[Gemma x Gemini CLI] Add an Experimental Gemma Router that uses a LiteRT-LM shim into the Composite Model Classifier Strategy

docs/cli/settings.md

@@ -140,6 +140,7 @@ they appear in the UI.
 | Plan                       | `experimental.plan`                      | Enable planning features (Plan Mode and tools).                                                                                                           | `false` |
 | Model Steering             | `experimental.modelSteering`             | Enable model steering (user hints) to guide the model during tool execution.                                                                              | `false` |
 | Direct Web Fetch           | `experimental.directWebFetch`            | Enable web fetch behavior that bypasses LLM summarization.                                                                                                | `false` |
+| Enable Gemma Model Router  | `experimental.gemmaModelRouter.enabled`  | Enable the Gemma Model Router. Requires a local endpoint serving Gemma via the Gemini API using LiteRT-LM shim.                                           | `false` |
 
 ### Skills
 

docs/reference/configuration.md

@@ -1014,6 +1014,23 @@ their corresponding top-level category object in your `settings.json` file.
   - **Default:** `false`
   - **Requires restart:** Yes
 
+- **`experimental.gemmaModelRouter.enabled`** (boolean):
+  - **Description:** Enable the Gemma Model Router. Requires a local endpoint
+    serving Gemma via the Gemini API using LiteRT-LM shim.
+  - **Default:** `false`
+  - **Requires restart:** Yes
+
+- **`experimental.gemmaModelRouter.classifier.host`** (string):
+  - **Description:** The host of the classifier.
+  - **Default:** `"http://localhost:9379"`
+  - **Requires restart:** Yes
+
+- **`experimental.gemmaModelRouter.classifier.model`** (string):
+  - **Description:** The model to use for the classifier. Only tested on
+    `gemma3-1b-gpu-custom`.
+  - **Default:** `"gemma3-1b-gpu-custom"`
+  - **Requires restart:** Yes
+
 #### `skills`
 
 - **`skills.enabled`** (boolean):

packages/cli/src/config/config.test.ts

@@ -2765,6 +2765,66 @@ describe('loadCliConfig approval mode', () => {
   });
 });
 
+describe('loadCliConfig gemmaModelRouter', () => {
+  beforeEach(() => {
+    vi.resetAllMocks();
+    vi.mocked(os.homedir).mockReturnValue('/mock/home/user');
+    vi.stubEnv('GEMINI_API_KEY', 'test-api-key');
+    vi.spyOn(ExtensionManager.prototype, 'getExtensions').mockReturnValue([]);
+  });
+
+  afterEach(() => {
+    vi.unstubAllEnvs();
+    vi.restoreAllMocks();
+  });
+
+  it('should have gemmaModelRouter disabled by default', async () => {
+    process.argv = ['node', 'script.js'];
+    const argv = await parseArguments(createTestMergedSettings());
+    const settings = createTestMergedSettings();
+    const config = await loadCliConfig(settings, 'test-session', argv);
+    expect(config.getGemmaModelRouterEnabled()).toBe(false);
+  });
+
+  it('should load gemmaModelRouter settings from merged settings', async () => {
+    process.argv = ['node', 'script.js'];
+    const argv = await parseArguments(createTestMergedSettings());
+    const settings = createTestMergedSettings({
+      experimental: {
+        gemmaModelRouter: {
+          enabled: true,
+          classifier: {
+            host: 'http://custom:1234',
+            model: 'custom-gemma',
+          },
+        },
+      },
+    });
+    const config = await loadCliConfig(settings, 'test-session', argv);
+    expect(config.getGemmaModelRouterEnabled()).toBe(true);
+    const gemmaSettings = config.getGemmaModelRouterSettings();
+    expect(gemmaSettings.classifier?.host).toBe('http://custom:1234');
+    expect(gemmaSettings.classifier?.model).toBe('custom-gemma');
+  });
+
+  it('should handle partial gemmaModelRouter settings', async () => {
+    process.argv = ['node', 'script.js'];
+    const argv = await parseArguments(createTestMergedSettings());
+    const settings = createTestMergedSettings({
+      experimental: {
+        gemmaModelRouter: {
+          enabled: true,
+        },
+      },
+    });
+    const config = await loadCliConfig(settings, 'test-session', argv);
+    expect(config.getGemmaModelRouterEnabled()).toBe(true);
+    const gemmaSettings = config.getGemmaModelRouterSettings();
+    expect(gemmaSettings.classifier?.host).toBe('http://localhost:9379');
+    expect(gemmaSettings.classifier?.model).toBe('gemma3-1b-gpu-custom');
+  });
+});
+
 describe('loadCliConfig fileFiltering', () => {
   const originalArgv = process.argv;
 

