Graphify Turns Your Codebase Into a Knowledge Graph That AI Agents Can Actually Navigate

Graphify is an open-source tool that converts any codebase into a structured knowledge graph, giving AI coding agents like Claude Code a map to navigate large repositories. Instead of dumping raw files into context windows, it builds a layered representation of your code, docs, and media that agents can query efficiently. It works with practically every AI coding tool on the market.

How the Three-Pass System Works

Graphify processes your codebase in three distinct passes, each handling a different type of content.

Pass 1 covers code structure. A tree-sitter parser reads your code files and comments across 25 supported languages, then extracts relationships deterministically. This pass is completely free and uses no API calls. No large language model is involved. The output is a structural map of how your code connects.

Pass 2 handles video and audio files. It transcribes media into Markdown files and inserts them into the knowledge graph. This pass also runs locally with no API calls required.

Pass 3 targets everything that is not code: documentation, papers, images, and other non-code assets. This is where the LLM comes in. Claude subagents read batches of files and output JSON fragments containing nodes, edges, and relationships. The tool extracts semantic meaning from these documents, functioning like a graph RAG system on this final pass.

Platform Support Is Broad

Graphify is not locked to a single tool. The video shows install commands for over 15 platforms, including Claude Code, Codex, OpenCode, Kilo Code, GitHub Copilot CLI, VS Code Copilot Chat, Aider, OpenClaw, Factory Droid, Trae, Gemini CLI, Hermes, Kimi Code, and Amp. The default graphify install targets Claude Code on Linux/Mac, while other platforms use a --platform flag.

Why This Matters for Large Codebases

AI coding agents struggle with large repositories because they have limited context windows and no built-in understanding of how files relate to each other. Graphify solves this by pre-computing the relationships between code, documentation, and media assets. The agent receives a structured map instead of guessing which files to read. The result, shown in the video as a colorful node graph, is a unified view that combines all three passes into a single navigable structure.

Key Takeaways

• Three passes, two free: Code structure and audio/video processing run locally without API costs. Only the semantic extraction pass (docs, papers, images) uses LLM calls.
• Works with any AI coding agent: Claude Code, Codex, Copilot, Aider, Gemini CLI, and many more are supported out of the box.
• Open source and free: The tool is available at no cost on GitHub.
• Deterministic first pass: The code structure analysis uses tree-sitter parsing, not LLM inference, so results are reproducible and fast.
• Graph RAG for codebases: Pass 3 applies semantic extraction to non-code files, creating a true knowledge graph that agents can reason over.

Resources

• Graphify GitHub Repository
• Claude Code Documentation
• Original TikTok by @chase_ai_