RunBerth Documentation
RunBerth gives your agents every computer they need. One account and one API key start sandboxes, Linux and Windows desktops and Android phones on demand, billed per second on one invoice. Drive all of it from the dashboard, the CLI, the API, the SDK or MCP.
- Agent boxes
- Sandboxes and swarms
- Computers on demand
Which one do I need?
| You need | Use |
|---|---|
| Run code a model wrote and throw the machine away | Sandbox |
| A workspace per user that keeps its files and tools | Agent box |
| The same job on many machines at once (evals, rollouts, batches) | Swarm |
| A browser or a Linux app with a screen, driven by your model | Linux desktop |
| Software that only runs on Windows | Windows desktop |
| A mobile app with no API | Android phone |
| A long-running agent with memory, channels and schedules | Agent (in its own Agent box) |
The machines
Agent boxes
Sandboxes
Computers on demand
Linux desktop
A desktop with a screen, mouse, keyboard and browser that your model drives over MCP or REST.
$0.12 an hour awake
Windows desktop
The same computer on Windows Server 2022, for software that only runs on Windows.
$0.18 an hour awake
Android phone
An Android phone your agent taps and swipes, with adb for your own tooling.
$0.15 an hour awake
Around every machine
- Agents and templates
Run OpenClaw, Hermes, Claude Code or your own container in an agent box.
- Add-ons
Static and residential IPs, private networks, warm pools and daily backups.
- End users
One machine per user of your product, keyed by your own id.
- Activity
Every start, sleep, wake and error across your account.
- One bill
Per-second usage from the plan credit, one invoice a month.
References
Driving RunBerth from a coding agent?
Copy the instructions into Claude Code, Cursor or Codex and it sets up RunBerth for you. The machine-readable contract is runberth guide --json; the full index is /llms.txt.