Currently only cloud-based models from OpenAI, Anthropic and Google can be selected. It would be great to have the option to connect a local or self-hosted agent (e.g. via Ollama, LM Studio or a custom OpenAI-compatible endpoint).

We work with sensitive data and cannot risk it being used for training public models. A local model integration would let us keep full data control while still using the same workflow features.

A generic "custom endpoint" configuration (base URL + auth token + model name) would be sufficient. This is already common practice for tools like Claude Code, where you can override ANTHROPIC_BASE_URL to route requests through a self-hosted proxy (e.g. LiteLLM) to a local model. The same flexibility here would let us keep sensitive data on our own infrastructure.