I’ve been trying to find a way to test Claude Code with Kimi K2 running on Groq. In my head, this could be a super fast, capable agent that would make trying larger tasks less time consuming to wait for.

Initially, I thought I could use the LiteLLM Anthropic /v1/messages API, but I couldn’t figure out how to get that to work. Looking at the exposed endpoints of the FastAPI server running the litellm proxy server, that path didn’t appear to be available.

I ended up finding this nice Github project fakerybakery/claude-code-kimi-groq that sets up a web server that does the proxying to Groq and now I am starting to experiment.