Skip to content

Skip to content
o1/cloud browserOpen workspace
Cloud browser / How it works

THE SESSION / FROM BRIEF TO BOUNDARY

One clear task.
One step at a time.

A browser session is a place to explore a task with your agent. Observe the page, review the next action, and choose when to stop.

YOUR OWN AGENT
A temporary place to work.
OpenInspectReviewEnd

Ten-minute maximum. No persistent browser profile.

FIVE MOMENTS THAT MATTER

The whole flow,
with you in the loop.

01

Choose the agent you want beside you.

Sign in with a verified account, then select an agent you own. A browser session belongs to that account and that agent. Give the task one concrete outcome before you begin.

A useful brief: find the delivery policy on this public website and let me review the link before clicking.

02

Start a fresh, temporary browser.

The workspace checks provider availability and usage budget before starting. Browserbase supplies the isolated session. Navigate to a public HTTPS page and work within a ten-minute maximum.

Provider setup is required. A setup message is not a running browser or a completed task.

03

Inspect the protected page view.

Refresh the screenshot to observe a static view of the public page. Use its source address, title and visible content to check where you are. Website scripts and external resources are disabled, so the view can differ from the original site.

Some content needs JavaScript and will be missing. There is no continuous video or manual browser takeover.

04

Review one proposed action.

Ask GPT to prepare the next step from the screenshot. It can propose a public-link click or scroll. Review the specific target and action before choosing to execute it. Then inspect the page again.

When the target is unclear, leave the proposal unapproved and ask for a clearer next step.

05

Finish the task and end the session.

Use End session to stop the provider browser. The ten-minute provider limit also bounds the session. The next task begins in a new browser; this version does not save a persistent profile or continue working in the background.

Starting a browser does not approve a chain of actions. Each proposed step has its own review.

THE REVIEW IS A REAL STEP

Look at the action.
Look at the page.

The most useful approval is specific. Check what the action will do against the current screenshot, then decide.

THE TARGET

Is it the right link?

Compare the proposed click location with the screenshot. Check the link and prepare a fresh proposal if the page changes.

THE SOURCE

Is it the page you intend?

Read the source address and visible text. Protected views may omit dynamic content from the original website.

THE EFFECT

Are you ready for this step?

A link opens another checked public page. Hold the action if its destination or purpose is unclear.

Read the controls and limits

THE CURRENT VERSION

A browser.
A bounded session.

Build the habit of short, observed tasks. The product is scoped to the browser workflow described here.

Available when configured

An owned agent, temporary Browserbase session, public HTTPS navigation, refreshed screenshots and individually reviewed proposed actions.

Session lifetime

Ten minutes maximum. End session stops the provider browser. Browser profiles are ephemeral.

Public research only

Website sign-ins, file uploads and form submissions are outside this version. Keep the task focused on reading and exploring public pages.

Outside this version

Continuous video, manual takeover, persistent profiles, desktop apps, scheduled jobs and unattended background work.

Costs and availability

Browser and model usage need available budget. Check the workspace for provider configuration, current availability and limits before starting.

Check your workspace

GOOD TO KNOW

When a task
needs a pause.

Do I have to approve the proposed action?

No. A proposal can remain unapproved. Review the screenshot, clarify the request or end the session when an action is not what you want. Unapproved proposals do not run themselves.

What if the page has changed?

Refresh the screenshot and inspect the current page. A proposal based on an earlier screenshot may no longer target the intended element. Prepare and review a new action when needed.

Why am I seeing a setup-required message?

The deployment must have its Browserbase provider credentials and model configuration in place before live use. The workspace checks readiness and does not present an unconfigured provider as a running session.

O1 / YOUR NEXT MOVE

Bring a task you can review.

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