It uses the apps that have no API
Consumer apps, internal tools, anything behind a login that never published an integration. If you can do it on a screen, it is in scope — no partnership, no API key, no waiting for a vendor roadmap.
ClawBot Crew · Computer Use Agent · Private beta
The Computer Use Agent works a real desktop in a session dedicated to you. It opens the applications you already have, clicks the buttons, types the text and waits for things to load — the way you would. That matters because the tools worth automating are usually the ones that never shipped an API.
Private beta, not a released product. Each tester needs their own desktop session, so places are limited by hardware rather than by a queue.
The point is not novelty. It is that a desktop is the one interface every tool already supports.
Consumer apps, internal tools, anything behind a login that never published an integration. If you can do it on a screen, it is in scope — no partnership, no API key, no waiting for a vendor roadmap.
Your desktop is dedicated to you. Your logins, files and windows are not on a shared machine with another user, because they are not on the same desktop at all.
It takes a screenshot, decides, acts, then looks again to confirm what happened. That is slower than an API call and far more honest about failure — it can tell you the button was not there.
It is a crew member, not an island. The Content Agent can ask it to publish something; the research or coding agents can hand it work that only exists behind a UI.
A real example, end to end, from the beta.
Describe the outcome, not the clicks. "Make me a 60-second video on this topic and post it everywhere." You are talking to your crew, the same as always.
YouOn first use it opens the app and hands you the login — a QR code or a sign-in screen — and you authenticate once, exactly as you would on any new machine. It does not hold your password.
YouIt writes a storyboard, asks the Gemini app for six ten-second clips, and assembles them into one sixty-second video.
Separately it has NotebookLM turn a trending topic into a podcast and a video, then shares both with a community.
AgentYou get what happened and what it produced. When a step fails — a layout moved, a login expired — it says which step and what it saw, instead of reporting success it cannot prove.
AgentWritten the way the beta testers actually phrase it.
A real desktop means real constraints. These are the ones worth knowing before you sign up.
It waits for pages, dialogs and uploads like a person does. For anything with a real API, the API is faster and this is the wrong tool. This exists for the rest.
Driving a desktop means each service's own rules on automated use apply, and some restrict it. You are operating your own accounts and you remain responsible for them.
When an app redesigns a screen, the agent has to find the new one. It is built to notice and tell you rather than click confidently in the wrong place, but expect a rough edge on the day a vendor ships a redesign.
Each tester needs a dedicated desktop session on real machines we run. That is a physical constraint, which is why this is a waitlist rather than a signup button.
Leave your email and you will hear from me directly as places open up — plus what the agent is actually doing as it gets better.
One email, no password. You will be added to my newsletter so I can send beta updates, and every issue carries a one-click unsubscribe. No spam, and I will not pass your address on.
Nothing yet, because there is nothing to buy yet.
Not on sale. Beta testers are not billed for it. When capacity allows a real product, the price will be published here before anyone is charged.
The released crew agents are self-serve, with a free trial and 1,500 credits a month, on top of a $9.99 base plan.
The Computer Use Agent operates a real desktop and signs into the accounts you give it, acting as you. Individual services set their own rules on automated use and some restrict it; you use this at your own risk and you remain responsible for anything done from your accounts. It is a private beta, not a released product, and should not be relied on for anything time-critical.
No. You sign in yourself on the agent's screen, once per app, the same way you would on a new laptop. The session stays logged in afterwards; the password does not pass through us.
A browser script drives one browser and breaks on the first selector change. This drives the whole desktop — any application, not just web pages — and it looks at the screen to decide what to do next.
The honest trade is speed: it is slower, and it is the only option for software that never shipped an API.
No. Each tester gets their own desktop session. There is no shared desktop, so there is nothing to see across users.
It depends on hardware — each tester needs a real session on a real machine. Join the waitlist and you will hear from me directly rather than from a queue position.
Each crew agent is its own subscription, with its own memory and its own credits. They can hand work to each other when you tag them in chat.
The tools worth automating are usually the ones without an API. This is how you reach them.