fix(cowork-service): audit H.7 — fix LLM timeout, harden test mocks, add Ollama config

Bug fixes from systematic review of H.7 LLM router wiring:
- lib/llm-router.ts: remove RUST_RUNTIME_TIMEOUT_MS (300s IPC timeout) override
  — let LlmRouter use its built-in 30s default, appropriate for cloud API calls
- server.test.ts: add debug/error to appMock.log — prevents fragile failures
  if any startup path hits those log levels
- server.test.ts: add OLLAMA_URL + OLLAMA_MODELS to config mock

New feature:
- config.ts: add OLLAMA_URL + OLLAMA_MODELS env vars for local Ollama support
- server.ts: wire Ollama env vars into initLlmRouter() — set
  OLLAMA_MODELS=model1,model2 to auto-add local Ollama as a provider

57 tests passing, 9 test files, typecheck clean

__LOCAL_LLMs/AI_IDE_CHAT_HISTORY/WINDSURF/.last-refresh.log

@@ -1,9 +1,9 @@
-Last refresh: 2026-04-03T06:00:06Z (2026-04-02 23:00:06 PDT)
+Last refresh: 2026-04-03T06:28:08Z (2026-04-02 23:28:08 PDT)
-Cascade conversations: 50 (297M)
+Cascade conversations: 50 (298M)
 Memories: 121
 Implicit context: 20
-Code tracker dirs: 106
+Code tracker dirs: 103
-File edit history: 4222 entries
+File edit history: 4225 entries
 Workspace storage: 37 workspaces
 Repo docs: 7 files across 2 repos
 Repo workflows: 54 files across 12 repos

services/cowork-service/src/lib/config.ts

@@ -29,6 +29,10 @@ const envSchema = baseBackendConfigSchema.extend({
   // Anthropic (passed through to Rust runtime)
   ANTHROPIC_API_KEY: z.string().optional(),
 
+  // LLM Router — local Ollama support (H.7)
+  OLLAMA_URL: z.string().default('http://localhost:11434/v1'),
+  OLLAMA_MODELS: z.string().optional(),
+
   // Ecosystem toggles
   TELEMETRY_ENABLED: z.coerce.boolean().default(false),
   FEATURE_FLAGS_ENABLED: z.coerce.boolean().default(false),

services/cowork-service/src/lib/llm-router.ts

@@ -10,7 +10,6 @@
  */
 
 import { LlmRouter, createLocalOllamaProvider, type ProviderConfig, type TelemetryEntry } from '@bytelyst/llm-router';
-import { config } from './config.js';
 
 let _router: LlmRouter | null = null;
 
@@ -47,9 +46,10 @@ export function initLlmRouter(opts?: LlmRouterOptions): LlmRouter {
   // If no explicit providers, use defaults (cloud free-tier providers)
   const routerConfig = providers.length > 0 ? { providers } : undefined;
 
+  // Use LlmRouter's built-in 30s timeout — appropriate for cloud LLM API calls.
+  // (config.RUST_RUNTIME_TIMEOUT_MS is 300s and only for IPC, not HTTP.)
   _router = new LlmRouter({
     ...routerConfig,
-    timeoutMs: config.RUST_RUNTIME_TIMEOUT_MS,
     onTelemetry: opts?.onTelemetry,
   });
 

services/cowork-service/src/server.test.ts

@@ -8,7 +8,7 @@ const appMock = {
   register: vi.fn(async () => undefined),
   inject: vi.fn(),
   get: vi.fn(),
-  log: { info: vi.fn(), warn: vi.fn() },
+  log: { debug: vi.fn(), info: vi.fn(), warn: vi.fn(), error: vi.fn() },
   addHook: vi.fn(),
 };
 
@@ -32,6 +32,8 @@ vi.mock('./lib/config.js', () => ({
     RUST_RUNTIME_TIMEOUT_MS: 300_000,
     JWT_SECRET: 'test-secret',
     ANTHROPIC_API_KEY: undefined,
+    OLLAMA_URL: 'http://localhost:11434/v1',
+    OLLAMA_MODELS: undefined,
   },
 }));
 vi.mock('./lib/product-config.js', () => ({

services/cowork-service/src/server.ts

@@ -73,8 +73,12 @@ try {
 }
 
 // Initialize LLM router (best-effort — works without API keys in dev)
+// Set OLLAMA_MODELS=model1,model2 to add local Ollama as a provider.
+const ollamaModels = config.OLLAMA_MODELS?.split(',').map(s => s.trim()).filter(Boolean);
 try {
   const llm = initLlmRouter({
+    ollamaModels,
+    ollamaBaseUrl: config.OLLAMA_URL,
     onTelemetry: (entry) => app.log.debug({ llmTelemetry: entry }, 'llm-router event'),
   });
   app.log.info({ providers: llm.getProviders() }, 'LLM router initialized');