Hey HN,
We've all seen tools like Cursor pull in context from an entire codebase to help LLMs understand large projects. I wanted an open-source way to get that same deep, structural understanding.
That's why I built Code-to-Knowledge-Graph.
It uses VS Code's Language Server Protocol (LSP) to parse your whole project and builds a detailed knowledge graph – capturing all your functions, classes, variables, and how they call, inherit, or reference each other. This graph is the "codebase-level context" to improve coding agents at scale.
The idea was inspired by research showing that knowledge graphs significantly improve retrieval-augmented generation and structural reasoning (such as "Knowledge Graph-Augmented Language Models" (Zhang et al., 2022 and "GraphCodeBERT")
Would love to hear your thoughts, feedback, or ideas for improvement!