[UnrealHenry]

An effort-only runtime selection stores a null provider and an empty model. Chat follows the active provider, but title generation returned before attempting the same fallback, leaving the session's placeholder title in place.

Allow the existing active-provider lookup to handle null, add an isolated loopback regression test that checks the selected model and returned title, and remove the obsolete comment about null providers in the chat resolver.

Closes #193.

Validation on Windows with Bun 1.3.13:

• The new regression fails on the original implementation and passes with the fix.
• bun test ./src/server/__tests__/title-service.test.ts: 15 passed.
• bun test --randomize --seed=1 ./src/server/__tests__/title-service.test.ts: 15 passed.
• Desktop bun run lint: passed.
• Desktop bun run test -- --run: 125 files passed, 1320 tests passed, 4 skipped.
• Full bun test: 1384 passed, 17 skipped, 35 failed. A separate pristine main worktree at 70b8321 has 1383 passed, 17 skipped, and the exact same 35 failing test names. These failures include shell/PATH, symlink, platform and config-directory scan behavior in this Windows environment. CI pins Bun 1.4.2; that version was not run locally.

This contribution was prepared and tested with Codex assistance.

[GoDiao]

Thanks @UnrealHenry, and welcome! This is exactly the fix #193 suggested: null now falls back to the active provider the same way chat does, the outdated comment is gone, and the new test checks both the title and the model the request used. Thanks also for comparing the Windows test failures against a clean main, that saved us a round trip.

Approving and merging. If you find DreamCoder useful, a ⭐ on the repo would mean a lot to us. Thanks again!