a full computer in4-6ms

Composable computers for Devs and Agents

Full, hardware-isolated Linux computers you compose once, spin up as many as you need, and never pay for idle time.

Set up once. Build on it forever

  • create
  • fork
  • snapshot
  • checkpoint
  • pause / resume
  • hibernate

Machines that remember

Running processes, open files, and shell state all survive between sessions.

Your tools are already signed in

Integrations & automations in every machine

Persistent by default

Persistent, hardware-level isolated Linux machines for your team and your agents. They cost nothing while they sleep.

Development workspaces

Give every developer a persistent Linux box in the cloud. It keeps your editor, dotfiles, and running processes between sessions, so you SSH back into exactly where you left off.

Learn more

Preview environments

Every pull request gets the whole stack running on a machine of its own. It comes up with its services and its database already seeded, so a reviewer clicks a link instead of pulling the branch.

Learn more

Coding agents

Every agent gets its own Linux machine. It keeps its state while the agent works, so Claude Code and Codex find everything where they left it.

Learn more

Personal assistants

An always-on machine your assistant lives on, holding its tools, its memory, and a messaging gateway. It hibernates between conversations and wakes on the next message with everything still loaded.

Learn more

Sandboxes

Run code nobody read first behind a hardware boundary. It gets root inside a kernel of its own with no path to the host, and one flag closes the routes into the rest of your fleet too.

Learn more

Agentic SaaS

Give every user of your product a machine of their own, created and destroyed through the SDK. The CLI is installed inside every machine, so your software does it without a person in the loop.

Learn more

Reproducible RL envs

Every rollout forks from one prepared machine, so each episode starts from the same bytes with the simulator initialised and the caches warm. The seed controls your policy, the fork controls the environment.

Learn more

Agent swarm intelligence

Fan one problem out to many agents, each on a machine of its own, then synthesize a single answer. Machines boot in milliseconds, so twenty workers cost about as much thought as one.

Learn more

OSS integrations

Open source agent frameworks run their agents on boxd machines. Flue, Bindu, and Harbor each give a run a whole Linux computer through the SDK.

Learn more

Fork a machine, mid-run

Live memory forking of machines, in 100-200ms, allows you to experiment at the speed of thought. Live memory forks include:

  • running processes
  • databases
  • everything on the machine

Stop rebuilding the same environment

Install your application or workspace once, and use it as the starting point for all your work. Every machine you use has everything set up and running inside: data, apps, rules, skills, and more.

Built for flexibility and control

boxd hands you the whole platform to configure, not a black box. Add checkpoints to your VMs and take snapshots to restore from, and give your organisation its own private domains, variables and secrets.

Configure your favorite integrations inside your private VMs and automate whatever you want.

A computer per agent. Per branch. Per developer

A dev box, an agent, and a preview for a pull request are the same thing underneath. These are the tools you get for all of them.

Machines

Persistent Linux VMs with 2 vCPU, 8 GB of RAM and 100 GB of disk by default. SSH in and it's yours.

boxd m new

Forks

Copy a running machine, disk and all, in 100 to 200ms. The copy is a real machine, not a container on top of yours.

boxd machine fork alice

Snapshots

Freeze a machine and bring it back later in the state you left it, down to the running processes.

boxd snapshots save alice

Checkpoints

Roll a machine back to exactly where it was. Memory and disk, restored in place — a save point on the machine you already have, not a new one.

boxd machine checkpoint save alice

Integrations

Connect GitHub, Linear or Slack once and every machine in the org has it. The MCP server installs straight into Claude Code, Codex or opencode.

boxd manage integrations connect linear

Domains

Every machine gets an HTTPS subdomain. Point your own domain at it when you're ready.

boxd machine proxy add api --vm alice --port 3000

Environment vars

Share secrets across machines, scoped to your org rather than pasted per box.

boxd env set STRIPE_KEY --secret

API keys

Drive all of it from the CLI, the API, or your CI job.

boxd auth keys create

Self-hosted

Run the whole platform on your own hardware.

FAQ

Things people ask first

All questions →

A real KVM virtual machine with its own Linux kernel, not a container sharing the host's. Every machine is Ubuntu 24.04 with root, systemd, and the docker group, so Docker, Postgres, and anything else that needs real kernel features run the way they do on a laptop. It also means a kernel exploit inside one machine does not reach the host or its neighbours, which is what makes it safe to hand an agent root and walk away.

It keeps running. Builds, background jobs, and agent tasks carry on after the SSH session drops, and your files, installed packages, and shell state are still there when you come back. Machines that go quiet suspend to disk and resume in under a millisecond on the next connection, so you land back in the same processes instead of waiting for a boot.

Billing is credit-based: €0.049 per vCPU-hour and €0.015 per GiB-hour of RAM while a machine runs, and disk at €0.0001 per GiB-hour on what is actually used. A default machine (2 vCPU, 8 GiB) is about €0.22 per hour while it works, and a hibernated machine pays for disk only, so a fleet of agent VMs only runs up a bill while it is working. Every new account starts with €30 of free credits. For bigger machines or a higher machine quota, email contact@boxd.sh.

A founder. Every account starts with a 25-minute call where we provision your first machine with you, and after that you have a direct line to the engineers who built the platform. There is no ticket queue and nobody working from a first-line script. Custom plans add dedicated support.

Every idea gets its own machine

Install the CLI, sign in once, and the first one is yours. Fork it the moment you need another.

Your laptop
curl -fsSL https://boxd.sh/downloads/install.sh | sh
boxd auth login
boxd m new