Can the agent act without me knowing?

No. computer use is off by default per computer, an on-screen border shows when an agent has control, and the emergency stop ends it immediately.

What stops a runaway session?

A per-chat step limit, the run's budget cap, the visible stop control, and an exclusive lease: one chat controls a computer at a time.

Does this work on cloud computers too?

Yes: the same consent, trace, and stop apply whether the desktop is your paired machine or a cloud computer.

All use cases

Computer use, safely

Desktop control with a hand brake.

Agents drive the screen under consent, a visible border, and an emergency stop.

Try now See pricing

Grant control, see control

Computer use stays off until you enable it per computer; while an agent drives, an on-screen border shows it, and emergency stop ends the run.

More about Computer Use

Scope the agent

The autonomy dial runs from read-only to full; ask mode confirms each consequential action before it happens.

More about Agents

Audit the machine

Every session lists in the computer's history: what ran, when, and what it did.

More about Computers

Built from these features

Computer Use

Agents see the screen, click, and type to finish real UI work: with your consent.

Computers

Daemon-connected machines with real filesystems and agent-friendly controls.

Agents

Skills, scoped permissions, persistent memory: any model you pick.

Frequently asked

Ready to work this way?

Create a free account and start in seconds.

Try now