Graphify: a free knowledge graph for any codebase

What it is

Imagine a brilliant new assistant who has to reread every file in your office before answering a single question. That's roughly how AI coding assistants like Claude Code, Codex and Cursor behave on big projects.

Graphify, made by Safi Shamsi, hands them a map instead. It reads your code, documents, PDFs, images and even video, then sketches the connections: this relates to that, which depends on this other thing. The assistant checks the map first and only opens what it actually needs.

Why it caught my eye

Safi built the first version in a 48-hour sprint, after Andrej Karpathy shared an idea he called the "LLM Wiki". It spread fast enough to become a company, Graphify Labs, backed by Y Combinator, with over 126,000 stars on GitHub.

The project says each question uses 71.5 times fewer tokens than giving the AI everything. One independent review measured closer to 7.3 times. I like that the gap is out in the open, and even the smaller number is a real saving.

What it means for you

If someone on your team, or a freelancer you hire, uses AI on a growing project, this is a cheaper, calmer way for the assistant to find its way around. The core engine is free and open source, there's a free hosted plan at app.graphify.com, and a version you can run on your own servers.

Words worth knowing

• Knowledge graph: a web of things and how they connect, like a mind map a computer can read.
• Tokens: the small chunks of text an AI reads. Fewer tokens usually means a smaller bill.
• Open core: the main engine is free, with paid extras on top.

A nice question for your next chat with your developer: does our AI reread everything, every time?