    expect(stages).toEqual([
+      'extracting',
+      'chunking',
+      'indexing',
+      ...Array.from({ length: chunkCount + 1 }, () => 'embedding'),
+      'done',
+    ]);
 
     // Case 15/16: the real filename + size come back from the list query.
     const docs = await ragService.getDocumentsByProject(PROJECT);

__tests__/integration/happy/imageBackends.happy.test.tsx

@@ -1,7 +1,7 @@
 /**
  * HAPPY-PATH (UI, BEHAVIORAL) — image generation across compute backends via the REAL ChatScreen: the user
  * turns on the image-mode toggle (ON/force) and sends; the generated image's details show the correct
- * backend label: NPU→"QNN (NPU)" · MNN(GPU)→"MNN (GPU)" · Metal(iOS)→"Core ML (ANE)".
+ * backend label: NPU→"QNN (NPU)" · MNN(GPU)→"MNN (GPU)" · Metal(iOS)→"Core ML".
  *
  * Only the native diffusion + LiteRT leaves are faked. The image model is DOWNLOADED (boundary) and ACTIVATED
  * by the real toggle gesture (not setState). Generation details are turned on via the real toggle.
@@ -21,7 +21,7 @@ jest.mock('@react-navigation/native', () => ({
 type Cfg = { label: string; backend: 'mnn' | 'qnn'; platform: 'ios' | 'android'; expected: string };
 const CONFIGS: Cfg[] = [
   { label: 'MNN GPU (Android)', backend: 'mnn', platform: 'android', expected: 'MNN (GPU)' },
-  { label: 'Metal (Core ML, iOS)', backend: 'mnn', platform: 'ios', expected: 'Core ML (ANE)' },
+  { label: 'Metal (Core ML, iOS)', backend: 'mnn', platform: 'ios', expected: 'Core ML' },
 ];
 
 describe('happy — image generation shows the correct backend label (heavy entry point)', () => {

__tests__/integration/image/imageParameterPolicy.test.ts

@@ -8,14 +8,24 @@ describe('shared image parameter policy at the Mobile generation boundary', () =
 
   it('uses current settings for both local and remote callers', () => {
     expect(
-      resolveMobileImageParameters(model, {
+      resolveMobileImageParameters({ id: 'juggernaut-xl', name: 'Juggernaut XL', backend: 'sd' }, {
         imageSteps: 42,
         imageGuidanceScale: 5.5,
         imageWidth: 768,
       }),
     ).toEqual({ steps: 42, guidanceScale: 5.5, size: 768 });
   });
 
+  it('keeps a native image pack within its fixed 512 size', () => {
+    expect(
+      resolveMobileImageParameters({ ...model, backend: 'mnn' }, {
+        imageSteps: 42,
+        imageGuidanceScale: 5.5,
+        imageWidth: 768,
+      }),
+    ).toEqual({ steps: 42, guidanceScale: 5.5, size: 512 });
+  });
+
   it('lets an explicit request win and falls back from damaged persisted values', () => {
     expect(
       resolveMobileImageParameters(

__tests__/integration/models/chatHomeSelectorParity.test.ts

@@ -36,7 +36,7 @@ import { handleModelSelectFn } from '../../../src/screens/ChatScreen/useChatMode
 jest.mock('../../../src/services/llm');
 jest.mock('../../../src/services/localDreamGenerator');
 jest.mock('../../../src/services/hardware');
-jest.mock('../../../src/utils/imageModelIntegrity', () => ({
+jest.mock('../../../src/services/imageModelIntegrity', () => ({
   validateImageModelDir: jest.fn(async () => ({ complete: true, missing: [] })),
   ensureImageExtractionComplete: jest.fn(async () => {}),
 }));

__tests__/integration/models/failedRemoteRefreshClearsSelection.rendered.test.tsx

@@ -17,6 +17,7 @@ function ModelsSheet() {
       labels={{
         text: textModel ?? 'Local text',
         image: imageModel ?? 'Local image',
+        video: 'Local video',
         voice: mediaServers.voice ? 'Cloud voice' : 'Local voice',
         speech: mediaServers.transcription ? 'Cloud speech' : 'Local speech',
       }}

