Describe a test. Watch it run. Review the result.
You write the task and the expected result. An agent runs it in a real browser and records what happened. There is no test code to write or maintain.
Write the test the way you'd explain it to a coworker
Type what a user should be able to do and what you expect to see. The agent works out the clicks. It also names the test and writes a short description, so the list stays readable as it grows.
- A sentence or two. No selectors, no scripts, no framework to install.
- Name and description filled in for you, editable any time
- Optional limits on steps and time, plus extra instructions when the agent needs them
Watch the agent work in a real browser
Each run uses a real Chrome session in the cloud. Watch it live, or close the tab and come back later. When the agent hits something it can't resolve on its own, like a two-factor code, it pauses and asks you.
- Live view of the browser while the run is in progress
- The agent asks when it's stuck. You can answer, or take over the browser.
- Runs keep going after you close the tab
See what happened when a test failed
Every action, console line, network request, and JavaScript error is recorded and lined up with screenshots and video. A failed run shows you the request that returned a 500 and the error that followed.
- Step-by-step timeline with a screenshot for each action
- Console, network, and error logs, with secrets redacted
- Click any row to jump the replay to that moment
Run the same tests after every deploy
Connect GitHub, GitLab, or Vercel and pick the tests that should run after a successful deploy. No pipeline config, and nothing to remember.
- GitHub App, GitLab project hook, or Vercel webhook
- Choose which branches and environments start a run
- One deploy starts one round of tests, even if the provider sends the event twice
A saved login for every role, kept out of the model
Save a login for each role your app has, like Admin, Viewer, or a second tenant. Passwords are encrypted, typed into the page by the system rather than the model, and redacted from logs. You can also set rules the agent has to follow on every step.
- Several labeled logins per application
- Passwords encrypted at rest and never included in model prompts
- Allowed domains, plus rules like “never submit a real payment form”
Attach the data the test should use
Upload a CSV or a fixture file, or pin exact values like a coupon code. Leave it empty and the agent makes up realistic data for each run. In a parallel run, each browser gets its own.
- Attach files to any test
- Pin exact inputs as key: value pairs
- Small text files are given to the agent in full on every run
Check the same flow in several browsers at once
Run one test in up to ten browsers at the same time and watch each one finish. It's a quick way to catch double submits and race conditions in signup or checkout. Each browser uses one run from your plan.
- Up to 10 browsers per run, depending on plan
- Each browser uses its own test data
- A separate pass or fail for each browser, with its own replay
Give your app a sanity check.
15 free runs a month. No credit card. Setup takes less than two minutes.