Read code like a map
CodeFlow Explorer
Upload a codebase and read how it works as a flow you can open up — from where information comes in, down to the steps inside one small function. Every box points at real lines of real code, and every explanation is written for someone who has never programmed.
Live demos
Each of these is the app's own one-file export of a real analysed project. Click a box to select it, use + to open it up, and read the code it points at — all in your browser, with nothing installed.
Flask todo app
Sign up, log in, keep a list.
125 steps · open it up → JavaScript · Express · PostgresExpress orders API
Place orders, update their status.
120 steps · open it up → TypeScript · ReactReact dashboard
Fetch sales figures and chart them.
77 steps · open it up → Python · CLIPython CSV tool
Read CSVs, clean them, write a report.
113 steps · open it up →Why it isn't just "ask an AI about my code"
- Grounded. tree-sitter maps the real files, functions and calls first. Claude may only describe things in that map — output naming a file, symbol or line that isn't in the upload is rejected and regenerated.
- Plain English. Five-word titles, 25-word explanations, and arrows labelled with the data that actually moves.
- Cheap on purpose. Static analysis does everything it can for free; prompt caching, batching and a content hash cache do the rest. You see a cost estimate first.
- Your key, your bill. No accounts. The key stays on the server and never reaches the browser or the logs.
- Never runs your code. Uploads are read as text and parsed.
Run the full app
The demos above are static snapshots. To analyse your own code you run the app locally — it needs its Python backend for unzipping, parsing and talking to Claude.
git clone https://github.com/Baharul0111/CodeFlow-Explorer.git cd CodeFlow-Explorer make install make dev # then open http://localhost:3000
No API key handy? LLM_MODE=mock runs the whole pipeline against a
deterministic stand-in, exactly like these demos.