AI models and API setup · Prompt token budgets

Add an OpenAI-compatible or Anthropic endpoint in settings. Use the base URL supplied by your provider and an API key when required. Test the connection before starting work. A local endpoint still needs a running model server; choosing it does not automatically make every connected tool local. Reserve response tokens before assembling a model request. Test a prompt exactly at the remaining limit and stop when the reserve exceeds the context window.

Uncensored, unfiltered and abliterated describe model choices you can connect through a compatible endpoint. They are not a guarantee that every provider accepts every request. Check the model licence, server configuration and tool support.