5.13 Local LLM / Ollama issues
Common symptoms
- Local LLM page reports “not installed” or “service stopped.”
- AI Dock warns the desktop model server is unreachable.
- Weights downloaded but Moss claims the model id is missing.
- Pull fails due to disk or port clashes.
Triage sequence
- Open AI capabilities → Local LLM.
- Runtime installed?
- Service alive?
- Model list contains the configured id verbatim?
- Run the built-in test harness.
- Promote passing models to Moss’s fast slot.
Key facts
| Term | Meaning |
|---|---|
| Desktop Ollama | Runs on your workstation—not the board |
| Model id | Must match ollama list exactly |
| Embedding models | Not valid chat backends for Moss |
| Stop service button | Only stops instances Studio launched—system‑wide daemons stop via OS tools |
Desktop Ollama fits Moss fast mode; avoid pointing OpenClaw at it unless the board can route to that endpoint.