Skip to content

Taking control and signing in

Take the wheel mid-task, sign in to a site yourself, hand back, and know what happens to the browser afterwards.

Last updated September 01, 2026

On this page

When the assistant is working in a browser, you are never locked out of it. You can take the controls with one click, do the part that needs you — a sign-in, a code, a decision the assistant should not make — and hand it straight back.

This page covers watching a run, taking over and handing back, and signing in to a site so later runs start signed in. For what computer use is, when the assistant reaches for a browser and what it costs, see Computer use.

Watching a task

The panel opens by itself when a browser task starts, headed Computer, with the live browser inside it. Its header carries Take control, Stop, a close button, and — on a wide screen — a toggle that expands and shrinks the panel.

Take control and Stop only appear while a browser is actually open. Once the browser has closed, they are gone.

The close button closes the panel, not the browser: the task carries on without you watching. The card in the conversation is how you bring the panel back — Watch while it is working, Open while you are driving, Wake if the browser has gone to sleep, Reopen once it has ended.

Taking the controls

There are two ways to take over:

  • Press Take control in the panel header.
  • Click anywhere on the live browser view. Hovering over it shows a pill reading Click anywhere to take control. That works while the assistant is working, and also while the browser is just sitting open after a task.

The assistant stops immediately, and it will not carry on clicking behind you. You get the confirmation You're driving. The assistant will wait., and the status chip and the card both change to You're driving. If the change does not go through, you will see Couldn't change who's driving instead.

While you are driving

The assistant is told, every turn, that you have the wheel. It will not click, type or navigate while you do. If you ask it to do something in the browser during that time, it says it will pick it up as soon as you hand back.

A browser you have deliberately taken control of is not put to sleep out from under you while you are working in it.

Handing back

Take control becomes Hand back. Press it again to return the wheel.

The hand-back is explicit, and the assistant is told at once: a short message is posted in the conversation on your behalf, saying the browser is back with it and which page it is on. You will see Handed back — the assistant knows and is picking it up., or, if that message could not be sent, Handed back. Ask the assistant to carry on. — in which case just tell it to continue.

The assistant then reads the page again before it does anything, and carries on from where you left it rather than starting the task over.

Signing in to a site

The assistant never signs in on your behalf. Instead it opens the browser on the site's own sign-in page and hands it to you. Unlike an ordinary browser task, this one always asks first.

Approve the sign-in

The assistant proposes it and you get a confirmation card with Cancel and Approve. Nothing opens until you approve. In the conversation, the card for it reads Sign in to continue, with an Open and sign in button.

Sign in yourself

The browser opens on the site's sign-in page with an amber strip across the top of the panel. Type your details, and any two-step code, yourself. The assistant is not driving at all during a sign-in — a person has the wheel for the whole of it.

Tell it you are done

Press I’m signed in. You will see Saved. Future runs will start signed in. If it could not be saved, you get I couldn't save that sign-in — nothing was stored. Try signing in once more.

Note. The panel says it in plain words above the box: Sign in below and the session is saved for future runs. This window isn’t recorded, and your password is never stored — only the signed-in session is kept.

Your saved sign-ins

Saved sign-ins are per site, and you name them yourself. Once one is saved, later runs on that site start already signed in, with nothing for you to do.

You manage them by asking the assistant in the conversation — there is no settings screen for them. Ask it to list them and it tells you which are ready and which have gone stale. Ask it to remove one and the signed-in browser it was tied to goes with it, so you really are no longer signed in there.

Sign-ins do go stale on their own: a password change, a “log out of all devices”, or a session the site expires. When a run finds itself logged out it says so and asks you for one more sign-in from the panel, rather than retrying blindly.

One saved sign-in cannot be used by two runs at the same time. The second is asked to try again in a minute, rather than risking the site logging you out.

What happens to the browser afterwards

When a task finishes, the browser does not close. It is left open on the page it reached: the card reads Done — browser still open and the chip reads Yours. You can click straight into the live view and carry on by hand, or ask the assistant for the next thing and it continues in the same browser.

After a stretch with nobody using it, the browser puts itself to sleep and the panel shows Your computer is asleep. Press Wake it up, or simply ask for the next thing and the assistant wakes it for you: it wakes up where you left it, still signed in, and you will see Woke your computer up where you left it.

Worth knowing. Because the browser stays open after a task, it keeps using credits for as long as it is open — including while you are driving it yourself. Asleep costs nothing, and pressing Stop closes it straight away.

If the live view drops

The panel tells you and offers a Reconnect button: The live view disconnected., or The live view keeps dropping. Reconnect to pick it back up. Reconnecting picks the browser back up where you left it, and confirms with Reopened your browser where you left it.

If the browser has actually gone, you will see That browser has closed. — sometimes with Ask the assistant to open a new one. — and if a reconnect cannot be done, Couldn't reopen the browser. Ask the assistant to start one. Asking the assistant for the task again starts a fresh browser.

Stopping a run

Stop ends the run and closes the browser. The chip reads Stopped, and it is recorded as you having stopped it rather than as a failure. This is also how you close a browser you have finished with instead of waiting for it to fall asleep.

Stopping is not the same as a refusal. If the assistant stops at a password field, a “confirm you're human” check or an action that would change something, the browser stays open on that page so you can take the controls, do that part yourself, and tell the assistant to carry on.

Next steps

Was this page helpful?