Cut Your Claude Token Consumption By 70x

A new open-source tool called Graphify reduces token consumption for AI coding assistants by up to 70x. The tool uses an AST parser to build a knowledge graph of code relationships and dependencies, allowing AI agents to navigate the graph instead of re-reading entire files. Graphify supports over 15 platforms including Claude Code, Cursor, and Copilot CLI, and has already garnered over 53,000 GitHub stars.

A recent open-source tool can help you cut your token consumption by up to 70x The tool is called Graphify and is enjoying a well-deserved hype Graphify directly addresses token consumption in a simple and smart way. You run it once. It reads your entire project using an AST parser tree-sitter, which supports 31 languages and builds a knowledge graph that maps relationships, dependencies, and the most interconnected concepts in your code. From that point forward, your AI agent doesn’t re-read files. It navigates the graph. The benchmark from the author: 70x fewer tokens consumed per session All of that boils down to three output files: Obviously, the first and most important reason is the reduction of token consumption. But this is not the only one. It’s not just for Claude Code. Cursor, Codex, OpenCode, Copilot CLI, Gemini CLI, GitHub Copilot Chat, Aider — the list goes on. If your agent can read files, it can query the graph. The tool supports over 15 platforms out of the box. Every relationship in the graph has a confidence tag. EXTRACTED — pulled directly from the code. INFERRED — logically deduced. AMBIGUOUS — uncertain. This is the kind of honesty most AI tools quietly skip. You always know what was found versus what was guessed. Then you can use Claude itself or other AIs to fix it. It identifies “god nodes.” God nodes are the most connected concepts in your project, the cornerstones, or the pieces that everything else flows through. If you’ve just inherited a large codebase you didn’t write, god nodes are the first place you look, the busiest intersections. It’s the architectural map you never had time to draw. On top of these reasons, the GitHub repo has over 53,000 stars. That’s a positive, strong signal. Once you try Graphify, you change the way you approach a codebase: the graph approach changes how you think about AI-assisted development. Instead of dumping context, you query it. Requirements: Python 3.10+ and uv. That’s it. uv tool install graphifyy graphify install Then open your AI assistant and type /graphify. And that’s it. MIT license. Free. Open source. I hope your $20 plan will take you further