Browser Use: Jev AI Browser Agent Demo

One goal becomes a sequence of clicks

Browser Use’s Jev experiment turns the visible page into an indexed list of controls. Jev chooses an operation and its target; a separate text model supplies text when typing is needed. The browser executes the action and the loop observes the page again.

The published recording searches for flights from Zürich to London and stops at matching results. It does not book a flight. This gives you a concrete task to follow without mistaking a search demo for a completed purchase.

Watch the handoffs

• Look at how the page’s available controls change between steps.
• Notice the difference between selecting a target and generating field text.
• Follow the final check: did the page reach the requested result?

Try the underlying pattern

Pick a short, read-only browsing task with an outcome you can verify. Keep a trace of observations and actions so that a wrong turn is inspectable. The original repository includes a local inspector and setup instructions; it requires your own environment and API configuration.

What this demo does not prove

Reported timings belong to the author’s particular run. Different websites, loading delays and task complexity can change the result. JevFlow’s card is an illustrative interface; the recording below is the creator’s original media.