Run Your First Test

With a project in place, you're ready to write a test case and watch the agent run it in a real cloud browser. Writing your first one takes about two minutes. This walkthrough keeps things simple: a tiny scenario written in plain English.

Write a test case

1. Open Test Cases in the project sidebar and click New test case. (On a fresh project the empty state has the same button.)
2. Fill in the Details form:
   • Name — a short label, e.g. Pricing page.
   • URL — where the agent opens the browser. Paste a full public address, e.g. https://example.com. It must be reachable over the public internet.
   • Description — the scenario in plain English: the setup, the actions, and the outcome to verify. Write it the way you'd brief a teammate.
   • Screen size — Desktop (1366 × 768) or Mobile (390 × 844).
   • Test suites — leave empty for now; you can group the case later.

A first description might be:

Find the "Pricing" link on the page and navigate to it. Make sure three plans are displayed,
including the enterprise one without a price. Switch the billing toggle to annual and confirm
the effective per-month price drops (annual is discounted versus monthly).

State assertions explicitly — what must be true for the test to pass. See Test Cases for writing guidance that makes runs reliable.

Run it and watch

Click Run test. For a brand-new case this saves it first, then starts the run automatically.

The panel on the right streams the run live:

• The browser preview shows the cloud browser as the agent works — the address bar tracks the current URL and the canvas streams what the agent sees.
• The Activity Log narrates each step as it happens, with timestamps:
  • Info — neutral notes, like opening the URL.
  • Reasoning — the agent's thinking about what to do next.
  • Action — a concrete step it took (clicking, typing, and so on).
• Click Stop test to cancel a run in progress.

Read the verdict

When the run finishes, a badge appears over the final frame:

• Passed — every assertion in your description held.
• Failed — the agent reached a conclusion and the app didn't do what the description required. The final Activity Log entry gives the reason.

If the verdict isn't what you expected, change the description to make the assertions specific and unambiguous. Run it again. Live previews are for iterating quickly on a single case.

That's it — you've just watched an AI agent testing your app in a real browser. Good job!

Next steps