__tests__/integration/models/imageDownloadRecovery.test.ts

@@ -14,7 +14,7 @@ const mockModelManager = {
 
 // Integrity is a boundary for this recovery test (it exercises the resume/move
 // orchestration, not the completeness rule — that has its own dedicated tests).
-jest.mock('../../../src/utils/imageModelIntegrity', () => ({
+jest.mock('../../../src/services/imageModelIntegrity', () => ({
   validateImageModelDir: jest.fn(async () => ({ complete: true, missing: [] })),
   ensureImageExtractionComplete: jest.fn(async () => {}),
 }));

__tests__/integration/onboarding/autoSetupImageChoice.rendered.test.tsx

@@ -61,6 +61,12 @@ const catalog: AutoSetupCatalogBoundaries = {
   ],
 };
 
+/** Open a plan's details; the device's own plan starts open, and its heading toggles. */
+function openPlan(ui: ReturnType<typeof render>, tier: string): void {
+  const heading = ui.getByTestId(`auto-setup-plan-heading-${tier}`);
+  if (!heading.props.accessibilityState?.expanded) fireEvent.press(heading);
+}
+
 test('Auto Setup shows Absolute Reality when the device catalog would choose NAI Anime', async () => {
   const navigation = { push() {}, replace() {} } as never;
   const ui = render(
@@ -74,7 +80,7 @@ test('Auto Setup shows Absolute Reality when the device catalog would choose NAI
     expect(ui.getByTestId('auto-setup-plan-lean')).toBeTruthy(),
   );
   for (const tier of ['lean', 'balanced', 'extreme']) {
-    fireEvent.press(ui.getByTestId(`auto-setup-plan-${tier}`));
+    openPlan(ui, tier);
     expect(ui.getByText('Absolute Reality (NPU 8gen2)')).toBeTruthy();
     expect(ui.queryByText('Absolute Reality (GPU)')).toBeNull();
     expect(ui.queryByText('NAI Anime (NPU 8gen2)')).toBeNull();
@@ -104,7 +110,7 @@ test('Auto Setup offers the GPU model when the device recommends GPU', async ()
   await waitFor(() =>
     expect(ui.getByTestId('auto-setup-plan-balanced')).toBeTruthy(),
   );
-  fireEvent.press(ui.getByTestId('auto-setup-plan-balanced'));
+  openPlan(ui, 'balanced');
   expect(ui.getByText('Absolute Reality (GPU)')).toBeTruthy();
   expect(ui.queryByText('Absolute Reality (NPU 8gen2)')).toBeNull();
 });
@@ -143,7 +149,7 @@ test('Android Auto Setup presents complementary image models across its three pl
     ['balanced', 'Absolute Reality (NPU 8gen2)'],
     ['extreme', 'Dream Shaper V8 (NPU 8gen2)'],
   ]) {
-    fireEvent.press(ui.getByTestId(`auto-setup-plan-${tier}`));
+    openPlan(ui, tier);
     expect(ui.getByText(name)).toBeTruthy();
   }
 });

__tests__/integration/onboarding/autoSetupJourney.test.tsx

@@ -68,6 +68,10 @@ class NativeDownloadBoundary implements DownloadProvider {
     if (request.modelType === 'image') {
       return this.begin(request.model.id, request.model.name, request.model.size);
     }
+    if (request.modelType === 'video') {
+      const size = request.model.files.reduce((sum, file) => sum + (file.sizeBytes ?? 0), 0);
+      return this.begin(request.model.id, request.model.name, size);
+    }
     return this.begin(request.modelId, request.modelId, 1 * MB);
   }
 
@@ -218,16 +222,21 @@ describe('Auto Setup release journey', () => {
       expect(ui.getByTestId('auto-setup-plan-balanced')).toBeTruthy(),
     );
 
