It also ships with an openai-compatible server implementation as well now that you could point your UI at (if you wanted to run leaner w/out ollama).

https://github.com/ggerganov/llama.cpp/blob/master/examples/...