Agent Sandboxes are available in every workspace. A sandbox is an isolated Linux machine for a coding agent, or for you, to work in for a session. Install rzv 0.20.0 and run rzv sandboxes create to start one.
What a sandbox is
Each sandbox is a Firecracker microVM with its own disk. Files in /workspace survive a stop and a start. A sandbox lasts eight hours from its first start. Stopping it does not pause that time, and when it ends the machine and its files are deleted.
How to use it
Start with the rzv CLI. It creates a sandbox, runs commands with streamed output and opens a terminal with rzv sandboxes ssh:
rzv sandboxes create --name agent-1
rzv sandboxes exec <sandbox-id> -- npm test
rzv sandboxes ssh <sandbox-id>
rzv sandboxes delete <sandbox-id>Agents can call the customer API directly: create a sandbox, run a command and read its output as a stream, or open the terminal as a WebSocket. Runzivo does not log the commands or their output. Start with the quickstart.
In the portal, open Build & Deploy, then Sandboxes, and select New Sandbox. The sandbox page has a terminal while it runs.
Billing
A running sandbox uses unit-minutes from your allowance or prepaid credit, at the same rate as a runner of its size. A stopped sandbox is not billed. See limits and billing for how runs are charged and what happens when credit runs out.
Limits
A workspace can run two sandboxes at once by default, and the lifetime cannot be extended. The Agent Sandboxes guide links the sizes, limits and API reference.