Choosing the server's model

LLM_URL takes any OpenAI-compatible API; it is OpenRouter unless you change it. Leave LLM_MODEL empty to use OpenRouter's current free models: the server reads OpenRouter's model list once a day, keeps the ones that can answer in JSON, and asks them in the order the last evaluation found best (spec/models.json).

LLM_MODEL is a comma-separated list used exactly as given, in order: when one is busy or finds nothing, the next one gets the word. An add answers within 25 seconds and asks at most 3 models; a word looked up before is answered from a 30-day cache. Each lookup is logged with the model that answered, never with the words.

For Ollama on the same machine:

LLM_URL=http://localhost:11434/v1
LLM_MODEL=<a model from `ollama list`>
LLM_API_KEY=

GET /api/v1/llm/status shows the models in use and, with OpenRouter, how many free lookups are left today. See the API reference.