-    expect(ui.getAllByText('INCLUDES')).toHaveLength(1);
+    // Only the chosen plan is expanded.
+    expect(ui.getAllByTestId('auto-setup-selected-plan')).toHaveLength(1);
     expect(ui.getByText('Gemma 4 E4B')).toBeTruthy();
-    fireEvent.press(ui.getByTestId('auto-setup-plan-lean'));
+    fireEvent.press(ui.getByTestId('auto-setup-plan-heading-lean'));
     expect(ui.getByText('Qwen 3.5 2B')).toBeTruthy();
     expect(ui.queryByText('SmolVLM 2B')).toBeNull();
-    fireEvent.press(ui.getByTestId('auto-setup-plan-extreme'));
+    fireEvent.press(ui.getByTestId('auto-setup-plan-heading-extreme'));
     expect(ui.getByText('Qwen 3.5 9B')).toBeTruthy();
     expect(ui.queryByText('Gemma 4 E4B')).toBeNull();
-    expect(useAppStore.getState().settings.modelLoadingMode).toBe('aggressive');
+    // A tier only previews downloads; it never changes the live loading policy.
+    expect(useAppStore.getState().settings.modelLoadingMode).toBe('balanced');
 
+    // Text is preselected; image and speech-to-text are opt-in.
+    fireEvent.press(ui.getByTestId('auto-setup-choice-image'));
+    fireEvent.press(ui.getByTestId('auto-setup-choice-stt'));
     fireEvent.press(ui.getByTestId('auto-setup-download'));
     await waitFor(() =>
       expect(ui.getByTestId('auto-setup-continue')).toBeTruthy(),
@@ -285,6 +294,8 @@ describe('Auto Setup release journey', () => {
       expect(ui.getByTestId('auto-setup-download')).toBeTruthy(),
     );
 
+    fireEvent.press(ui.getByTestId('auto-setup-choice-image'));
+    fireEvent.press(ui.getByTestId('auto-setup-choice-stt'));
     fireEvent.press(ui.getByTestId('auto-setup-download'));
 
     await waitFor(() =>
@@ -314,6 +325,8 @@ describe('Auto Setup release journey', () => {
     await waitFor(() =>
       expect(ui.getByTestId('auto-setup-download')).toBeTruthy(),
     );
+    fireEvent.press(ui.getByTestId('auto-setup-choice-image'));
+    fireEvent.press(ui.getByTestId('auto-setup-choice-stt'));
     fireEvent.press(ui.getByTestId('auto-setup-download'));
     await waitFor(() =>
       expect(ui.getAllByText(/STARTING|0%/).length).toBeGreaterThan(0),
@@ -364,7 +377,7 @@ describe('Auto Setup release journey', () => {
     await waitFor(() =>
       expect(ui.getAllByText(/STARTING|0%/).length).toBeGreaterThan(0),
     );
-    fireEvent.press(ui.getByTestId('auto-setup-plan-extreme'));
+    fireEvent.press(ui.getByTestId('auto-setup-plan-heading-extreme'));
 
     expect(ui.getByText('Gemma 4 E4B')).toBeTruthy();
     expect(ui.queryByText('Qwen 3.5 9B')).toBeNull();

__tests__/integration/pro/proExpiryRedirect.integration.test.tsx

@@ -59,8 +59,8 @@ describe('the live Pro expiry redirect', () => {
 
     expect(resetRoot).toHaveBeenCalledTimes(1);
     expect(resetRoot).toHaveBeenCalledWith({
-      index: 0,
-      routes: [{ name: 'ProDetail' }],
+      index: 1,
+      routes: [{ name: 'Main', params: { screen: 'HomeTab' } }, { name: 'ProDetail' }],
     });
   });
 });

__tests__/integration/pro/proPricingSettingsEntry.rendered.test.tsx

@@ -90,9 +90,9 @@ describe('Pro pricing reached from Settings (rendered, real App)', () => {
     expect(view.queryByText(/\$119/)).toBeNull();
     expect(view.queryByText('Intelligence, democratized.')).toBeNull();
 
