Graphify Turns Your Repo Into a Map
Graphify is pulling over a thousand stars a day on GitHub trending right now, sitting at 84,600 total, and it just shipped v0.9.14 yesterday. The pitch is one command: type /graphify . inside Claude Code, Cursor, Codex, Gemini CLI or any of 15+ coding assistants, and your entire project — code, docs, PDFs, even images and videos — becomes a queryable knowledge graph you can traverse instead of a pile of files you grep through.
The part that matters technically: the code parsing is local and deterministic. Tree-sitter AST extraction across roughly 40 languages, no LLM in the parsing loop, nothing leaves your machine. It resolves cross-file links — calls, imports, inheritance, mixins — then runs Leiden community detection to group related code into subsystems. Every edge carries a confidence tag, EXTRACTED versus INFERRED, so you always know whether the graph read something directly or guessed. Output is an interactive HTML graph, a markdown report, and a JSON graph your agent can query. There is even a post-commit hook that rebuilds the graph on every change and a git merge driver that union-merges graph.json so teams do not fight over it.
Remember the Is Grep All You Need debate from a couple months back? This is the other side of that argument, shipped as a product. Grep answers where does this string appear. A graph answers what calls this, what breaks if I change it, how do these two subsystems actually connect — the questions that eat most of an agent's context window when it has to figure them out by reading files one by one. Whether structured retrieval beats brute-force search for coding agents is one of the live fights in agent architecture, and a thousand stars a day says a lot of people want the structured side to win.
MIT licensed. Install with uv tool install graphifyy, then graphify install to register the skill. Repo: https://github.com/Graphify-Labs/graphify
← Back to all articles
The part that matters technically: the code parsing is local and deterministic. Tree-sitter AST extraction across roughly 40 languages, no LLM in the parsing loop, nothing leaves your machine. It resolves cross-file links — calls, imports, inheritance, mixins — then runs Leiden community detection to group related code into subsystems. Every edge carries a confidence tag, EXTRACTED versus INFERRED, so you always know whether the graph read something directly or guessed. Output is an interactive HTML graph, a markdown report, and a JSON graph your agent can query. There is even a post-commit hook that rebuilds the graph on every change and a git merge driver that union-merges graph.json so teams do not fight over it.
Remember the Is Grep All You Need debate from a couple months back? This is the other side of that argument, shipped as a product. Grep answers where does this string appear. A graph answers what calls this, what breaks if I change it, how do these two subsystems actually connect — the questions that eat most of an agent's context window when it has to figure them out by reading files one by one. Whether structured retrieval beats brute-force search for coding agents is one of the live fights in agent architecture, and a thousand stars a day says a lot of people want the structured side to win.
MIT licensed. Install with uv tool install graphifyy, then graphify install to register the skill. Repo: https://github.com/Graphify-Labs/graphify
Comments