Graphify Review: Knowledge Graphs for AI Coding Assistants

Graphify is an open-source knowledge-graph tool for AI coding assistants. Type /graphify in Claude Code, Cursor, Codex, Gemini CLI, and many other hosts, and it maps a project—code, docs, PDFs, images, and video—into a queryable graph so agents can ask, path-trace, and explain instead of grepping the whole tree. Searches for graphify usually want install steps (the PyPI package is graphifyy), what local AST parsing covers, and how Graphify Cloud differs from the free engine.

Graphify-Labs/graphify official GitHub repository page

This review summarizes the public site at graphify.com, docs at docs.graphify.com, and the official repository Graphify-Labs/graphify. Benchmark figures below are project-reported from the maintainers’ harness—not independent third-party audits. Check the live GitHub page for current stars, forks, and activity.

What Graphify is

Graphify’s pitch: cut agent token spend by mapping the repo once, then querying the graph. Code is parsed with tree-sitter AST—deterministic, no LLM, nothing leaves your machine for a code-only run. Docs, PDFs, images, and video can take an optional semantic pass through your assistant’s model or a configured API key. Edges are tagged EXTRACTED (explicit in the source) or INFERRED (resolved by Graphify), so you can tell what was read directly from what was inferred.

It is not a vector index: no embeddings store by default. You get a real graph you traverse with graphify query, graphify path, and graphify explain.

Install (official quickstart)

Needs Python 3.10+ and uv (or pipx). From the README:

uv tool install graphifyy      # PyPI package is graphifyy (double-y)
graphify install               # detects your assistant and registers the skill

Then in the assistant chat (not the shell):

/graphify .

That produces graphify-out/ with graph.html (interactive), GRAPH_REPORT.md, and graph.json. The CLI command is still graphify; other similarly named PyPI packages are not affiliated. Docs: Installation.

Platform registration covers Claude Code, Cursor (graphify cursor install), Codex, Gemini CLI, GitHub Copilot, OpenCode, and 20+ more. Project-scoped installs use graphify install --project.

What you get locally

God nodes — most-connected concepts
Communities — Leiden (or Louvain fallback) subsystems
Cross-file links — calls / imports / inherits across ~40 languages via tree-sitter
Query, path, explain — plain-language questions and shortest paths against graph.json
Beyond code — docs, PDFs, images, and optional video/audio into the same graph

Common follow-ups: /graphify ./docs --update, graphify hook install for commit rebuilds, and always-on assistant guidance so the agent prefers scoped graph queries over raw grep.

Benchmarks (vendor / project-reported)

The README’s table (same harness, same model budgets, blind-validated judges) headlines LOCOMO recall@10 of 0.497 vs mem0 0.048 and supermemory 0.149; LongMemEval-S QA accuracy 76% (tied with dense RAG in their write-up); and graph builds with 0 LLM credits for code-only AST extraction. Full tables and reproduction notes live in the repo’s BENCHMARKS.md. Treat every figure as publisher evaluation under their harness.

Graphify Cloud vs open-source

Graphify Cloud is the always-on hosted layer: live graph across monorepos and cross-repo SDLC, formal verification claims, code-review context, and connectors (Sentry, Jira, and more). The marketing site positions an ~80% token-spend reduction for agents that stop re-reading the codebase every session. Local graphifyy remains free to run yourself; Cloud is a separate product with a free trial at app.graphify.com.

Who it is for

Teams whose agents re-read large repos every session
Developers on Claude Code, Cursor, Codex, or Gemini who want a persistent codebase map
Orgs that need local-first code indexing (no API key for code-only graphs)

Caveats

Package name: install graphifyy on PyPI; the CLI is graphify
Semantic pass: docs/media need a backend or assistant model; code-only stays offline
Benchmarks: project-run; your repo and model will differ
Cloud vs local: hosted always-on features are not the same as the free CLI skill

Sources

Sources verified Oct 26, 2026 (Pacific Time)