I’m curious if anyone else has issues with sending chat message to Cursor and it only reads some of the codebase… I got to talking to Cursor back and forth and it revealed to me that, “I can only see what’s explicitly shared in the conversation under “## Potentially Relevant Code Snippets from the current Codebase”. I don’t actually have access to your entire codebase - I can only see what you specifically show me in each message.”
Kind of heartbreaking to be honest. I’m currently working on ways to make the algorithm deem all files in the codebase ‘relevant’… Not sure why all files in the codebase aren’t automatically deemed ‘relevant’, or at least a more expensive plan to have all files deemed relevant and shared with AI.
Any ideas? Any input? Really just wanted to vent my frustration.
This is my pain. I have a huge project, 4000+ files. Just one API request goes through more than 100 files, and if one in the chain is not available for the context, you lose the thread of communication. Previously, you could add 1200 files to the context, now it either asks for the code or starts inventing non-existent ones. You have to ask it 100 times and it’s really annoying, sometimes it’s faster to do everything yourself. This is about the chat. Composer on my project is talking complete nonsense. My code too complicated for them for now.
It is a real shame… There is so much potential nerfed by the scope of what files it can see. I’ve been messing with it and came up with a few potential ways to improve this but it almost seems like too much effort for what its worth.