Configure the LLM

Choose (or switch) the model that powers the orb's routing and replies. For the full list of providers and advanced options, see the Agent reference.

Pick a provider

1. Open Settings → Agent → LLM.
2. Choose a provider:
   • Local (private, no key) — Built-in (MLX) is the best on-device option on Apple Silicon; Ollama, LM Studio, and vLLM use a local server you run.
   • Hosted — protoLabs, OpenAI, Anthropic, and others. These need an API key.
3. If the provider needs one, paste the API key.
4. Save. Changes take effect on the next conversation.

Confirm it works

Just talk to the orb — if it replies, the model is reachable. If it hears you but never answers, the LLM is the usual culprit; re-check the URL and key, and see Voice isn't working.

Advanced (optional)

These live in orbis.yaml (or via env, no rebuild needed):

• Failover backup — add a fallback: model (a local one is ideal) so the orb keeps talking if your primary errors.
• Two-model routing — router_model (stronger, for the decision turn) + content_model (faster, for narration). OpenAI-compatible gateways only.
• Micro tier — micro_model for the cheap throwaway generation (fillers, acknowledgements).

See Agent reference → Advanced LLM options.