Claude and Codex can already run tasks in the cloud. Why this?
They can, and for a repo-scoped task it is genuinely good: hand off a branch,
close your laptop, review the diff later. The difference is that those sessions
are sandboxes, not machines. Each one starts from a fresh checkout, and what you
install is gone next session. A database or dev server it started does not
survive. Nothing gets a public URL. There is no shell.
| Their cloud | Your box |
|---|
| Task survives closing your laptop | yes | yes |
| Files and installs still there tomorrow | no | yes |
| Database or service that stays up | no | yes |
| Public URL for what it built | no | yes |
| A shell you can open | no | yes |
| Work outside a git repo | no | yes |
Use both. Their cloud is good at what it does; your box is the machine where
your project actually lives.
Why not just rent a VPS and install this myself?
You can, and it costs nothing: the self-hosted edition is the same code under AGPL, one install script. The $19 buys the part you would otherwise spend a weekend on and redo on every rebuild: a box provisioned, the agents signed in, a real address with TLS wired up, and updates that land without you. If running servers is the fun part for you, self-host, and we would still count that as a win.
Does my Claude Code login work on a server?
Yes. It is an ordinary Linux machine, so you run the same sign-in you run locally, once, from the browser terminal, and the session stays on your box. No API key, no separate billing. Antigravity signs in with your Google account the same way.
What do I need to bring?
One AI subscription: a Claude (Anthropic) or ChatGPT (OpenAI) plan, from $20/mo. That plan drives your agents directly; we never resell AI or add a markup on it.
Can I move my current project over?
It is a real machine, so it is a git clone. Your agents read the repo, run it on the box and carry on from there. Bring your dotfiles and your project's environment file too; they work the way they do on your laptop.
What exactly does $19/mo buy?
A dedicated cloud computer (not a shared sandbox) with ShellTeam pre-installed and kept up to date, a real HTTPS address of your own, and hosted voice input. The AI itself runs on your own plan, which is why there is no meter.
Is my stuff safe?
Your box is yours alone: a dedicated machine with no other users. Your agents have full control of it (that's what makes them capable) and of nothing else. Everything stays on your box; nothing phones home. The whole security model is public, and because the code is open source, any engineer you trust can verify it.
Can other people use my box?
It is yours alone by default. You can hand out a link to something your agents built, and that link works for whoever you send it to. Nothing else on the machine is reachable unless you publish it deliberately.
