Bi talks to providers through a common model interface, so most services connect directly.
Add a provider
In Settings, fill in:
- Base URL — the endpoint, for example
https://api.openai.com/v1
- API key — your key for that provider
- Model name — for example
gpt-4o
- API type — usually OpenAI-compatible; Ollama can also use its native API
Each model can declare its capabilities (vision, tools, reasoning) and context length, which the UI uses for display and filtering.
Where keys are stored
Keys are stored in the local .bi directory. Bi has no server of its own, so a key is only sent to the endpoint you configured.
Local models
Point the Base URL at a model service running on your machine and content need not leave it. Common choices:
Default model and failover
You can pin one model or let Bi choose automatically. Automatic selection picks among available models; if one is temporarily unavailable (connection failure or timeout), another available model continues the task instead of aborting it.
Long sessions
How context is compacted and continued is covered in Long conversations.