graphify: turn your codebase into a queryable knowledge graph

Graphify is a /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI that builds a knowledge graph from your project’s code, docs, SQL schemas, configs, and PDFs—then lets you query it instead of having your agent read files one at a time. No vector store, no embeddings API, no cloud dependency. It parses code locally with tree-sitter and calls your LLM only for docs and images.

How it works

Run /graphify . on a project folder and it produces three outputs: an interactive graph.html visualization, a GRAPH_REPORT.md summary, and a persistent graph.json you can query later. Edges between code entities are tagged EXTRACTED (statically found by tree-sitter), INFERRED (LLM-assisted from docs/images), or AMBIGUOUS, so you can always tell what was found versus what was guessed.

A SHA256 cache means re-runs only process changed files. The --watch flag monitors code changes in real time, and a git post-commit hook can rebuild the graph automatically on every commit. Output optionally includes an Obsidian vault, which is a nice touch if you already use it for notes.

The honest take

The structural clarity is genuinely useful on large codebases—think monorepos and multi-service backends where understanding how auth connects to the database means reading a dozen files. Having a queryable graph changes that interaction meaningfully. Claude can navigate the graph instead of grepping through files.

The token savings numbers, though, are all over the place. The project’s own README cites a 71.5× reduction—on a single mixed corpus of 52 files that they benchmarked themselves. Third-party hands-on reviews put it more like 5–10× on realistic projects. On a small repo (6 files, their own example), savings round to 1×. You’re paying for structural clarity on small projects, not compression. Don’t install this expecting magic math; install it if your project is actually big enough that re-reading the whole context every session is painful.

One small friction: the PyPI package is temporarily named graphifyy (double-y) while the graphify name on PyPI is being reclaimed. The CLI command is still graphify. Minor, but it will confuse you if you try pip install graphify and get nothing useful.

Install

pip install graphifyy && graphify install
# then inside any project:
graphify .

Requires Python 3.10+ and a supported coding agent (Claude Code, Cursor, Codex, or Gemini CLI). The graphify install step drops the skill file into your agent’s skills directory. If you prefer pipx: pipx install graphifyy.

Credit

Built by the Graphify Labs team. Source, docs, and MIT license at github.com/Graphify-Labs/graphify. Sitting around 124k stars as of October 2026, which is genuinely rare air for a developer tool in this category.