Interactive graph
Dots represent parts of your code; lines show recorded connections. Select a function to see what uses it and open its source.
Backed by
Combinator
CombinatorShared understanding of codebase
for agents and developers
Free plan, no card · run the open-source engine yourself
Works with
- GitHub stars
- 141,202
- PyPI downloads
- 9.8M
- Repositories indexed
- 15.2M
How it works
One map of the code
One graph across every repository and team, built once and kept current. Nobody re-learns the codebase by hand, and what one engagement learns carries into the next.
Answers with proof
Every answer is a path through the graph you can open, spanning code, docs, wikis and contracts. Evidence rather than opinion, for audited and regulated work.
See what breaks
Before the merge, the graph follows every edge to what a change breaks. The agent reads the graph instead of the repo, so the token bill buys answers.
Used by engineers at
Features
Understand your code. Know where to look next.
Build with code context, inspect reviews and supported verification, and recover shared repository knowledge.
Fully on-device
Run the open-source engine and parsing stays on your machine: 33 tree-sitter grammars, no uploads, no telemetry.
Query, don’t grep
Give your assistant source-linked code context. In Cloud, add shared Memory with inspectable sources and history.
20+ assistants
Your assistant queries the graph over MCP, in Claude Code, Cursor, Copilot, and 14 more.
Cross-repo graph
One persistent graph across every repo, so answers follow owners across services.
PR review
graphify prs triages CI, review, and merge-conflict risk, learning from each merge.
Formal verification
Compare supported behavior changes and inspect the evidence. Coverage depends on the workflow and language.
Side by side
Graph, chunks or grep.
Four tools an enterprise shortlists, as each vendor documents them in October 2026.
Graphify| Capability | Graphify |  CodeRabbit |  Greptile |  Sourcegraph |
|---|---|---|---|---|
 Differential verification | prove it or break it | no formal verification | review and runtime tests | search and batch changes |
 Cross-repo blast radius | every connected repo | linked repos, per PR | up to 7 linked repos | refs, no verdict |
 Typed code graph | persistent, queryable | per change (Codegraph) | stored, used in reviews | cross-repo (SCIP, opt-in) |
 Multi-hop reasoning | native traversal | agentic, in reviews and chat | agentic hops | agentic retrieval |
 Auditable answer | a path you can open | comments with evidence | findings, confidence scores | answers list sources |
 Provenance | every edge tagged | Blast Radius only (beta) | not documented | search and file sources |
 Deployment | cloud · self-run open source | self-host, 500+ seats | self-host / air-gap | cloud or self-host |
 Any AI assistant | any MCP client | CLI and plugins, MCP client | MCP server + CLI | MCP server (Enterprise) |
 License | open core (Apache 2.0) | proprietary | proprietary | proprietary |
Trust
Run it on your machine, or let us run it.
The open-source tool never sends your code anywhere. The hosted product is a choice, not the only way in.
On-device
The command-line tool parses locally with 33 tree-sitter grammars and writes the graph to a file on your own disk.
No telemetry
The tool sends no usage pings, no crash reports and no analytics. Nothing to opt out of, because nothing is sent.
Apache 2.0
The whole source is Apache 2.0 on GitHub. You need not trust this page; read what the code does.
Yours to host
Serve the graph over stdio or over HTTP on infrastructure you own, and on Enterprise run the whole thing there.
Proof
- 79× fewer tokens
After a weekend of using it on MemMachine, I’m not going back. We’re seeing 79x token reductions, and zero vector database needed.
Steve Scargall, MemVergeSenior Product Manager & Software Architect - Less re-reading
Instead of Claude Code re-reading every file, it queries the graph — which is persistent across sessions and costs a fraction of the tokens.
lucasrosatiSetup guide for Claude Code, Obsidian and Graphify - Rootly integration
The data already lives in Rootly. The graph makes the structure visible.
