Comment by cjonas

I could see this being quite useful in the background for apps like cursor when they need to perform a full codebase search. I imagine it could be more effective in breaking up larger codebases where embeddings start to fall out. If you could fit the entire document into context, you'd be able to "point the model" in the right direction.

The challenge is maintaining it... But you'd maybe ask the model to do that incrementally on every commit, or just throw it away and regenerate from scratch occasionally.