When you add a custom base_url AND a custom API key (not OpenAI), any operation related to embeddings doesn’t work (Web crawling, @Docs, codebase search). However, creating the embeddings themselves works (create new docs). If you have a custom base_url and your API key is OpenAI compatible, the above features work.
I suspect it’s because the API key is being used by Cursor servers and embeddings are stored and processed locally on Cursor servers.
The fix would be to do forward embeddings to the baseUrl rather than process the data on Cursor’s servers, but there are obvious compatibility problems here.
Would love to have a discussion about this if the developers are open to it, I’ve spent a substantial amount of effort into creating a proxy for Cursor using the base url feature.
-
If you can, please include a screenshot of your problem
-
Please include the name of your operating system
Windows 10 or Ubuntu (WSL). Tested on multiple computers with Windows / Linux.