Skip to main content
Back

Graphify Gives AI Coding Assistants a 71x Smarter Memory of Your Codebase

Graphify turns any codebase or document folder into a queryable knowledge graph inside your AI coding assistant, using 71x fewer tokens than raw file reads

Repo, Full Access
  • What it is: Graphify is an open-source skill that compiles any folder of code, docs, PDFs, images, or videos into a queryable knowledge graph inside your AI coding assistant.
  • Token efficiency: On a 52-file mixed corpus, it delivers 71.5x fewer tokens per query compared to raw file reading — savings compound on every subsequent session.
  • How it works: Three-pass pipeline: local tree-sitter AST for code (28 languages, no API calls), local Whisper for audio/video, and parallel LLM subagents for docs and images.
  • Platform support: Works with Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot CLI, Aider, VS Code Copilot Chat, Kiro, Devin CLI, Google Antigravity, and more.
  • Install: uv tool install graphifyy && graphify install, then type /graphify . in your assistant. MIT license, free, no telemetry.
  • Context: Built 48 hours after Andrej Karpathy publicly described the need for LLM-managed knowledge bases; now a YC S26 company with 61k+ GitHub stars.

Graphify is a new open-source tool that gives AI coding assistants a structural memory of your entire codebase. Instead of watching Claude Code or Codex grep through files one by one, you run a single command and your project gets compiled into a knowledge graph that the assistant can query directly. The repo crossed 61,000 GitHub stars in under two months, and the timing of its release is not a coincidence.

The Karpathy Spark

Andrej Karpathy , co-founder of OpenAI and former AI lead at Tesla , recently revealed a significant shift in how he uses LLMs: spending less time generating code and more time building personal knowledge bases, structured interlinked markdown wikis that an LLM compiles and maintains from raw source documents. Karpathy himself acknowledged that his current setup is a "hacky" collection of scripts, and that there is a huge opportunity for someone to build a polished product that makes this easy for everyone.

Graphify went viral 48 hours after Karpathy posted his LLM Knowledge Bases workflow. The creator, Safi Shamsi, shipped a working implementation almost immediately. The project is now a YC S26 company, has 128 releases, and its active development branch is already on v8.

The Problem It Solves

AI coding assistants are becoming stronger, but in real codebases they still frequently run into several problems: they do not know how key modules connect, they read many files but do not form an overall architecture map, and search finds text but not upstream and downstream dependencies.

The framing that clicked for many users: Graphify is a compiler, not a retrieval system. RAG (retrieval-augmented generation) chunks your code, embeds it, and retrieves at query time , every session re-reads the same files, every question triggers the same embedding lookup, and the knowledge stays flat with no accumulated structure, no cross-file understanding. Graphify processes your codebase once, at build time, and produces a structured knowledge graph. Subsequent sessions query that pre-compiled graph , they don't re-read your source files.