feat(proxy): enhance VS Code model handling with gateway integration and fallback logic

src/cli/commands/proxy/__tests__/connect-orchestrator.test.ts

@@ -302,7 +302,7 @@ describe('connectTargets — per-target dispatch, summary, partial-failure seman
     const { writeDesktopConfig } = await import('../connectors/desktop.js');
     const { writeVsCodeLanguageModelsConfig } = await import('../connectors/vscode.js');
     vi.mocked(writeDesktopConfig).mockResolvedValue('/desktop/config.json');
-    vi.mocked(writeVsCodeLanguageModelsConfig).mockResolvedValue({ configPath: '/vscode/models.json', requiresSecretConfiguration: false });
+    vi.mocked(writeVsCodeLanguageModelsConfig).mockResolvedValue({ configPath: '/vscode/models.json', requiresSecretConfiguration: false, modelIds: ['model-a'] });
     const { connectTargets } = await import('../connect-orchestrator.js');
 
     await connectTargets({ targets: { claudeDesktop: true, vscode: true } });
@@ -319,7 +319,7 @@ describe('connectTargets — per-target dispatch, summary, partial-failure seman
     const { writeDesktopConfig } = await import('../connectors/desktop.js');
     const { writeVsCodeLanguageModelsConfig } = await import('../connectors/vscode.js');
     vi.mocked(writeDesktopConfig).mockRejectedValue(new Error('disk full'));
-    vi.mocked(writeVsCodeLanguageModelsConfig).mockResolvedValue({ configPath: '/vscode/models.json', requiresSecretConfiguration: false });
+    vi.mocked(writeVsCodeLanguageModelsConfig).mockResolvedValue({ configPath: '/vscode/models.json', requiresSecretConfiguration: false, modelIds: ['model-a'] });
     const { connectTargets } = await import('../connect-orchestrator.js');
 
     await connectTargets({ targets: { claudeDesktop: true, vscode: true } });
@@ -381,7 +381,7 @@ describe('connectTargets — per-target dispatch, summary, partial-failure seman
   it('single successful target still prints the per-target summary (spec §3.4)', async () => {
     await withFreshDaemon();
     const { writeVsCodeLanguageModelsConfig } = await import('../connectors/vscode.js');
-    vi.mocked(writeVsCodeLanguageModelsConfig).mockResolvedValue({ configPath: '/vscode/models.json', requiresSecretConfiguration: false });
+    vi.mocked(writeVsCodeLanguageModelsConfig).mockResolvedValue({ configPath: '/vscode/models.json', requiresSecretConfiguration: false, modelIds: ['model-a'] });
     const { connectTargets } = await import('../connect-orchestrator.js');
 
     await connectTargets({ targets: { vscode: true } });

src/cli/commands/proxy/__tests__/index.test.ts

@@ -400,6 +400,7 @@ describe('proxy connect vscode', () => {
     vi.mocked(writeVsCodeLanguageModelsConfig).mockResolvedValue({
       configPath: '/mock/chatLanguageModels.json',
       requiresSecretConfiguration: false,
+      modelIds: ['gpt-4.1'],
     });
   });
 
@@ -450,6 +451,7 @@ describe('proxy connect vscode', () => {
     expect(daemonOptions).not.toHaveProperty('model');
     expect(writeVsCodeLanguageModelsConfig).toHaveBeenCalledWith(
       'http://127.0.0.1:4001',
+      'local-key',
       false
     );
   });
@@ -491,6 +493,7 @@ describe('proxy connect vscode', () => {
     expect(spawnDaemon).not.toHaveBeenCalled();
     expect(writeVsCodeLanguageModelsConfig).toHaveBeenCalledWith(
       'http://127.0.0.1:4001',
+      'local-key',
       false
     );
   });

