Agent Browser Overview

Agent Browser gives your application an isolated, short-lived Chromium instance through one API: create a browser session, navigate to pages within an allowed domain list, observe the DOM to get element references, interact with elements (click, fill, select), and take screenshots. It is built as a primitive for AI agents — every action returns structured JSON ready to feed a model as tool output. The semantic observation model means agents interact with element references instead of pixel coordinates.

What You Can Do

Capability	Endpoint	Returns
Create Session	`POST /v2/browser_sessions`	An isolated browser session
Navigate	`POST /v2/browser_sessions/:id/actions`	Page load with URL and title
Observe	`POST /v2/browser_sessions/:id/actions`	Element references (`refs`) and a text tree of the DOM
Click / Fill / Select	`POST /v2/browser_sessions/:id/actions`	Interaction with elements by ref
Screenshot	`POST /v2/browser_sessions/:id/actions`	A downloadable screenshot artifact
Batch	`POST /v2/browser_sessions/:id/actions`	Multiple actions in one request (1–50, sequential)

Next Steps

Get Started

First session to screenshot in a few minutes.

Sessions

Session lifecycle, allowed domains, and limits.

Actions

All action types, batch, and the interaction loop.

Observations

The observe → ref → click/fill flow explained.

Agent Browser

Features

API Reference

For AI Agents

Agent Browser Overview

What You Can Do

Next Steps

Get Started

Sessions

Actions

Observations

​What You Can Do

​Next Steps

Get Started

Sessions

Actions

Observations

What You Can Do

Next Steps