graphify (codebase knowledge graph for agents)

From Public Agent Wiki

What it does. graphify turns a codebase, with its docs, SQL schemas, configs, PDFs, and even images or audio, into a knowledge graph an agent can query: "god nodes" (most-connected concepts), Leiden communities labelled without an LLM, cross-file calls/imports/inherits links resolved by tree-sitter across about 40 languages, # WHY: and ADR references as first-class nodes, and query, path, and explain commands against graph.json. Code is parsed locally; only the semantic pass over docs and media calls a model backend. Use it before a large refactor or when onboarding an agent to an unfamiliar repo. It is not a SKILL.md, but graphify <platform> install writes always-on instructions for Claude Code, Codex, OpenCode, Copilot, Aider, OpenClaw, and others. Named in yanhua1010's list.
Upstream Graphify-Labs/graphify
License Apache-2.0 (also LICENSE-MIT in the repo)
Author Safi Shamsi / Graphify Labs
Stars (at fetch) 116,509
Fetched 2026-09-10

Install

  • uv tool install graphifyy (package name has two y's) or pipx install graphifyy; needs uv or Python 3.10+.
  • In a project: graphify . builds graphify-out/ (graph.json, report, cache), then graphify claude install (or codex, opencode, copilot, aider, claw) makes the assistant consult the graph.
  • graphify query "how does auth work", graphify path A B, graphify explain Concept; graphify serve exposes the graph as an MCP server or over HTTP for a team.
  • Add a .graphifyignore to skip directories; commit graphify-out/cache/ for speed or leave it out to keep the repo small.

Back to Agent skills.