I tried to prompt using Claude Code and Cursor, but I was still not happy with the results for medium-large codebases.

So four months later, I have now built an open-source repository (CodeBoarding) that analyzes your code-flows, and generates an abstract overview of your codebase that you can interact with.

To do this, I extract code-flows using an LSP server to reduce the context and serve it to an LLM in multiple steps (read metadata, explore packages, validate diagrams, etc.).

The final results are now hosted on the website url, and you can find the repo from there. Would be super happy to hear feedback, thanks HN!