Graphify
ResearchTurns any folder of code, SQL, docs, PDFs, images, or video into a queryable, interactive knowledge graph.
Install
$ pip install graphifyy && graphify install
Update
$ pip install -U graphifyy && graphify install
Re-running graphify install fetches the latest version and re-registers the skill.
When to use it
- You drop into an unfamiliar codebase and want a map of its real structure and relationships before reading files
- You want to query a body of work (code, SQL schemas, docs, papers, screenshots, even video) instead of re-reading raw files every time
- You want a persistent, interactive graph you can revisit weeks later without reprocessing, and that auto-syncs as files change
- You want to export understanding into Obsidian, Wikipedia-style articles, JSON, SVG, GraphML, or Neo4j
Commands & flags
- /graphify .
- Graph the current directory.
- /graphify ./raw --mode deep
- Aggressive, deeper extraction.
- /graphify ./raw --update
- Merge new files into the existing graph.
- /graphify ./raw --watch
- Auto-sync the graph as files change.
- /graphify ./raw --wiki
- Emit Wikipedia-style articles from the graph.
How to use it
Install once (needs Python 3.10+), then drive it with the /graphify slash command pointed at any folder.
pip install graphifyy && graphify install
graphify install registers the skill globally (~/.claude/skills/graphify/) and auto-detects your runtime; add --project only if you want the SKILL.md vendored into a single repo. Then, inside the agent:
/graphify . # graph the current directory
/graphify ./raw --mode deep # aggressive extraction
/graphify ./raw --update # merge into the existing graph
/graphify ./raw --watch # auto-sync as files change
/graphify ./raw --wiki # emit Wikipedia-style articles
You get an interactive HTML graph (clickable nodes, search, community filtering) plus analysis — "god nodes" (highest-degree concepts), surprising connections, and suggested questions.
Why it earns its place
Most agents understand a repo by repeatedly reading raw files, which burns context and misses cross-file structure. Graphify builds the structure once into a knowledge graph and lets you query it — the author claims ~71x fewer tokens per query — while surfacing relationships you didn't know were there. It is multimodal (code in a dozen languages, SQL, shell/R scripts, Markdown, PDFs, and images), persists graphs so you can return without reprocessing, can --watch for live updates and rebuild on commit, and exports to Obsidian, JSON, SVG, GraphML, and Neo4j. MIT-licensed, ~70k stars, and runs across Claude Code, Codex, Cursor, Gemini CLI, and more.
Try it
> /graphify . --wiki