Sylvain Kalache, RootlyHead of AI Labs · on Rootly’s own Graphify plugin - Fewer blind searches
Instead of Claude blindly searching through files, Graphify helps it understand relationships, dependencies and affected areas first. Less blind searching. Better context. Smarter implementation.
Audran Yematha@WolfhardsA - A second brain for code
It takes the best of the wiki index and graphs, but helps you more with code without leaving any document or md file behind — it eats everything and still works like a 2nd brain.
TheSaid@sineidesis
From the blog
Latest posts
Graphify made it to GitHub's Octoverse 2026
GitHub's Octoverse 2026 lists Graphify-Labs/graphify among the most-engaged open source repositories on GitHub this year. Thank you to everyone who built it.
By Syed Fahad
Graphify accepted into NVIDIA Inception: what it means
Graphify Labs was accepted into NVIDIA Inception, NVIDIA's free program for AI startups. Here's what changes for you and what we're building next.
Reduce Claude Code token usage: what actually works
How to reduce Claude Code token usage: see where tokens go (context, CLAUDE.md, MCP tools, file reads), which fixes cut them, and where a code graph fits.
Why vector search isn't enough for coding agents
Vector search finds code that sounds right. A code knowledge graph returns the dependency path, with every hop labelled parsed or inferred. One real trace.
FAQ
Frequently Asked Questions
Common questions about how Graphify works.
RAG splits your code into chunks, embeds them, and retrieves the top-k that look similar to your question. Graphify builds a real graph and traverses it, so every answer is an explicit path with file:line citations. There are no embeddings and no vector store. You follow the actual call and import edges instead of guessing which chunks are relevant.
That is your choice. The open-source engine parses code locally with tree-sitter, which is deterministic AST extraction with no model call and no telemetry, so nothing leaves your machine unless you point the non-code pass at a hosted model. The hosted product builds and keeps the same graph for you in the cloud, on the repositories you connect.
Yes. The hosted product at app.graphify.com has a Free plan with no card, and Pro and Teams by seat above it (those need a card). The open-source engine is free under Apache 2.0, runs on your own machine, and needs no account. Graph-aware review and formal verification are in every hosted plan, Free included. The early-access Enterprise layer adds self-hosting, SAML single sign-on and legacy migration.
A review on every pull request and a graph that stays current without you maintaining it. The CLI builds the same kind of graph on your machine, but you keep it fresh yourself and it has no automatic reviews. On Pro, Graphify re-indexes your default branch on every push, reviews each pull request against the graph, runs formal verification on changed functions, and serves the graph to your assistants over the hosted MCP server. On Teams, everyone shares that one graph. Start on the Free plan with no card, or try Pro free for two weeks.
17, including Claude Code, Cursor, Copilot, Codex, Gemini CLI and Aider. Run graphify install --platform <name> for each assistant you use; bare graphify install sets up Claude Code. The graph is also served over an MCP server any client can query.
LSP answers one hop: where a symbol is defined, where it is referenced. A graph answers questions that span the codebase, such as what breaks if you change this, every path between two functions, and how a doc or config connects to the code. It is persistent, queryable by your assistant, and every edge carries a provenance tag.
ctags builds a flat symbol index for jump-to-definition. Sourcegraph is enterprise code search, run in its cloud or self-hosted. Graphify builds a typed graph your assistant traverses over CLI or MCP. It is not a symbol table and not a search box.
Yes, incrementally. A code change re-parses only the changed file through the local AST pass, with no re-embedding and no full re-index. Install the git hook and the graph rebuilds on every commit.
Yes. Code parsing is always local. Point the semantic pass at a local model such as Ollama, or any OpenAI-compatible server like vLLM, and nothing leaves the machine: tree-sitter on code, your local model on docs.
It can store embeddings as one signal on a node, but retrieval is a graph traversal rather than a similarity search. Results come back connected and explainable instead of as a ranked list of lookalike chunks.