packages/cli/src/config/config.ts

@@ -856,6 +856,7 @@ export async function loadCliConfig(
       // eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
       format: (argv.outputFormat ?? settings.output?.format) as OutputFormat,
     },
+    gemmaModelRouter: settings.experimental?.gemmaModelRouter,
     fakeResponses: argv.fakeResponses,
     recordResponses: argv.recordResponses,
     retryFetchErrors: settings.general?.retryFetchErrors,

packages/cli/src/config/settingsSchema.test.ts

@@ -444,6 +444,60 @@ describe('SettingsSchema', () => {
       expect(hookItemProperties.description).toBeDefined();
       expect(hookItemProperties.description.type).toBe('string');
     });
+
+    it('should have gemmaModelRouter setting in schema', () => {
+      const gemmaModelRouter =
+        getSettingsSchema().experimental.properties.gemmaModelRouter;
+      expect(gemmaModelRouter).toBeDefined();
+      expect(gemmaModelRouter.type).toBe('object');
+      expect(gemmaModelRouter.category).toBe('Experimental');
+      expect(gemmaModelRouter.default).toEqual({});
+      expect(gemmaModelRouter.requiresRestart).toBe(true);
+      expect(gemmaModelRouter.showInDialog).toBe(true);
+      expect(gemmaModelRouter.description).toBe(
+        'Enable Gemma model router (experimental).',
+      );
+
+      const enabled = gemmaModelRouter.properties.enabled;
+      expect(enabled).toBeDefined();
+      expect(enabled.type).toBe('boolean');
+      expect(enabled.category).toBe('Experimental');
+      expect(enabled.default).toBe(false);
+      expect(enabled.requiresRestart).toBe(true);
+      expect(enabled.showInDialog).toBe(true);
+      expect(enabled.description).toBe(
+        'Enable the Gemma Model Router. Requires a local endpoint serving Gemma via the Gemini API using LiteRT-LM shim.',
+      );
+
+      const classifier = gemmaModelRouter.properties.classifier;
+      expect(classifier).toBeDefined();
+      expect(classifier.type).toBe('object');
+      expect(classifier.category).toBe('Experimental');
+      expect(classifier.default).toEqual({});
+      expect(classifier.requiresRestart).toBe(true);
+      expect(classifier.showInDialog).toBe(false);
+      expect(classifier.description).toBe('Classifier configuration.');
+
+      const host = classifier.properties.host;
+      expect(host).toBeDefined();
+      expect(host.type).toBe('string');
+      expect(host.category).toBe('Experimental');
+      expect(host.default).toBe('http://localhost:9379');
+      expect(host.requiresRestart).toBe(true);
+      expect(host.showInDialog).toBe(false);
+      expect(host.description).toBe('The host of the classifier.');
+
+      const model = classifier.properties.model;
+      expect(model).toBeDefined();
+      expect(model.type).toBe('string');
+      expect(model.category).toBe('Experimental');
+      expect(model.default).toBe('gemma3-1b-gpu-custom');
+      expect(model.requiresRestart).toBe(true);
+      expect(model.showInDialog).toBe(false);
+      expect(model.description).toBe(
+        'The model to use for the classifier. Only tested on `gemma3-1b-gpu-custom`.',
+      );
+    });
   });
 
   it('has JSON schema definitions for every referenced ref', () => {