Skip to content

Skip to content
o1/cloud browserOpen workspace

O1 / A BROWSER FOR YOUR AGENT

Give your agent
a browser of its own.

A little space to explore. A clear view of the next step.
Your agent works in a temporary cloud browser, with you reviewing the actions.

Open your workspace See how it works
Ten-minute sessions Review before execution Screenshot inspection

Protected static page views. Website scripts and external resources are disabled. Provider setup and a verified account are required.

EXPLORE THE SESSION FLOWIllustration · no browser running
Cloud browserONE AGENT / ONE SESSION
Your public HTTPS destinationSCREENSHOT
SPACE FOR YOUR NEXT TASKA fresh browser.
A focused start.

Provider setup is required before a real session can launch.

Your verified account owns both the agent and the session.

Temporary browser · authenticated screenshots · review before execution

Give a task its own browser.

Select an agent you own, check availability, and open a temporary browser. Each session has a ten-minute maximum.

A SMALL WORKSPACE / A USEFUL NEXT STEP

Your agent brings the brief.
You keep the decision.

Move from a question to a page you can inspect. The browser gives an agent a bounded place to work; the review step keeps each proposed action visible.

One task, in its own space.

Choose an agent you own, give it a clear browser task, and inspect a public page. Keep the session focused on an outcome you can review.

Follow the session flow

Your agent. Your session.

Verified access connects the browser to an agent you own. Session controls and screenshots belong to that signed-in account.

Observe before moving.

Refresh a screenshot, inspect the actual page, and compare the next proposal with what is on screen.

Approve a specific action.

Review the proposed public-link click or scroll. Approval applies to that action, one step at a time.

Finish with a boundary.

End the session when you are done. The provider also limits the browser to ten minutes; profiles are temporary.

START WITH SOMETHING SPECIFIC

A better brief makes
a clearer browser task.

Bring one public website, one goal and a point where you want to review the result. These are example task briefs, not completed work.

01 / FIND THE DETAIL

“Find the delivery policy
on this product site.”

Inspect the page, identify the relevant link, then review the proposed click.

02 / CHECK THE JOURNEY

“Find the getting-started
guide for this product.”

Explore public documentation and inspect the next page. Review the target before navigating.

03 / EXPLORE THE SOURCE

“Find the support article
for this setting.”

Use a public documentation site. Check the screenshot and source page before trusting the answer.

Shape your browser task

CONTROL IS PART OF THE PRODUCT

Keep a hand
on the next step.

Browser pages can change, and an agent can misunderstand a target. Screenshots and explicit action review help you decide what happens next.

Read the security and control guide
01

Inspect the current page.

Check the screenshot before approving an action. Refresh after meaningful page changes.

02

Check the proposal itself.

Review its target, text and effect. Hold an action when the page or intent does not match.

03

End a session you no longer need.

Stop the provider browser from the workspace. Sessions have a ten-minute outer limit.

A FEW CLEAR ANSWERS

Before you
open a browser.

The first version keeps the scope small and the next step visible.

Does my agent get a permanent computer?

This version gives your agent a temporary browser session with a ten-minute maximum. It does not persist a browser profile, install desktop apps, or run an always-on computer. Start a fresh session for the next task.

Can I watch or control the browser live?

You can refresh authenticated screenshots to inspect the page. This version does not provide a continuous live video stream or manual remote takeover. You review proposed actions in the workspace.

Can my agent act without asking?

GPT proposes one action at a time. A proposed public-link click or scroll waits for your review before execution. Starting a session is not approval for future actions.

Can it sign in, upload files or submit forms?

This version is for research on public pages. Website sign-ins, uploads and form submissions are outside its supported scope. Use screenshot inspection and reviewed actions to explore public information.

Will every website look and work the same?

The browser displays a protected static view of a public page. Website scripts, external images, fonts, embedded media and forms are disabled. Content that needs JavaScript may be missing. Use the source address to verify information and review public-link clicks or scrolling.

What do I need before starting?

Sign in with a verified account and select an agent you own. Live sessions also need the deployment's Browserbase provider setup and available usage budget. The workspace checks availability before you can start.

What happens when I end a session?

End session requests that the provider stop the browser, and the session becomes unavailable for further actions. Each browser also has a ten-minute provider limit. A finished session does not become a saved browser profile.

O1 / YOUR NEXT MOVE

Put a clear task in motion.

Choose your own agent and check session availability in the workspace. A verified account and provider setup are required for live browser use.

Open your workspace