Thirty seconds to set up. Then you stop thinking about it.

No CLI, no config file, no key to look after. Your AI app signs in the way every other app you use does.

STEP 1

Get a computer

One click builds it: real Ubuntu, root, 20GB of disk. It belongs to your project rather than to a conversation.

STEP 2

Paste one URL, approve one screen

Give your MCP URL to Claude, Cursor or VS Code. It opens a browser, you approve, and it is connected — no secret ever touches a config file.

STEP 3

Let it move in

Your AI installs its tools, writes files and runs your app on real hardware — and every change it makes to the machine sticks.

STEP 4

Come back whenever

Close it, switch apps, take a fortnight off. Reconnect and the computer wakes exactly as it was — nothing to reinstall, nothing to re-explain.

What happens when it sleeps

This is the part that makes a computer cheap to keep and fast to sit back down at.

After 20 minutes without activity, your computer archives its entire filesystem and copies it to storage. The running container is then destroyed — you are not paying to keep an idle box warm, and nothing about your project is lost.

When you or your AI touch it again, the archive is restored onto a compute node and unpacked before the machine reports itself ready. Your files, your installed packages, your database contents and your environment variables are all exactly as they were. The same connection keeps working, so nothing needs reconfiguring.

It is worth being clear about what that buys, because it is easy to read as a billing trick. The archive is the whole disk, so everything your AI ever did to this machine is still done. The tool it installed in March is installed. The dependency tree it spent four minutes resolving is resolved. The note it left itself is where it left it. That is not a cache — it is the accumulated result of every session, and it is the closest thing to long-term memory an assistant currently has.

You can also put a computer to sleep deliberately, from the dashboard or by asking your AI to do it.

One computer runs at a time. Extra machines exist so you can keep separate projects around and switch between them — waking one automatically sleeps the other, archiving it first. Nothing is ever discarded to make room.

Where your work actually lives

Running computers sit on fast local disk. Sleeping ones sit on cheap bulk storage. Neither is shared with anyone else.

Every computer is sealed off from the hardware it runs on and from every other customer's. Archives are written to storage that only that one machine can reach.

Your AI connects directly to the hardware running your computer rather than through a central gateway, which is why commands feel local rather than proxied.

Ready when you are.

Build a computer and sign your first AI app into it in about thirty seconds.

Get started