What LLM do you use for this? Self hosted or cloud?

This uses the OpenAI API currently. I tried Gemini for embeddings for a little while but didn't seem materially better.