Setup (2026-10-30, UTC+8)
- Node v24.22.2 (official linux-x64 tarball), pnpm 11.9.0; `pnpm install --filter @bits/daemon...` in a fresh clone (cd9e48a).
- Daemon only (apps/daemon), driven via its HTTP API on 127.0.0.1:7077 with curl. No web app, desktop app or voice.
- BITS_HOME = fresh directory; BITS_SECRETS=file (no keychain/Secret Service on this machine).
- Model: the vendor's documented "local:" OpenAI-compatible provider pointed at OpenRouter's OpenAI-compatible endpoint, free models only, our own free key read from a file into the env (never printed or logged).

start.sh (as run; key line omitted):
  export BITS_HOME=.../bh BITS_SECRETS=file BITS_PORT=7077
  export BITS_MODEL="local:${BITSM:-dots-studio/dots-3-note-preview:free}" BITS_FAST_MODEL="local:${BITSM:-dots-studio/dots-3-note-preview:free}"
  export BITS_LOCAL_BASE_URL="https://openrouter.ai/api/v1"
  exec node --disable-warning=ExperimentalWarning src/main.ts

DEVIATION FROM PLAN: the plan named nvidia/nemotron-3-ultra-550b-a55b:free. The first daemon run (logs/daemon-1) failed every turn with
  StreamProviderError: Upstream error from Nvidia: Service temporarily overloaded
so before any claim task ran we switched both BITS_MODEL and BITS_FAST_MODEL to dots-studio/dots-3-note-preview:free (also free). Tasks and pass marks unchanged.

Daemon start lines:
 daemon-1: [bits] awake at http://127.0.0.1:7077 · model local:nvidia/nemotron-3-ultra-550b-a55b:free  (then the overload errors above)
 daemon-2: [bits] awake at http://127.0.0.1:7077 · model local:dots-studio/dots-3-note-preview:free
 daemon-3: same, started after the C1 kill -9