How do I watch the browser and take the controls?

Every browser task runs in front of you. The panel opens itself when a run starts, shows you the browser moving in real time, and lets you take over with a single click whenever you want to.

The panel opens itself

The moment a browser task starts, a panel headed Computer slides in from the right-hand edge of the chat. You do not open it, and you do not have to keep it open — it keeps itself up to date while a run is going, so it is current whenever you come back to it.

Note: there is no separate button or menu entry for the Computer panel. It opens itself when a task starts, and after that the card in the conversation is the way back to it. The card's button reads Watch while it is working, Open while you are driving, Wake if the browser has gone to sleep, and Reopen once the run has ended.

What is in the panel

The header has four controls: Take control, Stop, an expand and shrink button, and a close button. Take control and Stop are only there while a browser is actually open; once the run has ended they disappear.

The header also carries a small status label. It reads "Starting", "Working", "You're driving", "Yours", "Done", "Failed", "Stopped", "Closed" or "Asleep" — except while it is actually working, when it narrates the current step instead of just saying "Working", so you can read what it is doing rather than only that it is busy.

Making it bigger

The expand control widens the panel and folds the app's side navigation away to make room; shrinking puts everything back. Your choice is remembered for next time on that device.

On a wide screen the panel sits beside the conversation and the chat makes room for it. On a narrower screen it slides over the conversation with the background dimmed, tapping the dimmed area closes it, and the expand control is not shown at that size. On a phone it takes the full width.

Taking the controls

Two ways, and they do the same thing:

• Press Take control in the panel header.
• Click anywhere on the live browser view. Hover over it and a pill appears reading "Click anywhere to take control".

That pill is there while the assistant is working, and also while the browser is just sitting open after a task has finished. It disappears once you are the one driving.

The moment you take over, the assistant stops — not at the end of what it was in the middle of. It will not carry on clicking behind you. A short confirmation appears saying you are driving and the assistant will wait, and both the status label and the card headline read "You're driving".

For as long as you have the wheel, the assistant knows it. It will not click, type or navigate. If you ask it to do something in the browser while you are driving, it tells you it will pick that up as soon as you hand back.

Handing back

The Take control button becomes Hand back. Press it and the assistant has the wheel again.

Handing back is explicit, and the assistant is told immediately: a short message goes into the conversation on your behalf — "I've handed the browser back to you. Carry on from what's on screen now." (when the page is known, it names the page too). A confirmation appears saying the assistant knows and is picking it back up.

If that message could not be sent, the confirmation says instead to ask the assistant to carry on — in which case do exactly that in your next message. And if the switch does not go through at all, you will be told it could not change who is driving; press the button again.

What the assistant does with what you did by hand

This is the part worth knowing. After you hand back, the assistant reads the page again before it does anything, and carries on from where you left it. It does not start the task over.

So doing a step yourself is a normal part of the flow, not an interruption you have to undo. Sign in, clear a "confirm you're human" check, pick the option you actually wanted, fix a field it filled in wrong, click through to the right page — then hand back, and it picks up from what is on screen.

When it hands over to you on its own

Some steps the assistant will not take. When it reaches one it stops and asks you:

• A password, passcode or security code field.
• A "confirm you're human" check — it never attempts one.
• A payment card number.
• The last click on anything that would change something — buying, paying, submitting, sending, deleting, cancelling and the like — unless that is exactly what you asked it to do. It will fill a form in and stop short of sending it, including where pressing Enter would send it.
• A site outside the ones you scoped the task to.

None of these is a failure and none of them closes the browser. It stays open on the page it reached, the run waits for you, and the assistant tells you what it stopped at and asks you to do that part. Do it in the panel, hand back, and tell it to carry on.

Rephrasing your request will not get past any of them. They are enforced underneath the assistant, so asking again a different way only spends more time and more credits on the same answer.

Stopping it

Stop ends the run and closes the browser. The status label reads "Stopped", and it is recorded as you having stopped it — not as something that failed.

If the browser has already gone to sleep, pressing Stop does not throw anything away. It stays wakeable.

When the task finishes, the browser stays open

A finished task does not close the browser. It is deliberately left open on the page it reached, so you can carry on from there or ask for the next thing. The card reads "Done — browser still open" and the status label reads "Yours".

Whatever you ask next continues in that same browser — same tabs, same history, same sign-ins — including in a brand-new conversation.

When it goes to sleep, and waking it

After a stretch with nobody touching it, the browser puts itself to sleep. Asleep is not closed: your sign-ins, cookies, site preferences (down to a cookie banner you already dismissed) and the page it was on are all kept.

The panel shows a moon and "Your computer is asleep.", with a line underneath saying it will wake up where you left it, still signed in, and a Wake it up button. In the conversation, the card reads "Asleep — tap to wake" with a Wake button.

Usually you need neither. Ask for the next thing and the assistant wakes it itself — same page, still signed in, nothing for you to do first.

Two things worth knowing about sleep. A browser you deliberately took control of is not put to sleep out from under you while you are working in it. And it only stays awake while you are genuinely using it, so leaving the panel open on a second screen while you are somewhere else will not keep the browser running.

Note: waking brings you back to the page you were on, still signed in, but the page is opened again. Do not count on something you had typed into a page but not saved still being there.

If the live view drops out

Sometimes the picture goes while the browser itself is fine. The panel tells you which is which.

• "The live view disconnected." or "The live view keeps dropping. Reconnect to pick it back up." — press Reconnect to pick it back up.
• "That browser has closed." or "That browser has closed. Ask the assistant to open a new one." — that one is gone; asking for the task again opens a fresh browser.
• "Couldn't reopen the browser. Ask the assistant to start one." — do exactly that.

One thing to remember about cost

The meter runs for as long as the browser is open, including the time you spend driving it yourself and the time it sits open after a task has finished. It stops when the browser closes — you press Stop, or it goes to sleep. So if you are finished with it, pressing Stop closes it there and then rather than leaving it open until it falls asleep.