Graphify

Overview

Graphify builds a knowledge graph out of a project's code, docs, PDFs, images, and video/audio, then lets an AI assistant query, traverse, and explain it instead of re-reading raw files on every question. Code is parsed locally with tree-sitter (deterministic AST extraction across ~40 languages, no LLM call, nothing leaves the machine); docs, PDFs, images, and media go through the assistant's own model (or a configured API key) for a semantic pass. It's built as a /graphify skill for Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot and 15+ more platforms.

How it works

1. Extract. graphify install registers the skill; running /graphify . parses the target folder into nodes (concepts, files, symbols) and edges (calls, imports, inherits, references).
2. Confidence-tag every edge. Each connection is marked EXTRACTED (explicit in the source), INFERRED (resolved by graphify), or AMBIGUOUS, so nothing is silently guessed.
3. Cluster into communities. Leiden community detection groups the graph into subsystems with LLM-free or LLM-named labels, surfacing "god nodes" (the most-connected concepts).
4. Query instead of grep. graphify query "<question>", graphify path A B, and graphify explain "<concept>" return a scoped subgraph rather than dumping the whole codebase into context.
5. Stay current automatically. graphify hook install rebuilds the graph on every commit and branch switch (AST only, no API cost); graphify update . re-syncs it after a pull or merge.
6. Optional MCP server. python -m graphify.serve graphify-out/graph.json exposes query_graph, get_node, get_neighbors, shortest_path, list_prs, get_pr_impact, and triage_prs as MCP tools, over stdio or Streamable HTTP for a shared team server.

Examples

$ graphify explain "APIRouter"
Node: APIRouter
  Source:    routing.py L2210
  Community: 2
  Degree:    47

Connections (47):
  --> RequestValidationError [uses] [INFERRED]
  --> Dependant [uses] [INFERRED]
  --> .get() [method] [EXTRACTED]

$ graphify path "FastAPI" "ModelField"
Shortest path (3 hops):
  FastAPI --uses--> DefaultPlaceholder <--references-- get_request_handler() --references--> ModelField

Licensing note

The project relicensed from MIT to Apache-2.0 (per the LICENSE file and NOTICE: "Copyright 2026 Safi Shamsi and the Graphify contributors... licensed under the Apache License, Version 2.0"). Portions contributed before the relicensing remain available under the original MIT terms, retained verbatim in LICENSE-MIT. If redistributing any of the source, preserve the NOTICE file as Apache-2.0 requires; listing, linking, and quoting the README for our directory is fine with attribution.

Installation

You need Python 3.10+ and uv (recommended) or pipx.

1. Install the CLI: uv tool install graphifyy (the PyPI package is graphifyy, double-y, while the graphify name is being reclaimed; the command itself is still graphify).
2. Register the skill with your assistant: graphify install (add --project to install into the current repo instead of your user profile, or use a per-platform command like graphify cursor install).
3. Restart or reopen your AI assistant, then type /graphify . in the project you want mapped (on Windows PowerShell, use graphify . without the leading slash).
4. Verify it worked: confirm graphify-out/graph.html, GRAPH_REPORT.md, and graph.json were created, then ask graphify query "<a question about your project>" and check the answer cites real files/lines.