A map of any GitHub repo, on your machine

For your first week on a new codebase. Repo Atlas reads a repository once and lays it out in five layers, from whole areas down to single functions. Every arrow is a real call.

npx @cpldxx/repo-atlas

Free to run. Needs Node 18+ and git.

Repo Atlas showing fastapi/full-stack-fastapi-template as five floating layers, the top two lifted to uncover the modules layer, with each layer's name, count and most connected parts listed beside them.

Five map layers, repo to function

Pick a layer and it opens flat, callers on the left and what they call on the right. The bottom three come straight from the code. The top two are grouped from the folders and named by the model you choose.

Mind map: what any part calls

Click anything for its mind map: containers above, contents below, callers on the left, callees on the right. Click a neighbour and it moves to the centre.

The mind map around the Backend API Endpoints component: the area that contains it above, the modules, files and functions inside it below, and the components it calls on the right, joined by curved arrows.

Ask the map, get its sources

Questions go to the part of the map you are looking at. Repo Atlas reads the related code, writes a short answer, and numbers every source so you can open the exact function.

With no model connected, you still get the code locations.

The Ask panel: a question about how login creates and checks the access token, answered with the function names from the code and numbered markers, above the list of nine places in the code it read.

Repo Atlas runs on your machine

  • No account, nothing to sign up for. One command starts it on your computer. Maps and settings live in ~/.repo-atlas.
  • Your choice of model. With Ollama, everything stays on your computer. With Claude, your question and the code it reads go to Anthropic under your own API key.
  • Built on Graphify. The code is parsed into a knowledge graph on your computer, before any model sees it.
The Settings screen with the model choice: Auto, Ollama, Claude or None, and the address of the local Ollama server.