Docs menu

Operate

Browser checks

Load your app's important pages in a real browser every few minutes, click through a flow, and hear about failures with a screenshot.

Explain it

A browser check opens a page of your app in a real browser every few minutes, fills in forms and presses buttons, and checks the right thing appears. It catches a broken sign-up page or checkout before your customers do.

How to do it: on the project page, under Browser checks, pick an app, a page and some text that should appear on it. If it fails twice in a row you get an alert with a screenshot.

"Add a browser check that our sign-up page shows 'Create your account'."

Each check run costs a small amount from the project's credit.

A health check says the server answers. A browser check says the sign-up page actually works: it opens the page in headless Chrome, fills in the form, presses the button and checks that the right thing appears. It runs every few minutes, so a broken deploy, an expired certificate or a third-party script that took the page down gets noticed in minutes, not when a customer complains.

Add one

On the project page, under Browser checks, pick an app, a page and text that should appear. For a full flow, use the API:

API
POST /v1/projects/:id/browser-checks
JSON
{
  "name": "Sign-up works",
  "serviceId": "svc_…",
  "intervalMinutes": 5,
  "maxLoadMs": 3000,
  "steps": [
    { "action": "goto", "path": "/signup" },
    { "action": "fill", "selector": "#email", "value": "[email protected]" },
    { "action": "click", "selector": "button[type=submit]" },
    { "action": "expectText", "text": "Check your email" }
  ]
}
ActionDoes
gotoOpens a path on the app (or a full URL)
click, fill, pressClicks an element, types into a field, presses a key
expectText, expectSelector, expectUrlFails unless the text, the element or the URL shows up
waitPauses up to 10 seconds

Use url instead of serviceId to check any https page. Each check also fails if the first page answers 400 or above, or takes longer than maxLoadMs.

When it fails

Two failures in a row raise browser_check_failed (one blip doesn't wake anyone), with the error, the step that failed and a screenshot of the page at that moment. JavaScript errors in the browser console are recorded too. The next pass raises browser_check_recovered. GET /v1/browser-checks/:id lists recent runs, and GET /v1/browser-check-runs/:id/screenshot returns the image.

Use test accounts and data for flows that write: the check runs for real every few minutes.

Cost and setup

Each run costs a twentieth of a cent. The server needs Playwright's Chromium (npm i playwright && npx playwright install --with-deps chromium on the control plane). Without it, checks load the page over HTTP and check text in the HTML; steps that need a browser report that they couldn't run. Agents use cli_browser_check.