What locally hosted (on device, or on local network) LLMs does it support? I can only see mention of using cloud providers.

It’s purely API based communication, but we route it through our own API and your communication with it gets sent to OpenAI or Mistral. No unique identifiers for sent messages, we do have some custom content filtering in place for safety.

Is the privacy because it’s running locally, or you’re anonymizing user data when using open ai, or both?

The second one