Werkzeuge
·
4 MIN
Graphify: giving AI coding assistants a map
Graphify turns a codebase into a queryable knowledge graph: local-first, no code leaves the machine, cheaper agent runs.
LOCATION
Germany
SERIES
AI Tooling & Dev Harnesses
AUTHOR
Aashwin Shrivastava
PUBLISHED
Graphify turns an entire codebase into a queryable knowledge graph, locally, so an AI assistant gets structure without the code ever leaving the machine. It has become one of the fastest-growing developer tools of the year: 58K+ GitHub stars and 1.2M PyPI downloads since spring.
01.
What it does
Its origin shows how fast this space moves: it was inspired by an Andrej Karpathy post on 1 April 2026 about using LLMs to build personal knowledge bases, and within 48 hours Safi Shamsi (YC S26) had published Graphify on GitHub.
Run it on a project and it reads every file locally, parsing 33 languages via tree-sitter with zero API calls, so your code never leaves the machine. It outputs three artifacts a whole team can query: an interactive HTML visualisation, a markdown report, and a JSON graph, and it plugs into around 20 AI coding assistants.
The reason it spread is concrete: instead of an agent grepping through thousands of files, and tokens, to understand structure, it queries the graph. One demo measured a 72x reduction in Claude Code tokens on the same question; Jack Roberts' walkthrough frames it simply: what Graphify does is give Claude Code a map.
02.
Why we are paying attention
Two reasons. First, the architecture is right for how we work: local-first, no code exfiltration, deterministic artifacts you can commit and share. That is the sovereignty principle applied to developer tooling, the assistant gets context without the codebase ever leaving the client's environment.
Second, it is a sharp reminder that the bottleneck for AI coding assistants is rarely the model, it is giving the model the right context efficiently. A knowledge graph the agent can query beats brute-force file reading on cost, speed, and accuracy at once. We treat that as a pattern worth building around, not just a tool to install. For background, see this overview.
