The tool now supports local models. In particular, you can pass an alternative base URL for both embeddings and completions so if you can serve a local model over an OpenAI-compatible API (e.g. using LM Studio) then you can point the tool to that.