Skip to content
Simon Rendon Arango

Lab

Give my agent a job

Name a small web app. A model writes the code and 8 tests. The tests run in your browser, and when one fails, the model repairs the code up to 2 times.

3 jobs a day

Try
  1. Read: not started
  2. Plan: not started
  3. Build: not started
  4. Test: not started
  5. Repair: not started
  6. Pass: not started
00:00

Waiting for a job.

The plan appears here.
Sandbox. Scripts only, no network.

The app appears here while the agent writes it.

Tests

The model writes them from the plan, without the code. Each test runs on a fresh copy of the app.

No tests have run yet.

The tests appear here.

A job makes 3 calls to an OpenAI model, plus up to 2 repair calls. What you type goes to OpenAI. The app runs in a sandboxed frame with scripts only. It cannot reach this page, your cookies or the network. The model refuses phishing pages, password forms and other harmful jobs.