What it is
Graphify turns a project's code, docs, PDFs, images and video into a knowledge graph. Code is parsed locally with tree-sitter across about 40 languages, with no LLM involved. Docs and media go through an optional second pass using a backend you configure (Claude, Gemini, OpenAI, DeepSeek, Kimi, Ollama, Bedrock or Azure). Each edge is labelled as extracted from the source or inferred.
It writes an interactive graph.html, a GRAPH_REPORT.md and a graph.json, and registers a skill in your agent. In Claude Code and Gemini CLI a hook steers the agent to the graph before it searches or reads files; other agents get instruction files such as AGENTS.md. It installs from PyPI as graphifyy (two y's). The repo moved from Safi Shamsi's account to Graphify-Labs, a Y Combinator S26 company that also sells a hosted Graphify Cloud.
Our review
Graphify gives a coding agent a map of the codebase before it starts reading files, and on big repos that is a real saving in wasted tool calls. It has about 133,500 GitHub stars and released v0.9.94 on 22 October 2026, often shipping several releases a week, which says as much about churn as about momentum.
The code graph is deterministic and costs no LLM credits to build. The README's benchmarks are self-reported: 82% key-fact coverage on six ERPNext questions against 70.8% for a grep-and-read baseline, and 0.497 recall@10 on LOCOMO. A sample of six is thin, and the "80% less token spend" headline belongs to the paid cloud product, not the open-source tool.
Privacy depends on configuration. Code never leaves the machine, but the semantic pass sends docs, PDFs and images to whichever backend you pick, and the list includes Kimi, which sends data to Moonshot AI in China. A September 2026 high-severity advisory showed the other exposure: indexing an untrusted repo could make the Fortran extractor read and exfiltrate host files through #include paths, fixed in 0.9.70. The optional query log, off by default, stores your questions and paths in ~/.cache when enabled.
The practical failure mode is staleness. The git hook rebuilds the graph on commit and branch checkout, but after a git pull the graph is out of date until you run graphify update, and an agent steered by a hook towards a stale graph is worse off than one that simply greps. Lookalike packages on PyPI add a supply-chain trap: only graphifyy, with two y's, is official.
Where it earns its place
A staff engineer on a 4,000-file Python and TypeScript monorepo
Stop Claude Code spending half a session grepping for call sites
uv tool install graphifyy, graphify install, then /graphify . builds the code graph locally; the PreToolUse hook sends the agent to graph queries first. Free and no LLM cost for code; add graphify hook install and run graphify update after each pull.
A security-conscious consultancy auditing client repositories
Map an unfamiliar codebase without sending client material anywhere
Build the code graph only, with no backend key set, or point the semantic pass at Ollama. Upgrade to 0.9.70 or later before indexing untrusted repos because of the Fortran include advisory.
An engineering manager at a 40-developer SaaS company
Give every developer and agent the same picture of the system
Commit graph.json and GRAPH_REPORT.md so the team queries one shared graph, and rebuild it in CI. Graphify Cloud adds cross-repo links and Jira and Sentry connectors, but per-seat prices are not published.
Before you commit
- Setup
- Needs Python 3.10+ and uv or pipx: uv tool install graphifyy, then graphify install for your agent; the README puts a first graph at about 30 seconds on a small repo. Extras add PDF, Office, video, Neo4j and backend support.
- Integrations
- Claude Code, Codex, Gemini CLI, Cursor, OpenCode, Kimi Code, Pi and other agents via skills, hooks or AGENTS.md; an optional MCP stdio server; exports to Neo4j and FalkorDB.
- Data and privacy
- Code is parsed locally; docs and media go to the configured backend only when you run the semantic pass. No telemetry is documented; the query log is off by default. One high-severity advisory (host-file read via the Fortran extractor), fixed in 0.9.70.
- Cost at scale
- The open-source tool is free and the code graph uses no model tokens; the semantic pass costs whatever your chosen backend charges per document. Graphify Cloud Pro and Teams are per seat with prices not published.
- Limits
- About 1,620 open issues and pull requests, releases several times a week, ambiguous file extensions (.h, .inc, .m) that parse silently as the wrong language unless declared, and staleness after pulls.
What's good
- Deterministic, local code parsing in about 40 languages
- Edges labelled extracted or inferred
- Hooks or instruction files for 20+ agents
- Code graph costs no LLM credits
What's not
- Doc and media pass sends content to the backend you choose
- Benchmarks are self-reported, some on six questions
- Graph goes stale after git pull until updated
- Lookalike packages share the name
Best for
Teams with large codebases who want their agent to navigate structure instead of grepping.
Alternatives
Graphify: common questions
- Is Graphify worth it?
- Worth installing on large repos where your agent keeps grepping, provided you treat the benchmark numbers as the vendor's own and choose the docs backend deliberately. We rate it 4.1 out of 5.
- How much does Graphify cost?
- Open-source engine free (Apache-2.0, with an MIT licence file also in the repo). Graphify Cloud has a free hosted tier and per-seat Pro and Teams plans with prices not published; self-hosted Enterprise is in early access.
- What are the downsides of Graphify?
- Doc and media pass sends content to the backend you choose; benchmarks are self-reported, some on six questions; graph goes stale after git pull until updated; lookalike packages share the name.
- Who is Graphify best for?
- Teams with large codebases who want their agent to navigate structure instead of grepping.
- How much does Graphify cost for a team or in production?
- The open-source tool is free and the code graph uses no model tokens; the semantic pass costs whatever your chosen backend charges per document. Graphify Cloud Pro and Teams are per seat with prices not published.
- Is Graphify safe for company data?
- Code is parsed locally; docs and media go to the configured backend only when you run the semantic pass. No telemetry is documented; the query log is off by default. One high-severity advisory (host-file read via the Fortran extractor), fixed in 0.9.70.
- What are the best Graphify alternatives?
- The closest alternatives we have reviewed are claude-mem, claude-obsidian and rtk (Rust Token Killer).
Answers checked Oct 23, 2026.
Make Graphify? Show this rating on your site
