Nice, I wanted to build something similar for a long time. The coolest thing is to start summarising clusters for very large codebases, which essentially provides an LoD system for the context.