-    // The Settings upsell banner (its LIVE SYNC feature grid) carries its own "Get Pro" button.
+    // The Settings upsell banner carries its own "Get Pro" button.
     // Home stays mounted behind the tabs with a "Get Pro" of its own, so scope to the banner.
-    let banner = view.getByText('LIVE SYNC').parent;
+    let banner = view.getByText('Keep your work in context with memory, Sync, and actions you approve.').parent;
     while (banner && rtl.within(banner).queryAllByText('Get Pro').length === 0) {
       banner = banner.parent;
     }

__tests__/integration/rag/embeddingFlow.test.ts

@@ -44,6 +44,7 @@ const deterministicEmbed = (text: string): number[] => {
 
 jest.mock('../../../src/services/rag/embedding', () => ({
   embeddingService: {
+    runExclusive: <T,>(operation: () => Promise<T>) => operation(),
     load: jest.fn(() => Promise.resolve()),
     embed: jest.fn((text: string) => Promise.resolve(deterministicEmbed(text))),
     embedBatch: jest.fn((texts: string[]) => Promise.resolve(texts.map(deterministicEmbed))),
@@ -90,7 +91,8 @@ describe('Embedding Flow Integration', () => {
 
       // Verify embedding service was called
       expect(embeddingService.load).toHaveBeenCalled();
-      expect(embeddingService.embedBatch).toHaveBeenCalled();
+      // Each chunk is embedded on its own so indexing can report per-chunk progress.
+      expect(embeddingService.embed).toHaveBeenCalled();
 
       // Verify embeddings were inserted into the database
       const embInserts = mockExecuteSync.mock.calls.filter(

__tests__/integration/remote/ollamaKeyedDiscovery.rendered.test.tsx

@@ -79,7 +79,7 @@ describe('Ollama over HTTPS with a saved key', () => {
 
     // Back on Home, the picker lists the Ollama's model.
     fireEvent.press(ui.getByLabelText('Back'));
-    await waitFor(() => expect(ui.getByText('Select a text or image model to start')).toBeTruthy());
+    await waitFor(() => expect(ui.getByText('Select a text, image, or video model to start')).toBeTruthy());
     fireEvent.press(ui.getByTestId('browse-models-button'));
     await waitFor(() => expect(ui.getByText(MODEL)).toBeTruthy());
   });

__tests__/integration/settings/liteRTReleaseBackends.rendered.test.tsx

@@ -17,21 +17,20 @@ it('offers only CPU and GPU in Chat and Model Settings even with a saved NPU pre
   const { useAppStore } = require('../../../src/stores');
   await useAppStore.persist.rehydrate();
   const { GenerationSettingsModal } = require('../../../src/components/GenerationSettingsModal');
-  const { TextGenerationSection } = require('../../../src/screens/ModelSettingsScreen/TextGenerationSection');
+  const { GeneralGenerationControls } = require('../../../src/components/settings/GeneralGenerationControls');
   const model = createDownloadedModel({ id: 'gemma-litert', engine: 'litert', filePath: '/models/gemma.litertlm' });
   useAppStore.getState().setDownloadedModels([model]);
   useAppStore.getState().setActiveModelId(model.id);
   useAppStore.getState().updateSettings({ liteRTBackend: 'npu' });
   const chat = render(React.createElement(GenerationSettingsModal, { visible: true, onClose: () => {} }));
-  fireEvent.press(chat.getByText('TEXT GENERATION'));
-  fireEvent.press(chat.getByTestId('modal-text-advanced-toggle'));
+  // Acceleration is a General control on both surfaces.
+  fireEvent.press(chat.getByTestId('modal-general-accordion'));
   expect(chat.queryByText('NPU (Beta)')).toBeNull();
   expect(chat.getByText(/Run on GPU via OpenCL/)).toBeTruthy();
   fireEvent.press(chat.getByText('CPU'));
   expect(useAppStore.getState().settings.liteRTBackend).toBe('cpu');
   chat.unmount();
-  const settings = render(React.createElement(TextGenerationSection));
-  fireEvent.press(settings.getByTestId('text-advanced-toggle'));
+  const settings = render(React.createElement(GeneralGenerationControls));
   expect(settings.queryByText('NPU (Beta)')).toBeNull();
   fireEvent.press(settings.getByText('GPU'));
   expect(useAppStore.getState().settings.liteRTBackend).toBe('gpu');

