<aside> 📌 Most people burn through tokens re-reading the same files every single session. There's a fix — and it takes 2 minutes to set up.

</aside>

What This Is (And Why You Should Care)

Andrej Karpathy — the guy who coined "vibe coding" — posted about a problem every Claude Code user has: every time you start a new session, your AI has to re-read your entire project from scratch. That's thousands of tokens wasted before you even ask your first question.

Someone built the fix within 48 hours. It's called Graphify — an open-source Claude Code skill that reads your files once, builds a knowledge graph, and then lets you query your entire project in plain English without re-reading a single file.

Step 1: Install Graphify

Open Claude Code and paste this one command:

pip install graphifyy && graphify install

That's it. Two things happen: the Python package installs, and Graphify registers itself as a Claude Code skill. Takes about 30 seconds.

Step 2: Build Your Knowledge Graph

Inside Claude Code, navigate to your ~/.claude folder and run the Graphify skill:

/graphify ~/.claude

This scans every file in your Claude configuration folder — your CLAUDE.md, skills, memory files, everything — and builds a persistent knowledge graph from it. First run takes a bit depending on how much you have in there. After that, it's incremental.

Once it finishes, your AI can query the entire structure of your setup in plain English without re-reading a single file.

Step 3: Add Context Navigation to Your CLAUDE.md

This is the step most people skip — and it's the one that actually saves you tokens. Add this to your ~/.claude/CLAUDE.md file:

## Context Navigation
When you need to understand the codebase, docs, or any files in this project:
1. ALWAYS query the knowledge graph first: `/graphify query "your question"`
2. Only read raw files if I explicitly say "read the file" or "look at the raw file"
3. Use `graphify-out/wiki/index.md` as your navigation entrypoint for browsing structure