src/cli/commands/proxy/connect-orchestrator.ts

@@ -36,7 +36,6 @@ import {
 import { fetchManagedMcpServers } from './connectors/managed-mcp-remote.js';
 import { writeVsCodeClaudeCodeConfig } from './connectors/vscode-claude-code.js';
 import { writeVsCodeLanguageModelsConfig } from './connectors/vscode.js';
-import { VS_CODE_SUPPORTED_MODELS } from './connectors/vscode-models.js';
 import { checkProxyHealth } from './health-check.js';
 import {
   discoverCodexModels,
@@ -442,15 +441,15 @@ async function runVscodeByok(
   verbose: boolean
 ): Promise<TargetResult> {
   try {
-    const result = await writeVsCodeLanguageModelsConfig(state.url, insiders);
+    const result = await writeVsCodeLanguageModelsConfig(state.url, state.gatewayKey, insiders);
     logger.info(
       '[proxy] VS Code BYOK configuration written',
       ...sanitizeLogArgs({
         configPath: result.configPath,
         gatewayUrl: state.url,
         profile: state.profile,
         project: state.project,
-        modelCount: VS_CODE_SUPPORTED_MODELS.length,
+        modelCount: result.modelIds.length,
         clientType: state.clientType,
         requiresSecretConfiguration: result.requiresSecretConfiguration,
       })
@@ -459,7 +458,7 @@ async function runVscodeByok(
     if (verbose) {
       console.log(`  Config:  ${result.configPath}`);
       console.log(`  Gateway: ${state.url}`);
-      console.log(`  Models:  ${VS_CODE_SUPPORTED_MODELS.length}`);
+      console.log(`  Models:  ${result.modelIds.length}`);
       console.log(`  Project: ${config.codeMieProject || '(not configured)'}`);
     }
 

src/cli/commands/proxy/connectors/__tests__/vscode.test.ts

@@ -74,7 +74,11 @@ describe('writeVsCodeLanguageModelsConfigAtPath', () => {
     const provider = providers[0];
     const models = provider.models as Array<Record<string, unknown>>;
 
-    expect(result).toEqual({ configPath, requiresSecretConfiguration: true });
+    expect(result).toEqual({
+      configPath,
+      requiresSecretConfiguration: true,
+      modelIds: [...EXPECTED_MODEL_IDS],
+    });
     expect(provider).toMatchObject({
       name: 'CodeMie',
       vendor: 'customendpoint',

src/cli/commands/proxy/connectors/gateway-models.ts

@@ -0,0 +1,77 @@
+import { ConfigurationError } from '@/utils/errors.js';
+import { logger } from '@/utils/logger.js';
+import { sanitizeLogArgs } from '@/utils/security.js';
+
+/** SSO-backed catalog endpoint; `include_all` returns every model the tenant may use, not just the defaults. */
+const MODELS_PATH = '/v1/llm_models?include_all=true';
+
+interface ModelsListResponse {
+  data?: Array<{ id?: string }>;
+}
+
+interface CodeMieLlmModel {
+  id?: string;
+  base_name?: string;
+  deployment_name?: string;
+}
+
+function extractModelIds(json: ModelsListResponse | CodeMieLlmModel[]): string[] {
+  const ids = Array.isArray(json)
+    ? json.map((model) => model.id || model.base_name || model.deployment_name)
+    : (json.data ?? []).map((model) => model.id);
+  return [...new Set(ids.filter((id): id is string => typeof id === 'string' && id.length > 0))];
+}
+
+/**
+ * Fetch every model ID the gateway serves for the active profile.
+ *
+ * Unlike the Claude Desktop path this applies no family filter — callers decide
+ * what to expose. Always throws {@link ConfigurationError} on failure so callers
+ * can pick their own fallback instead of silently receiving an empty catalog.
+ */
+export async function fetchGatewayModelIds(
+  proxyUrl: string,
+  gatewayKey: string
+): Promise<string[]> {
+  const endpoint = new URL(MODELS_PATH, proxyUrl).toString();
+
+  let response: Response;
+  try {
+    response = await fetch(endpoint, {
+      headers: { Authorization: `Bearer ${gatewayKey}` },
+    });
+  } catch (error) {
+    throw new ConfigurationError(
+      `Local proxy model discovery could not reach ${endpoint}. ` +
+      `Reason: ${error instanceof Error ? error.message : String(error)}`
+    );
+  }
+
+  if (!response.ok) {
+    throw new ConfigurationError(
+      response.status === 401
+        ? `Local proxy model discovery was rejected with 401 Unauthorized at ${endpoint}. ` +
+          'The local gateway key was not accepted by the proxy or was forwarded upstream incorrectly.'
+        : `Local proxy model discovery failed at ${endpoint}: ${response.status} ${response.statusText}`
+    );
+  }
+
+  const contentType = response.headers.get('content-type') ?? '';
+  if (!contentType.includes('application/json')) {
+    throw new ConfigurationError(
+      `Local proxy model discovery received an unexpected response (${contentType || 'no content-type'}) from ${endpoint}. ` +
+      'Your SSO session may have expired — run `codemie proxy stop && codemie profile login` to re-authenticate.'
+    );
+  }
+
+  const ids = extractModelIds(await response.json() as ModelsListResponse | CodeMieLlmModel[]);
+  logger.info(
+    '[proxy] Gateway model discovery completed',
+    ...sanitizeLogArgs({
+      endpoint,
+      modelCount: ids.length,
+      models: ids,
+    })
+  );
+  return ids;
+}