+1 looking for an answer here as well. My guess is that the context limit for all the source code for some large projects will almost always exceed the LLM allowed token limits, so the AI editors generally need to implement some RAG like solution for these. Sourcegraph's cody[0] claims to be able to answer questions across your entire codebase. I haven't tested it extensively but I'm guessing it's competitive with some other AI editors that would offer this.

If I just have a github file open in browser for reference, then I'll sometimes use Rocky AI[1] to explain the code to me - helps avoid the constant copy past into chat gpt from your tab.

[0] https://sourcegraph.com/cody [1] https://rockyai.me/