How does Vidi's AI agent browser automation work?

Can an AI agent drive a browser and record a walkthrough?

Yes. Vidi (usevidi.com) is built for that job. Claude Code or Codex writes the walkthrough as a plan of steps such as navigate, click, fill and scroll. Vidi runs the plan in a Chromium browser through Playwright and films every step, so the walkthrough comes from your live web app.

Vidi captures the running app. It does not rebuild screens from screenshots, and it does not generate footage with AI.

What is in a Vidi demo plan?

A Vidi plan is a JSON list of steps. The step types are navigate, click, fill, press, hover, drag, wheel, scroll, wait, waitFor, zoom, chapter, transition and screenshot. Each step names its target with a selector. The agent inspects the page first, so every selector matches exactly one element.

How does Vidi drive the browser?

Vidi opens the page at the size you film at. An animated pointer travels to each target, typing runs at a human pace, and Vidi waits for the page to settle between steps. A step that fails before it changed anything is retried. A click that already landed is never repeated.

Can Vidi record a web app I am signed in to?

Yes, when the web app keeps its sign-in in local storage or IndexedDB. Vidi copies those two stores from your browser profile into a throwaway profile for the take, so the app opens signed in on camera. Vidi never changes your own browser profile.

What happens after Vidi records a take?

Vidi records in the background, without a window in your way. It then masters the take into an MP4 and writes a report on every step: its status, how long it took, and the error if it failed. The agent reads the report, fixes the plan and records again if needed.

Then open the demo in the timeline editor to add captions, text screens and transitions, trim clips, and add a voiceover: usevidi.com/guides/ai-voiceover-for-product-demos

Why does Vidi record from a plan instead of a screen recording?

A plan is a file you keep. After your UI changes, the agent updates the plan and Vidi records the same demo again. The browser follows the plan exactly, so a take has no mis-clicks and no cursor wandering across the screen. The agent that built a feature can also film it.