Its own machine
A real desktop with a browser and your apps, not a hidden script.
Agents and their computer
Each agent works on a real computer with a browser, files and your apps. Watch its screen, take the keyboard when it needs you, then hand it back.
Book a callA real desktop with a browser and your apps, not a hidden script.
Take over for a sign-in or a code, then hand it back.
Each action it takes is logged with when it happened and what it touched.
1.0Live screen
The agent’s screen sits in the conversation, live. Watch it full size, or pop it into a corner.
Every reply that used the computer carries its screen, with Take over right under it.
Type while it works. Steer changes the task at its next step; Queue waits until it’s done.
Keep the screen in a floating window that snaps to a corner.
Each step is named as it happens: “Reading Stripe webhook logs”.
Under each reply: the steps it took, files it saved, apps it used.
One button, or ⌘ full stop, ends the task where it is.
2.0Take control
When an agent hits a sign-in, a code or a captcha, it asks and waits. You finish the step and hand it back.
Its row in the sidebar says it needs you, and the chat names the step it’s waiting on.
The agent, you, or nobody: the screen names it.
One button gives the keyboard back, and the task continues.
Expired sign-ins are named, with the one step to fix them.
The screen turns amber until you return control. The task waits for you.
If a task fails, its progress is saved. Fix the blocker and the agent picks up from there.
You can always see what it’s doing. Every agent works on a screen you can watch, and the keyboard is yours whenever you want it.
Bring a real task. We’ll show an agent doing it on its own computer, in your apps, and asking before it acts.