Get Atlas

Atlas maps your system the way the map on the homepage maps Atlas. One engine file, one JSON per system. Tell us where to send the setup directions and we'll follow up with the install steps, the agent skills, and help pointing it at your first project — the quick version is on the right, and it works today.

What you'll be doing

Atlas is in private beta — the public repo is being prepared. Sign up and we send you the engine plus these directions. Node 18+ is the only requirement: no build step, no framework, no account, nothing phones home.

Get the engine. We send you the repo access (or a zip, your call). It is one HTML file, one sync script, and four agent skills.
```
cd Atlas
```
Scaffold an instance inside the project you want to map. One command creates the map file, the sync script, an empty data file, and a history folder.
```
node scripts/init-instance.mjs ../my-project/atlas my-project
```
Install the agent skills into whatever harness you use, so your agents can build and maintain the map for you.
```
node scripts/install-skills.mjs
```
Let an agent map the system. Point it at the atlas-generate skill — it fans out readers over your docs, code, tracker, and incident history, then synthesizes one deduplicated graph into atlas-data.json.
Sync and open it. The sync script validates the graph (it refuses dangling edges), writes it into the page, and snapshots history for the time scrubber.
```
cd ../my-project/atlas && node sync-data.js
```
Open atlas.html in a browser. That's the whole loop — edit the JSON, sync, reopen. Host it anywhere static when you want to share it.

From then on, atlas-update keeps it honest as things ship and break, atlas-navigate answers "what should I work on" straight out of the data, and update-instance.mjs pulls engine upgrades without touching your map's content.