Your own LLM provider

Until you have a provider of your own, an agent runs on the free tier's platform model. That is enough to try things out, but the model is not yours to pick and the spend is capped. Your own provider lifts both: tokens go at your own rate with your own provider.

Add a provider

LLM Providers → "Add Provider". The type is OpenAI, Anthropic, Gemini or OpenAI-compatible; the last one also needs a base URL. Then the API key.

The key is stored encrypted on the server and is never returned to the interface again, neither to you nor to anyone else. You can replace it with "Rotate Key", and the previous one becomes invalid at once.

Right after creation the platform goes to the provider for the list of models. If the list did not load, the provider is created all the same; fix the credentials and press refresh in its row.

Purposes and priorities

Every purpose carries its own model: chat is the agent's main model, and the other purposes (images, vision, audio) are tool models. One model per purpose.

A provider carries priority lists per purpose. A purpose the agent has not set is resolved through the list of the provider its chat runs on, and if that is empty too, through the platform provider. An empty list means "the purpose is switched off": such a call will fail, and the agent will say so in the chat.

Pick a model for an agent

The agent's page → Models. One row per purpose: your own provider and model, or "provider's list", meaning there is no model of its own and the purpose is resolved by priorities.

A model the provider has stopped listing is marked "Not listed". The row stays saved, but calls through it may fail.

How much was spent

Token spend is visible per run and per model. The counter counts tokens: the platform does not store prices at call time, and the bill comes from your provider.