Meet the Yantra Computer: A Cloud Desktop Your Agent Drives

Most AI assistants can tell you how to do something. Fewer can do it. The ones that can usually do it somewhere you cannot see.

The Yantra Computer is a desktop that belongs to your account. It has a browser, a terminal and a file system, it persists between sessions, and the agent in your chat drives it. You watch it happen in a window beside the conversation. When a site needs your password, a verification code or a payment, you take control, do that part, and hand it back.

What it is for

• Operating a live website. "Open my analytics dashboard and export last month's report." The agent opens the site, clicks through, downloads the file, and tells you where it is.
• Multi-step web tasks. Filling a signup form, comparing five products with screenshots, checking whether you are still signed in somewhere.
• Anything that needs a shell. Clone a repository, run the tests, read the failures.
• Work that should keep going. The workspace is durable. Files you leave there are there tomorrow.

It is not for simple lookups. A question web search can answer is answered in chat, without waking a desktop.

What it is not

It is not your own computer. The Yantra Computer runs in the cloud, on its own IP, with its own sessions. That is a feature when you want isolation and a machine that is always available. It is a limitation when a site only trusts your own logged-in browser. For that case, the Yantra browser extension lets the agent drive your real Chrome instead.

How it feels to use

Ask for something in chat. If the request needs a browser, Yantra routes it to the Computer and wakes it the moment the turn is routed, so the cold start is mostly hidden. A strip appears in the conversation: Working on your Computer, with a Watch link. Click it and the desktop opens on the right: a Browser window with an address bar, a Terminal, a Files window, and a dock.

When the agent is done the strip says Done on your Computer. Open it to see the result, or keep chatting.

Taking control

Press Take control of the browser and the browser is yours: click, type, sign in, approve the payment. Press Stop controlling and the agent continues from where you left it. Nothing you typed is visible to the agent unless it is on the page afterwards.

Settings that matter

• Skip authorization for this browser. Requests from this browser are approved automatically. Other browsers still ask.
• Allow cross-browser tasks. Use this browser for tasks you start from your phone or another client.
• Auto-open Computer panel. Show the desktop whenever the agent uses computer tools.

An email notification for Computer tasks tells you when something finishes while you are away.

Where it shows up elsewhere

• Shopping uses the Computer's browser to check watched prices every hour.
• Your assistant page shows a live glance of the same desktop.
• Workspaces can provision an OpenClaw agent workspace with a browser runtime of its own.

Try it

• "Open example.com and tell me what the page says."
• "Fill out the signup form on this site for me."
• "Show me the computer so I can watch you do it."

More in the Computer help article.