Swapping models
Model selection is data, not code. The server reads models.toml once at startup; nothing else in the system names a model.
Swap the answer model
Edit one line
[models]synthesis = "anthropic/claude-sonnet-4.5"The value is an OpenRouter model slug by default, so hundreds of models are one edit away.
Set the key
export OPENROUTER_API_KEY=sk-or-...A different variable name can be configured with endpoint.api_key_env.
Restart the server
No feature code changes anywhere.
Use any OpenAI-compatible endpoint
Point the endpoint at any server that speaks the OpenAI chat-completions API:
[models]synthesis = "llama3.1"
[endpoint]api_base = "http://localhost:11434/v1"api_key_env = "OLLAMA_API_KEY"