Type one command, get a map of everything
Graphify is a skill you invoke from inside an AI coding assistant. You type /graphify . and it maps your entire project, code, docs, PDFs, images, and videos, into a knowledge graph you can query instead of grepping through files one at a time. It works across a long list of agents, including Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Aider, and many more.
What sets it apart from the other code-graph tools climbing GitHub is the breadth of what it ingests and how portable the result is. It does not stop at source code: SQL schemas, R scripts, shell scripts, papers, and media all land in the same graph, so application code, database schema, and infrastructure live together. Under the hood it leans on GraphRAG techniques with Leiden community detection and tree-sitter parsing, which is how it finds structure across such different artifact types.
What you actually get
A run drops three files into a graphify-out/ directory, and the three-file output is the quiet genius of the design:
graph.html, which you open in any browser to click nodes, filter, and search.GRAPH_REPORT.md, a written digest of key concepts, surprising connections, and suggested questions.graph.json, the full graph you can query anytime without re-reading your files.
Because the output is plain files, the graph is portable and re-queryable without rebuilding. For a readable architecture page with Mermaid call-flow diagrams, graphify export callflow-html produces one.
Why one graph for code, schema, and infrastructure
The reason to put SQL schemas and infrastructure in the same graph as application code is that the useful questions cross those layers. Which API endpoint touches this database table? What breaks if this config value changes? A code-only index struggles with those, because the answer lives in the seams between code and the systems around it. Graphify’s heterogeneous ingestion is aimed squarely at that class of cross-layer question, which is also why it pulls in docs and papers: they carry the context for why the code is shaped the way it is.
Install
Graphify needs Python 3.10 or newer, and the recommended runner is uv. On macOS with Homebrew:
brew install [email protected] uvOn Windows:
winget install astral-sh.uvThen install the tool and wire it into your assistant:
uv tool install graphifyy
graphify installNote the trap the README flags explicitly: the PyPI package is graphifyy with a double y, and other graphify* packages on PyPI are not affiliated, though the CLI command is still graphify. pipx and pip are documented alternatives to uv. Once installed, you invoke the skill from your coding assistant rather than running a long CLI session, which is the whole point of the /graphify entry point.
Where it sits among the code-graph tools
Three knowledge-graph tools are trending together, and they are not interchangeable. The cleanest way to place graphify is against its neighbors:
| graphify | CodeGraph | Understand Anything | |
|---|---|---|---|
| Stars | 64,545 | 46,365 | 56,235 |
| Form | a skill you invoke (/graphify) | an MCP server for agents | a Claude Code plugin dashboard |
| Ingests | code, SQL, docs, images, video | code | code |
| Output | portable html, json, report files | runtime index | interactive web dashboard |
Counts are from GitHub as of June 2026. CodeGraph optimizes an agent’s runtime token cost by serving an index over MCP. Understand Anything builds a visual dashboard for a human to learn a codebase. Graphify’s distinct bets are the one-command skill invocation, the heterogeneous ingestion that pulls SQL and infrastructure into the same graph as code, and the portable file output you can keep and re-query. If your project is more than just source, graphify is the one that captures the rest of it.
The gotcha to expect
With 332 open issues as of 2026-06 and frequent releases (v0.8.36 in June 2026), this is a fast-moving pre-1.0 tool. The integration surface is where friction shows: a discussed issue is a PreToolUse failed error under Codex, the kind of agent-integration breakage that comes with supporting twenty-odd assistants at once. If your assistant errors on invocation, check the tracker for your specific agent rather than assuming the graph build itself failed.
Related
For the runtime-index sibling of this idea, see CodeGraph, and for the visual-dashboard one, Understand Anything. Graphify works in the same agents that cc-switch helps you configure. For what else is climbing, see LLM tooling, the daily digest, and the weekly report.
FAQ
How do I use graphify? Invoke it as a skill from your AI coding assistant: type /graphify . and it builds the graph and writes the output files.
What can it ingest? Code, SQL schemas, R and shell scripts, docs, PDFs, papers, images, and videos, all into one graph.
What does it produce? A graph.html to explore, a GRAPH_REPORT.md digest, and a graph.json you can query without re-reading your files.
How is it different from CodeGraph or Understand Anything? It is a one-command skill that ingests non-code artifacts too and emits portable files, rather than a runtime MCP index or a dashboard plugin.