An AI assistant that uses your computer and browser
GAIA can do real work on your computer, within clear limits. From the desktop app it can take a screenshot, read and write your clipboard, open links and apps, and list your open windows. For anything on the web it can run a browser task: a browser agent in a cloud browser that navigates sites, fills in forms and reports back, with you able to step in. What it cannot do is take over your mouse and keyboard to drive any app on your machine.
What GAIA can do on your desktop
These actions run through the GAIA desktop app, only in conversations that start there. They act on your own machine and are easy to undo, so GAIA runs them without an approval prompt:
- Take a screenshot of the display you are on and read it
- Read the text on your clipboard, or put text there for you to paste
- Open a link in your default browser
- Open an app by name (macOS only, for now)
- List your open windows by app and title (macOS only, for now)
What browser tasks can do
For jobs that live on websites without an integration, GAIA hands the work to a browser agent. It runs in a headless Chromium browser that GAIA hosts, not in the browser on your laptop, so it keeps going while you do something else. You follow it in a card in the chat, and the result comes back to the conversation when it is done.
Good fits are the tedious, click-heavy errands: filling in a booking or signup form, checking an order status on a site with no API, collecting details from several pages into one summary, or looking something up behind a login.
Using sites you are logged in to
If a task needs your account on a site, you can import your browser logins with the gaia connect tool. It detects your browser, asks which sites to sync, and copies cookies only, never passwords. Saved logins appear in GAIA's settings with an expiry date, and you can forget one site or all of them at any time.
Where you stay in charge
In GAIA's Always ask or Auto approval modes, GAIA checks with you before it starts a browser task at all. During a task, the agent hands control back to you for payments, sign-ins, CAPTCHAs and anything it cannot undo, instead of pushing through on its own. One browser task runs at a time per account.
What it can't do
- It cannot drive arbitrary desktop apps with your mouse and keyboard. Its computer actions are the specific ones listed above.
- Browser tasks run in GAIA's cloud browser, not inside the Chrome or Safari window on your screen.
- It does not make phone calls, answer calls or control your phone.
- Opening apps and listing windows are macOS-only today; Windows and Linux get screenshots, clipboard and links.
Common questions
Keep reading
An AI assistant that can see your screen
How GAIA's screenshots work and what it does with them.
What is an AI agent?
The difference between answering and acting.
Human-in-the-loop
Why the best agents hand risky steps back to you.
ChatGPT Atlas alternative
How GAIA compares with a browser built around an agent.
