I see. BoltAI currently uses the OpenAI-compatible server from Ollama. Maybe that's why it's slower than querying the model directly.

I will do more benchmarking and maybe switch to direct connection in the future.