Graphify: the open source tool that turns your project into a knowledge graph, so Claude Code queries the graph instead of reading files one by one
You type /graphify in your AI assistant and the tool maps your code, docs, PDFs, even images and videos into one knowledge graph. After that, when the assistant looks for an answer, it checks the graph instead of opening dozens of files one by one. The code part is extracted entirely on your machine, with no AI calls. It is Apache 2.0 licensed and free, and works with more than 20 assistants including Claude Code, Cursor, Codex and Gemini CLI. Below: a verified install, the commands you will use every day, and two overstatements in the usual pitch: "71x fewer tokens" does not hold for every project, and "nothing leaves your machine" is only true for code.
The tool itself
The repo is Apache 2.0 licensed and completely free. The company behind it also offers a paid platform at graphify.com that updates in the background; but the command line tool covered here does not depend on it and works on its own.
What is extra below: the trap in the package name, the PATH issue that breaks the install on Windows, the separate command that makes the assistant actually use the graph, and which projects the token saving claim really holds for.
What it does
When Claude Code looks for an answer in a large project, it usually greps and opens files one by one. Every opened file goes into context, so it burns tokens and the assistant can still miss connections. Graphify does this work once and saves the result as a graph: which function calls what, which file imports what, which doc refers to which code.
From install to first question
- 1
Install
Install the CLI and register the skill with your assistant
- 2
/graphify .
Code is parsed locally with tree-sitter; docs, PDFs and images go to your assistant's model
- 3
Three files
graph.html (clickable map), GRAPH_REPORT.md (summary), graph.json (the graph itself)
- 4
Query
The assistant asks the graph instead of rereading files
It is not a vector database; there are no embeddings. It is a real graph, so you can trace the path between two things step by step. Every connection carries a tag too: EXTRACTED means it is explicit in the source, INFERRED means the tool resolved it. You can tell what was read from what was guessed.
Install
You need Python 3.10 or newer. The recommended route is uv; on Windows, install uv first if you do not have it.
winget install astral-sh.uvuv tool install graphifyy
graphify installThen open Claude Code and type this in your project folder:
/graphify .The package name has two y's: graphifyy
The official package on PyPI is graphifyy. The command itself is graphify, with one y. Other packages starting with graphify exist on PyPI and are not affiliated with the project; install the wrong one and you run entirely different code.
Common install traps
Installed it, but it says graphify not found
Run uv tool update-shell, then open a new terminal. The tool's folder is not on PATH yet
uvx graphify install
uvx --from graphifyy graphify install. uvx reads the first word as the package name, and the package is graphifyy
Plain pip install graphifyy on Windows
uv tool install or pipx install. With pip, the skill can pick up a different Python and fail with "No module named graphify"
If you want PDFs, Word/Excel or video files in the graph too, you install the extras on purpose; for example uv tool install "graphifyy[pdf]", or "graphifyy[all]" for everything.
To make the assistant actually use the graph
This is the step most people miss. Building the graph does not mean Claude Code will use it on its own; it keeps grepping out of habit. After the graph is built, run this once in the project:
graphify claude installThis installs a hook that fires right before the assistant searches or reads files, and steers it to ask the graph first. For Cursor, Codex and others the command differs (graphify cursor install, for example); the full list is in the repo.
The graph goes stale over time. You can install a hook so it updates itself on every commit; after a git pull you still update it by hand:
graphify hook install # once: rebuilds on every commit and branch switch
graphify update . # after a git pullCommands you will use every day
The assistant calls these itself, but you can ask from the terminal too:
graphify query "what connects auth to the database?"
graphify path "UserService" "DatabasePool"
graphify explain "RateLimiter"| Command | When |
|---|---|
| query | You ask a plain question, it returns the relevant subgraph |
| path | Shows step by step how two things connect |
| explain | Explains one concept: where it is defined and what it connects to |
Take a look at GRAPH_REPORT.md once too. It lists the most connected concepts, unexpected links between different folders, and sample questions the graph can answer. It is the fastest way to understand a project you just joined.
The claims, one by one
What is said, what is true
You spend 71x fewer tokens
71.5x was measured on one 52 file test project. The same table in the repo shows 5.4x for a 4 file project and almost no difference for a 6 file one. Savings grow with project size
Nothing leaves your machine
True for code. Docs, PDFs and images are sent to your assistant's model to be understood. If you only want code processed, there is a --code-only flag
Building the graph is free
The code part really is free, no AI calls. On the first /graphify run, any docs cost tokens; the savings come in later queries
The README is out of date on query logging
The README's privacy section says every query is written to a log file on your machine. In the current code (0.9.77) that log is off by default and only turns on if you set GRAPHIFY_QUERY_LOG_ENABLE=1. The log is never sent anywhere anyway; the tool has no telemetry.
DOA: Yapay Zeka ve Otomasyon
Installing these tools on your own is one thing; actually building with them is another. The community has people using these daily and people building systems from scratch.
If you want a system that actually runs in your business, let's talk for 10 minutes; I'll look at what you're trying to build and tell you which path fits. Free, and not a sales pitch.
Book a 10-minute callIf you'd rather learn this alongside people doing the same work instead of on your own, the community is always open:
Join the communityThis is an affiliate link.