I’m currently working with VScode +Aider, and it works reasonably well. I’m interested in Cursor because of the vector indexing of the codebase, as my current workflow gets very expensive, as I like to ask questions about my code. Currently, it has to feed all code directly to OpenAI API. Now if the code is vector-embedded instead it would bypass most of this. I assume many of the calls can be removed (as it appears you already do as core functionality?).
But I have some questions,
1 Are the codebase embedding updated on the fly so re-embeddings happen seamlessly?
2: A thing I like in Aider is adding files y and z to the chat and then asking it to debug and auto-fix blocks across these files. In a video demonstration of Cursor, it seems like we need to either copy-paste the fixed block or edit a specific block inline. Inline editing is cool, but it requires first understanding where the issues are located in the codebase which can be hard with bigger codebases. (I mean this the power of AI to give me some help to fix weird bugs)
Using the latest version of Cursor you can include multiple files. Also with codebase indexing all the code is vectorized based on my understanding.
Hello Mat,
Yes it appears to be vectorized, but is it kept updated when files are changed?