__tests__/integration/settings/modelSettingsSurfaceParity.test.tsx

@@ -40,6 +40,40 @@ describe('model settings surface parity', () => {
     _clearSlotsForTesting();
   });
 
+  it('keeps shared controls in General on both surfaces and preserves saved values', () => {
+    useAppStore.getState().updateSettings({
+      modelLoadingMode: 'aggressive',
+      showGenerationDetails: true,
+    });
+    const chatSettings = render(<GenerationSettingsModal visible onClose={() => {}} />);
+    fireEvent.press(chatSettings.getByText('TEXT GENERATION'));
+    expect(chatSettings.queryByText('Model Loading')).toBeNull();
+    expect(chatSettings.queryByText('Show Generation Details')).toBeNull();
+    fireEvent.press(chatSettings.getByTestId('modal-general-accordion'));
+    expect(chatSettings.getByText('Text Model Backend')).toBeTruthy();
+    expect(chatSettings.getByTestId('model-loading-mode-aggressive-button')).toBeTruthy();
+    expect(chatSettings.getByTestId('show-gen-details-on-button')).toBeTruthy();
+    chatSettings.unmount();
+
+    const modelSettings = renderModelSettings();
+    fireEvent.press(modelSettings.getByTestId('text-generation-accordion'));
+    expect(modelSettings.queryByText('Model Loading')).toBeNull();
+    fireEvent.press(modelSettings.getByTestId('general-generation-accordion'));
+    expect(modelSettings.getByText('Text Model Backend')).toBeTruthy();
+    expect(modelSettings.getByTestId('model-loading-mode-aggressive-button')).toBeTruthy();
+    expect(modelSettings.getByTestId('show-gen-details-on-button')).toBeTruthy();
+  });
+
+  it('shows and edits the saved default system prompt in chat Text settings', () => {
+    useAppStore.getState().updateSettings({ systemPrompt: 'Saved instructions' });
+    const chatSettings = render(<GenerationSettingsModal visible onClose={() => {}} />);
+    fireEvent.press(chatSettings.getByText('TEXT GENERATION'));
+    const input = chatSettings.getByTestId('default-system-prompt-input');
+    expect(input.props.value).toBe('Saved instructions');
+    fireEvent.changeText(input, 'New instructions');
+    expect(useAppStore.getState().settings.systemPrompt).toBe('New instructions');
+  });
+
   it('caps output by context on both surfaces and writes one shared setting state', () => {
     useAppStore.getState().setModelMaxContext(262144);
     // A context wide enough for the output this test chooses. Max tokens is capped BY the context,

__tests__/pro/sync/stateSync.integration.test.tsx

@@ -297,9 +297,12 @@ describe('Pro mobile state sync journey', () => {
         created_at: createdAt,
       },
     );
+    // Twenty real edits: a put that repeats the current value is not a new op, so each revision
+    // carries its own value and the Desktop's final 0.55 is genuinely later than the phone's
+    // own first-run settings.
     for (let revision = 0; revision < 20; revision += 1) {
       remoteLog.record(CORE_SYNC_ENTITIES.modelSetting, 'temperature', 'put', {
-        value_json: revision === 19 ? '0.55' : '0.5',
+        value_json: revision === 19 ? '0.55' : (0.3 + revision * 0.01).toFixed(2),
       });
     }
 
@@ -318,7 +321,7 @@ describe('Pro mobile state sync journey', () => {
     );
     fireEvent.press(ui.getByTestId('settings-tab'));
     fireEvent.press(ui.getByText('Model Settings'));
-    fireEvent.press(ui.getByTestId('text-generation-accordion'));
+    fireEvent.press(ui.getByTestId('general-generation-accordion'));
     fireEvent.press(ui.getByTestId('show-gen-details-on-button'));
     fireEvent.press(ui.getByLabelText('Back'));
     fireEvent.press(await waitFor(() => ui!.getByTestId('open-sync-settings')));