Skip to content
AI Building

Build an Atrium app with AI

One minimised file with everything an AI tool needs: how to work, the hard rules, the full app contract and the design system. Upload it to any LLM, then describe your app.

ATRIUM-AI-BUILD.mdMarkdown
  1. Upload the file to ChatGPT, Claude, Cursor or any AI tool (or paste its contents).
  2. Describe the app you want in plain English.
  3. Let the AI ask you for the API key when it needs one.
View the file
Auto

App sample

hello-python

View .md

Minimal Atrium web app using Python stdlib (http.server, hmac, hashlib, urllib).

Default port: 5101 (PORT env override).

Terminal window
python3 app.py

Your process reads ATRIUM_SETUP_SECRET to verify the first lifecycle.setup call. That is your app’s own setup secret: COHO mints it when you register the app and shows it once in the create response (regenerate it from My apps if lost). On Atrium Hosting, COHO injects the env var for you; when self-hosting, set it yourself before connecting. Set ATRIUM_DATA_DIR to persist connection secrets and config across restarts (hello-python-state.json). Set ATRIUM_FRAME_ANCESTORS and ATRIUM_PARENT_ORIGIN to the Store origin for the environment you embed from (same value for both). Playground and production values are in developer spec §6; when unset, these kits default to http://localhost:4200 for local ng serve.

No pip install required.

Same runtime contract as the other hello samples:

  • /health
  • signed /webhooks/atrium/*
  • signed native config at /config/schema and /config
  • signed iframe quick view at /ui (launch URL includes atriumUsageToken — prefer that Bearer for interactive Public API calls; connection apiKey for background — spec §6 / §8)
  • external app-owned page at /configure

For local Store registration, use manifest.example.json.

Signature verification is pinned to the published test vectors in the app developer spec.

Terminal window
python3 -m unittest test_app.py -v