Browser Use is for tasks that need your authenticated session or a real browser: checking messages, posting, reading analytics, extracting from pages behind a login. For public web questions the agent uses web search and page fetching instead, which are faster and free.
How a browser run works
You never pick a tool. Describe the task, and when it calls for a real browser the agent starts a browser run:1
The run starts and a card appears
The card in chat shows the goal, the live status, and the current step. A side panel opens with the live browser on top and a rolling activity log below: what the browser is doing, what the agent is thinking, and what it says along the way.
2
Watch, or step in when asked
The live browser is interactive. If a site wants a sign-in, a code, or a human check, do it there. Use the expand control for small forms. Passwords and codes go to the site, never to the agent.
3
The agent reports back
When the run finishes, the agent writes the result into the chat the way it answers any other question, with follow-up suggestions. The card collapses into a receipt: duration and credits charged. The panel keeps the full report, any files the browser produced, and the activity log.
Signing in for the first time
Sites you have not signed into yet stop the run at their sign-in page. The agent tells you so and points you to the live browser on the card.1
Sign in inside the live browser
Open the panel if it is closed, sign in on the site as you normally would, and finish any two-factor step. The browser stays available for about ten minutes after a run ends.
2
Say “continue”
The next run picks up in the same browser, already signed in. From then on every run starts signed in, in this conversation and in new ones.
Continuing and stopping
- Continue. After a run ends, asking for more on the same site continues in the same browser: page, login, and what the browser already did carry over. Say “continue” or just describe the next step.
- Cancel. The card and the panel have a Cancel control. The browser finishes its current step and hands back what it found, which can take up to two and a half minutes; the card and panel say “Stopping” while that happens. The agent then reports the partial findings in chat, and you can continue or ask for something else. Credits are charged for the work the browser did before the stop.
- Time limit. A run stops on its own after 15 minutes. The card shows “Stopped”, and the agent explains in chat that the run reached its limit and what it had found so far; “continue” picks up where it stopped. Large jobs go faster as several smaller asks than as one long one.
Using files in a browser task
Ask the agent to use a file you uploaded in chat or one it generated, such as “Upload the six images you just created to this draft.” The agent transfers the selected files into the browser before starting. You do not need to download generated files and attach them again. A browser task accepts up to 10 files, 10 MB per file and 50 MB total. If a selected file is missing, empty, too large, or cannot be transferred, the run stops before the browser starts. Files the browser writes, such as a CSV it extracted, are copied into your conversation and appear as download chips on the card and in the panel.Your saved logins
Your sign-ins live in a single browser profile shared by every run: the cookies and site sessions accumulate there as you sign in, so each new run on a site you have used before starts already signed in. Signing in always happens inside a run — there is no separate place to sign in ahead of time or manage saved logins outside of one.What you can automate
Browser runs act as you. For irreversible actions such as posting, sending, or deleting, the agent confirms with you first unless your request clearly authorized it.
Privacy and security
- Saved logins are stored by our browser provider, Browser Use Cloud, as browser state (cookies, local storage) scoped to your profile. CREAO stores only a reference to the profile, never your passwords or session cookies.
- You sign in inside the live browser. Credentials go to the site; the agent never sees them.
- Page content the browser visits during a run, including pages behind your login, is processed by the provider to run the task. The provider’s public privacy policy permits it to use content submitted to its agent to train or improve its models, and it describes no opt-out; retention and the full disclosure are on the privacy page. Use browser runs on accounts and pages you are comfortable sharing with the provider under those terms.
- Your profile is scoped to your account. Sign out on the site itself to end that site’s session there.
- The provider API key stays server-side and is never exposed to the agent sandbox or the browser.