I’ve used Cursor and Windsurf. Windsurf does have Cascade memories which are supposed to help with this, although tbh I haven’t seen it make much difference.

IMO they’re all struggling with this problem right now. There’s not much I’ve found to help besides being surgical with the context I provide to the prompt.

You could try Cline. But mono repos suck with LLMs right now because context windows are too small. This might be fixed once Google releases Gemini 2.5 Pro with 2 million tokens limit. But even then, large code bases will still hit a limit at some point.

Until we get billion token context windows, I find the microservices approach works well, where each repo contains a specific microservice. That way the LLM can know everything about the code.

No there isn't. The problem is that all these products and subscriptions operate at loss (even openai pro). What you want is economically not feasible, but technically possible. You can always decide to use API metering (and get max context window) with these tools but then you would see how it gets very expensive very fast although the results may be better.

Claude Code + Vim (+ occasional JetBrains IDE usage) is working phenomenally for me

Windsurf feels more organic. Context size will likely still be a problem.

Zed works fine for me

Try this: cursoraitips.com

Vscode