Great idea! I agree that AI models should be able to understand the project architecture instead of wasting tokens trying to trace function calls across files. It's incredibly frustrating to see such smart models wasting time parsing the codebase. Do you plan to open source the actual tool? Starred anyway for the cool concept!

Btw I tried it with my switcheroo-control port repository and it understood this low-level Linux Rust project quite well. I'd love to try running it locally with the llama.cpp backend.