Understand Any Codebase in 10 Minutes

New engineers waste weeks reading files one by one. There is a faster way: turn the codebase into a map, look at the biggest nodes first, then ask it questions.

graphify (80,000+ GitHub stars, MIT) is a skill for A.I. coding agents like Claude Code. Point it at any folder and it builds a knowledge graph — an interactive graph.html you explore in your browser, plus a written report and a queryable graph.json. It runs entirely locally (no server, no Neo4j).

We did this live on python-requests and asked one question: “How does a request travel from Session to the network?” It answered with the actual path — Session.send() → HTTPAdapter.send() → urllib3's HTTPConnectionPool — file and line numbers included.

Install (2 commands)

pip install graphifyy && graphify install

Notes: the PyPI package is temporarily named graphifyy (double y) while the original name is reclaimed — the CLI and skill are still graphify. On a Mac with an “externally managed” Python, use pipx install graphifyy. You need Claude Code and Python 3.10+.

Build the map (1 command)

Inside Claude Code, in the repo you want to understand:

/graphify .

For a big repo, start with just the source folder (that’s what we did: /graphify src/requests — ~20 files, builds in a couple of minutes; code extraction is AST-based via tree-sitter, so it’s fast). You get graphify-out/ with:

• graph.html — the interactive map: click nodes, search, filter by community
• GRAPH_REPORT.md — god nodes, surprising connections, suggested questions
• graph.json — the raw graph for path/query commands

The 10-minute checklist

1. Open graph.html. The big nodes (“god nodes”) are the code that everything touches — in requests: Session, HTTPAdapter, Response. Learn those first.
2. Read the top of GRAPH_REPORT.md. It names the communities (subsystems) and the edges worth knowing. Every edge is tagged EXTRACTED, INFERRED, or AMBIGUOUS — you always know what was found versus guessed.
3. Ask three questions:

/graphify query "how does a request travel from Session to the network?"
/graphify path "Session" "HTTPAdapter"
/graphify explain "Response"

query gives a narrated answer with file references; path gives the deterministic shortest path; explain describes one node and its neighbours.

Honest notes

• It extracts structure — classes, functions, calls — plus LLM-extracted concepts from docs. It doesn’t “understand” code like a human, and some edges are inferred; the tags keep it honest.
• Supported inputs (per the README): code (.py .ts .js .go .rs .java .c .cpp .rb .cs .kt .scala .php), docs (.md .txt .rst), PDFs, and images. Doc/PDF/image passes use Claude, so those aren’t offline.
• The graph is a starting map, not a replacement for reading the code that matters — it just tells you which code matters first.