CodeGraph Turns Repos into Queryable Graphs to Predict Breakage

CodeGraph ingests any public GitHub repository, parses every function with tree-sitter, and stores call relationships as a directed graph in Neo4j AuraDB. The goal is letting engineers see the downstream impact of a change before making it — a real concern given that developers reportedly spend 58% of their time reading code rather than writing it. Where relational databases need costly recursive JOINs to trace multi-hop dependencies, a single Cypher query resolves 5-level call chains in milliseconds, even across nearly 500,000 relationships.

The project computes a 'Blast Radius' score per function, weighting direct and indirect callers to flag risky change points, and pairs this with a GraphRAG-style AI agent: a Groq-hosted LLaMA model calls tools that query Neo4j directly, so answers are grounded in actual graph data rather than model guesswork. A Cytoscape.js frontend visualizes the graph with color-coded risk levels, letting developers click any function to see its full dependency chain before touching the code.