Until now, every Dock agent ran in one place: your machine. That is still the right home for a lot of work, because an agent with their hands on your real files, your logged-in browser, and your own Claude account can do things no hosted bot can. But it carried one limit that everyone eventually hit. When your Mac slept, your agents waited.
Now you can hire a cloud agent instead. They get a machine of their own, they stay on when yours is off, and the work that suits them best is exactly the work you never wanted tied to your laptop: the overnight watch, the standing sweep, the routine that should fire at 6am whether or not anyone is awake. Hire a cloud agent to watch your Sentry issues and tell you what broke overnight, and the telling happens before you sit down.
The sign-in is one click
An agent that runs in the cloud needs a Claude login of their own, and until now that meant copying a token by hand. Now the machine does it. When you choose cloud at hire time, it opens an authorize URL; you click approve in a browser you are already signed into; that is the whole interaction. Nothing typed, nothing pasted. You are asked at the moment you choose cloud, not after you have finished hiring and wondered why nothing happens.
The part worth slowing down on is whose credential it is. It is issued to you, not to us. The agent's own Claude CLI is the OAuth client: it picks the client id, the redirect, and the PKCE challenge, and it performs the exchange. Dock never constructs an OAuth request, and a token minted by us would be refused at the exchange anyway. The credential is written to that agent's own volume, kept at owner-only permissions, stored exactly as the CLI wrote it, and refreshed by the CLI itself when it expires. A rebuilt machine still has the login it had before.
Always-on is the feature. A credential that is yours, on the agent's own disk, is the design.
Local and cloud, side by side
This is a choice you make per agent, not a migration. The agent that works your repo and drives your browser stays local, and their work still queues while your machine sleeps rather than being lost. The researcher on a standing watch goes cloud. Both kinds hold their own identity, write to the same shared workspaces, and leave the same attributed trail, so the team reads their work the same way regardless of where the run happened. If your Mac already has a Claude login, a cloud agent's machine can inherit it, which makes the second agent quicker to set up than the first.
Two honest edges to know about. Cloud agents come with a cap that depends on your plan, and hitting it now tells you so instead of silently giving you a local agent. And a cloud turn that nothing can pay for is refused and told to you as that, rather than starting and going quiet.
Where to start
Pick the one job you most wish happened before you woke up. Hire a cloud agent, give them that objective, and read the result in the workspace with your coffee. The deeper decision guide, including when local is still the better answer, is in cloud vs local agents, the security model is in are cloud AI agents safe, and the full setup detail is in the changelog.