Skip to main content
A boxd machine is a persistent, hardware-isolated Linux computer for developers, agents, apps, and automations. The recommended way to use boxd is the CLI. For orchestration and automation, use the TypeScript and Python SDKs. The quickest way to get started is the onboarding in the console at app.boxd.sh. If you prefer the terminal, follow the steps below.

1. Install

This script auto-detects your platform, verifies checksums, and installs the latest boxd CLI to ~/.local/bin/boxd.From an AI agent, or anywhere without a terminal to answer the installer’s questions, add --yes. Without it only the binary is installed; with it the installer also adds the agent skills, the PATH line and shell completions:

2. Authenticate

Sign in with GitHub or Google:

3. Create a machine

boxd new boots a fresh machine in milliseconds.

You’ll see:
This is your first machine. It has a persistent disk and its own HTTPS domain, with TLS already terminated. Visit the URL and you’ll see a default landing page until you start a server. The URL also says where the machine runs. The CLI puts it in the region closest to you, and the SDKs put it in Frankfurt unless you pass region. myapp.boxd.sh is in Frankfurt, and a machine in New York gets myapp.nyc.boxd.sh. See Regions to pick one.

4. Get into the machine

You get a shell on the machine straight away. Type exit or Ctrl-D to come back to the shell on your laptop.

5. Serve something on your public URL

Every machine already has a public HTTPS address at https://myapp.boxd.sh. It serves whatever listens on port 80, 443, 8080, 8000, 3000, 5000 or 5173, and falls back to port 8000. To see it working, start the nginx that comes pre-installed:
Now open https://myapp.boxd.sh in your browser. You are looking at a server running on your machine, reachable by anyone you send the link to. Swap nginx for your own app on one of those ports and the same URL serves that instead.

6. Fork your machine

A fork is a full copy of the source machine exactly as it is right now, with the disk, the memory, and every running process included. Disk and memory are copy-on-write, so the fork arrives in under 200ms. Fork the machine you just set up:
You now have a second machine, myapp-test, with its own URL at https://myapp-test.boxd.sh. Running processes come with it. The nginx you started in step 5 is already serving on the fork’s URL. Nothing had to boot or start up again. Open both URLs side by side and you will see the same page twice.

7. Snapshot your machine

A snapshot is a saved copy of a machine’s memory and disk, kept under its own name. Where a fork gives you a live copy of a machine that is still running, a snapshot is a starting point you create machines from. No machine has to keep running for that, so it costs less. It stays around even after you delete the source machine. Save myapp as a snapshot:
The machine keeps running while it saves. Now you can create machines from that saved state whenever you want:
Each one comes up in the saved state with its processes already running, so this is how you keep a known-good setup and hand copies of it to your organization or to an agent.

8. Make a checkpoint

A checkpoint is an undo button for one machine. Save one before you try something risky, and if it goes wrong, rewind the same machine back to that moment. The machine keeps its name and its URL. Save a checkpoint:
Then if a change goes wrong, roll the machine back:
The machine reboots into exactly the state it was in when you saved. Checkpoints live on the machine, so they go away when you delete it. Use a snapshot when you want something that outlives the machine.

9. Talk between machines

Machines can manage each other. The boxd CLI and both SDKs are already installed on every machine. Inside a private machine they act as the machine’s owner, fenced to that organization. Inside a shared machine they act for the organization’s shared machines only. There is no API key to copy in and nothing to configure. Get into myapp as you did in step 4, then run any of these from in there:
To control which machines are allowed to reach which, put them on networks. See Machine to machine.

10. Share a machine

Machines are private to you by default. Sharing one opens it to every member of your org, so a colleague can connect to it and work in the same machine.
Only the owner can share or unshare a machine, and sharing never moves it between organizations. It only changes who can reach it.
Sharing wipes the agent logins inside the machine, so a colleague with a shell can never read your Claude Code, Codex, or OpenCode tokens. Do not share a machine an agent is working in, because it gets logged out immediately. If you want a copy with your logins intact, fork the machine instead. A fork of a shared machine is private to you.
See Machine sharing.

Next steps

Dive deeper into the guides, or explore the use cases.

Development workspaces

A real computer for you and your team.

Preview environments

A URL for every pull request.

Coding agents

Give an agent its own machine to work in.

Personal assistants

An always-on machine your assistant lives on.

Sandboxes

Run untrusted code behind a hardware boundary.

Agentic SaaS offerings

Give every one of your users their own machine.