The Live Browser

The Browser tab in Studio Mode is a real browser (Chrome) running for you in the cloud. The agent can use it like a person would, and you can watch every step. You can also click and type in it yourself, for example to sign in.

What the agent can do

The agent sees a screenshot after each step, so it knows what's on the screen. It can also show you a screenshot in its answer.

Try asking:

• "Open our pricing page and check that every button works."
• "Compare the prices of these three laptops on their official sites."
• "Review the design of example.com and tell me what to improve."

Watch the agent work

While the agent uses the browser:

• A blue arrow shows its pointer, labelled with what it's doing: Agent · Clicking, Typing, Scrolling down, Pressing Enter, Opening example.com or Taking a screenshot.
• A blue ring flashes where it clicks.
• The page gets a blue border and the note "The agent is using the browser".

While the agent is working, you can't click or use the address bar, so you don't get in each other's way. Control comes back to you a few seconds after the agent's answer ends.

Use the browser yourself

When the agent isn't working, the browser is yours:

• Click, scroll and type right on the page.
• Use Back, Forward and Reload.
• Type a website or a search in the address bar ("Search or enter a website") and press Enter.
• Save a screenshot downloads a picture of what you see.
• Full page screenshot saves the whole page, top to bottom, in the Screenshots tab.
• Open in a new tab opens the same page in your own browser.

The browser stays open after the agent is done, so you can keep using it.

Sign-ins, passwords and payments

You and the agent share the same browser, so when a site needs you to sign in, you type your password in the Browser tab. Once you've signed in, the agent can carry on.

If a page blocks the agent, for example with a captcha or a login wall, it stops and tells you instead of trying to get around it.

Your sign-ins are remembered in your browser, across all your Studio chats.

Each chat has its own page

• A new chat starts with an empty browser.
• When you go back to an older chat, the browser opens the last page that chat was on.

Open your app while you build it

The agent can test the app you're building:

1. Ask it to start your project's dev server, for example "Start the dev server and open it in the browser."
2. It runs the server in the background (you'll see it in the Terminal tab) and opens it at localhost in the live browser.
3. Now it can click through your app, check the design and fix what's broken.

Only your own Studio can reach your dev server. It needs a project open, and the server must be started in the same chat.

When the browser closes

The browser keeps running while you use it. After about 30 minutes with nobody using or watching it, it shuts down to save resources. Your sign-ins and each chat's last page are kept, so it opens where you left off next time.

If something goes wrong

FAQ