Run your agents in parallel, each on its own machine.
Claude Code, Codex and OpenCode come installed and set up for boxd. They keep working after you close your laptop.
Connect your accounts
Sign in to Claude Code once and every new machine has you logged in. Connect GitHub the same way. Codex and OpenCode log in once per machine.
Your agents, on every device
Run Claude Code in tmux on the machine with remote control on. Close your laptop and pick the same session up in the Claude app, on your desktop or your phone.
Build your agent factory
Snapshot your repo once, then start a machine per task, each with its own agent. The primitives are yours to build on.
Also in every machine
Agents that know boxd
Built-in instructions let agents create, use and clean up machines themselves.
Secrets
API keys sealed into the machine as env vars.
Checkpoints
Roll the machine back to before the agent broke it.
Isolated machines
No integrations, no agent logins, no reach to your other machines.
Egress allowlist
The agent only reaches hosts you name.
Hibernate
An idle agent pays for disk only.
API and SDKs
Start and fork machines from Python or TypeScript.
Self-host
Run boxd in your own cloud or datacenter.