Tools· Research & Knowledge Management

graphify

graphify converts code, documentation, PDFs, images and video into a searchable semantic knowledge graph that reads far more token efficiently than ingesting every file individually.

structuringresearchingexecutableOpen Source

Source code

uv tool install graphifyy, then graphify install to register it with your AI environment. Alternatively pipx install graphifyy or pip install graphifyy also work. Invoke via the CLI as graphify <path> or /graphify <path> in Claude Code.

Open provider

Description

The open source tool works in two phases: a local extraction via Tree-sitter analyzes source code for free and without any content leaving your machine, then a subsequent semantic pass brings together documentation, PDFs and media through an AI model and resolves the relationships between the elements. From any given path this produces three output files: an interactive visualization, a summary report and a queryable graph as JSON.

The main benefit is understanding an unfamiliar or organically grown project without having to read every file individually, and being able to ask targeted questions against the graph instead of searching the whole codebase again each time. At Voyage Digital, graphify is part of the standard approach for exploring third party open source repositories before their actual code is read or adapted.

Strengths

Local code analysis

Source code is parsed purely locally via a tree-sitter AST, deterministic and without any LLM call, so this part never leaves your machine.

Traceable edges

Every connection in the graph is tagged EXTRACTED, INFERRED or AMBIGUOUS, so you can always tell what was read directly from the source and what was derived.

A real graph, not a vector index

Without embeddings or a vector store you can traverse the graph directly, trace a path between two concepts, or get a single node explained.

Broad platform support

The skill can be registered with, among others, Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot and more than 15 other environments.

Benchmark validated

Results run on the same test harness with the same model and budget, and are scored by a judge model that was cross checked against a second judge model with 90.6 percent agreement and a Cohen's kappa of 0.81.

Assessment

AI features

Semantic pass on non code content Documentation, PDFs, images and video material run through the model of the respective AI assistant or a configured API key.
Targeted graph queries Commands such as graphify query, graphify explain and graphify path return a scoped subgraph for a plain language question instead of the full report.
Automatic report The generated GRAPH_REPORT.md names the most connected nodes, surprising links between modules, reasoning documented in the code such as NOTE or WHY comments, and suggested questions to ask the graph.
PR triage graphify prs --triage lets the configured AI backend rank your own review queue by priority.
Adding external sources graphify add fetches a paper, for example from arXiv, or transcribes a YouTube video and adds it to the graph.

Suitable for

Developers who need to get up to speed quickly in an unfamiliar or organically grown codebase
Teams who want to understand or port a third party open source repository before adopting its code
Users of Claude Code, Cursor or similar AI coding environments who want to ask targeted questions against a knowledge graph instead of the full set of files
Projects with mixed material of code, documentation, PDFs and media that should be evaluated as one connected body of content

Limitations and notes

Not fully local Only the code extraction runs locally without an LLM, documentation, PDFs, images and video require the AI assistant's model or your own API key.
Mind the Python environment Installing via pip install can leave graphify looking for the package in a different environment than the one it runs in at runtime, which triggers a ModuleNotFoundError. uv tool install or pipx install isolate the package in its own environment and avoid the issue.
Watch the package name On PyPI the official package is called graphifyy with a double y, other graphify packages are not affiliated. The CLI command itself is still called graphify.
uvx call differs When running via uvx or uv tool run, you must name the package graphifyy rather than the command graphify, otherwise resolution fails.
Google Workspace files .gdoc, .gsheet and .gslides files from Google Drive for desktop are shortcut pointers without document content, evaluating them also requires the gws CLI with authentication.

Quick start

Install the package: uv tool install graphifyy (alternatively pipx install graphifyy).
Register the skill with your AI assistant: graphify install, adding --project for a project scoped instead of a global install.
In your AI assistant, type /graphify . to build the graph for the current directory.
Use the three generated files: open graph.html in a browser, read GRAPH_REPORT.md for the overview, use graph.json for targeted queries.
Optionally run graphify hook install so the graph rebuilds automatically on every git commit.

Tips

Use graphify query "<question>" for targeted questions instead of reading all files again, it costs far fewer tokens than the full report.
Use graphify path "A" "B" to trace the shortest path between two concepts in the project.
Commit graphify-out/ to the git repository so the whole team has the same graph available right after checkout.
Create a .graphifyignore in the project root, syntax like .gitignore including ! negation, to exclude irrelevant files from the graph.
After an update or reinstall, run graphify hook install again so the interpreter path embedded in the git hook stays current.

Access

The repository is public at https://github.com/Graphify-Labs/graphify and is released under the MIT license. Recommended installation via uv: uv tool install graphifyy, followed by graphify install to register it with your AI environment. Alternatively, pipx install graphifyy or pip install graphifyy also work. It is then invoked via the CLI as graphify <path> or as the skill trigger /graphify <path> in Claude Code.