Learn

What is Graphify?

The plain-language answer, how it works under the hood, what it deliberately is not, and where it falls short.

Graphify is a free, open-source (MIT) command-line tool and skill that turns a codebase into a knowledge graph for AI coding assistants. It parses code on your machine with tree-sitter, writes the graph to local files, and lets assistants like Claude Code, Cursor, and Codex query structure instead of grepping.

How it works

Three steps. Install the CLI from PyPI (the package is graphifyy, the command is graphify), register the /graphify skill with your assistant, then run /graphify . inside the assistant to map the repo. Parsing is on-device tree-sitter across 36 code languages, and three files land in graphify-out/.

terminal

$ uv tool install graphifyy✓ installed graphify $ graphify install✓ /graphify skill registered with Claude Code # then, inside your assistant:/graphify . ✓ graphify-out/graph.html        interactive graph✓ graphify-out/GRAPH_REPORT.md   architecture report✓ graphify-out/graph.json        machine-readable graph

From then on, your assistant (or you, from the terminal) queries the graph instead of grepping. Answers come back as explicit paths with file:line citations:

terminal

$ graphify query "what connects auth to the database?"AuthService → SessionStore → DatabasePool        [EXTRACTED]  src/auth/service.py:42 → src/db/pool.py:17

The same graph is also exposed over MCP (Model Context Protocol), so assistants can call graph tools natively instead of shelling out. The quickstart walks through all of it in about five minutes, and concepts explains the graph model itself.

What Graphify is not

Not RAG or vector search
There are no embeddings and no similarity retrieval. Graphify extracts explicit relationships (calls, imports, references) into nodes and edges, so answers are exact paths with file:line citations, not nearest-neighbor chunks.
Not a cloud service
There is no hosted backend, no account, and no telemetry. Everything runs on your machine and the output is plain files in your repo. The only network calls are the ones your own assistant already makes.
Not a code generator or agent
Graphify does not write code or take actions. It is a map: your existing assistant queries it (over the CLI or MCP) to answer architecture questions with grounded citations before it edits anything.

For a deeper comparison of the graph approach against embeddings-based retrieval, see Graphify vs RAG.

Honest limits

Inferred edges are guesses, and say so
Static extraction cannot see everything (dynamic dispatch, reflection, config-driven wiring). Graphify keeps those edges but tags every relation EXTRACTED, INFERRED, or AMBIGUOUS so you always know what is grounded in code.
The graph goes stale as code changes
graph.json is a snapshot. Re-run /graphify . --update to re-scan only what changed, use graphify watch to rebuild on save, or graphify hook install to rebuild after each commit.
Deep semantic extraction needs a model
The structural graph is pure AST work and needs no API key. Richer semantic passes (community naming, deep mode) run through your assistant session or a model backend you configure for headless CI runs.

Where to go next

Ready to try it? The tutorial takes you from install to your first query. The docs cover the CLI and MCP server in depth, and integrations lists every supported assistant with its exact setup command.

Common questions

What is Graphify in one sentence?

Is Graphify free?

Yes. The whole tool is open source under the MIT license. Install it from PyPI (the package is graphifyy, the command is graphify) with no account, no API key, and no card.

Does my code leave my machine?

No. Parsing runs on-device with tree-sitter and the graph is written to local files in graphify-out/. Graphify has no cloud backend and sends no telemetry. Deeper semantic extraction runs through the AI assistant session you already use, under the terms you already accepted for it.

Which AI coding assistants does Graphify work with?

20+ assistants, including Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot, Aider, OpenCode, Amp, Devin, Kiro, and more. Run graphify install for Claude Code, or a per-assistant command like graphify cursor install.

Is Graphify a RAG or vector-search tool?

No. Graphify does not embed your code or retrieve similar-looking chunks. It extracts explicit nodes and edges (calls, imports, references) into a graph, and answers come back as paths with file:line citations, each edge tagged EXTRACTED, INFERRED, or AMBIGUOUS.

Where is the graph stored?

In a graphify-out/ directory inside your repo: graph.json (the machine-readable graph), GRAPH_REPORT.md (an architecture report), and graph.html (an interactive visualization that opens in any browser). It is designed to be committed to git so a whole team shares one map.

Start in one command