On-device Knowledge Graph engine for Enterprise Software
Hey everyone, I'm Safi, founder and CEO of Graphify Labs.
For fifty years, the hard part of software was writing it. AI flipped that overnight. It writes more code than any team can read, and a lot of it is slop. That leaves two hard problems: no one can hold the whole codebase in their head, and no one can be sure an AI's change didn't quietly break something. Graphify solves both.
By the numbers
- 108K+ GitHub stars: one of the top five open-source projects ever from a YC company, and the first to cross 100K⭐ mid-batch
- 5M+ downloads of the open-source engine
- 6,000+ signups on the platform in its first weeks
- Enterprises already running it in production
The memory. One shared, always-current knowledge graph of your whole codebase, across every language and every repo, plus its docs and tickets, so a change in one service shows its impact across all the others. Instead of grepping through files one at a time, your engineers and their AI agents trace how everything connects, who calls what and what a change touches, and answer questions across the entire system, at a fraction of the tokens. Agents connect over a single MCP endpoint, so the graph works inside Claude Code, Cursor, and every major coding tool. It groups the code into communities, the real subsystems, so teams can plan, onboard, and understand the architecture across the whole SDLC.
The proof. Graphify reviews every pull request, including changes that span multiple repositories, and verifies each one. It writes and runs a test that reproduces the bug, proves the new code behaves exactly like the old, or hands back the exact input that breaks it. Most AI reviewers just have one model comment on another model's code, which is still a guess. Graphify proves its findings, because it runs on a real code graph and formal verification.
It learns. Graphify remembers what your team ships and what it rejects, and feeds that back in, so its reviews get sharper over time and stop flagging what your team has already decided is fine.
Migrating a large codebase with AI is like changing a train's engine while it is moving. Graphify lets you do it without stopping the train, and proves the new version does exactly what the old one did.
Here's the launch film:
It also proposes a fix and keeps it only if it re-verifies through the fix loop. Graphify connects to Jira, Linear, and Sentry so it knows the ticket a change should satisfy and the production bug it should resolve.
The enterprise layer runs on-prem, so your code never leaves your environment.
The founder. Safi Shamsi is a published knowledge graph researcher and the author of The Memory Layer. His research on knowledge-graph retrieval improved retrieval 50% and cut hallucinations 67%. Now he's applying that work to one of the hardest problems in software: giving every engineer and AI agent a shared memory of the codebase, and a way to prove their changes still work.
Soon every line of software will be written or reviewed by an agent. That software will need a memory of itself and proof that it works. We're building both.
The ask: Before September 10 we're taking 10 design partners: a free 14-day build where we deploy the enterprise graph on-prem, with your team
Reach us at founders@graphify.com
Dig in
- Run it yourself: https://app.graphify.com
- Open-source engine: https://github.com/Graphify-Labs/graphify
- Docs: https://graphify.com/docs
- Connect any agent over MCP: https://api.graphify.com/mcp
- Benchmarks and methodology: https://github.com/Graphify-Labs/graphify/blob/v8/BENCHMARKS.md
- Talk to us: founders@graphify.com
