Kodama IA: how to turn any project into a map your AI understands
What a code knowledge graph is, why it saves tokens (and time) for any AI assistant working on your project, and how Kodama IA works in practice.
If you work with an AI assistant (Claude Code, Cursor, GitHub Copilot, or another) on a project of any real size, you've probably already lived this: every new session, the AI has to re-read code and documentation to understand what exists and how it's connected — and that costs tokens, time, and sometimes incomplete answers because the AI never got to see the whole project. Kodama IA was built to solve exactly that problem.
What does Kodama IA do?
Kodama IA analyzes any project folder — code, documents, whatever it is — and generates a knowledge graph: what files exist, how they relate to each other, and a plain-language report of the overall architecture. That map can be browsed as an interactive HTML page (self-contained, no server needed), or handed directly to an AI assistant in JSON format so it can use it as persistent project memory across sessions.
Why this saves your AI tokens
The map extraction uses local syntax analysis (AST) — not a language model — so generating the map doesn't cost a single token, no matter how many times you run it. In an in-house evaluation on two real projects (one in Java, one in Python), the report Kodama generated cut the tokens the AI needed by up to 90.6% compared to reading the full code and documentation, and up to 87.9% compared to reading just the code. The result varies by project and task — not everything gets solved by the report alone — but the savings are consistent.
Who is it useful for?
- Development teams — new hires understand where everything is without someone having to explain it out loud.
- Freelancers and consultants — they get a client project and understand the full structure in minutes, not days.
- Businesses with scattered files — anyone in the company can see at a glance what's in each folder, without opening an IDE.
- AI assistant users in general — the map serves as long-term project memory for Claude Code or a similar tool.
How it works in practice
It installs fully offline (once downloaded, it needs no internet connection or calls to any external service — just Python 3.10 to 3.13 installed). It works the same on Windows and Linux. The Interactive Explorer it generates is a single .html file that you can share by email or USB and open with a double click. And everything it generates can be exported as PNG and PDF for presentations, or as JSON and CSV to paste directly into your AI's context.
Frequently asked questions
Do I need internet to use Kodama IA?
No, once installed. The offline installation includes every dependency; you just need Python 3.10–3.13 already installed on your machine.
Does it spend AI tokens when generating the map?
No. Extraction uses local syntax analysis (AST), not a language model — mapping your project is free, as many times as you want.
How much does it cost and what's included?
A one-time payment of $60, no subscription: every command, an Interactive Explorer with unlimited projects, fully offline installation, full export (PNG, PDF, JSON, CSV), and email support.
Does it work with Claude Code or other AI assistants?
Yes — the generated map (graph.json, GRAPH_REPORT.md) is specifically designed for an AI assistant to use as persistent project memory across sessions.