Agent Sandboxes

A Linux workspace for your next command.

Create an isolated Firecracker microVM. Run commands, inspect results and open a terminal from the rzv CLI, API or portal. Delete it when the work is done.

Isolated LinuxA Firecracker microVM per sandbox
CLI · API · PortalCommands and interactive terminals
Up to 8 hoursFrom first start, running or stopped

The lifecycle

Create. Run. Inspect. Clean up.

A short illustration of the command sequence, followed by copyable examples.

Illustrative animation, not a recording of a live sandbox. Commands use a placeholder sandbox ID. Follow the steps below with the ID returned in your workspace.
  1. 01

    Create

    Create and start a Linux microVM in your workspace. Wait until its status is running.

    Terminal · command example
    rzv sandboxes create --name code-check --size 2vcpu-8gb
    rzv sandboxes list
  2. 02

    Run commands

    Stream stdout and stderr. The CLI exits with the command’s status. Commands run in /workspace by default.

    Terminal · command example
    rzv sandboxes exec <sandbox-id> -- "printf 'hello sandbox\n' > result.txt"
  3. 03

    Inspect results

    Read the result or open an interactive terminal to investigate. Use exec on Windows and in scripts.

    Terminal · command example
    rzv sandboxes exec <sandbox-id> -- cat result.txt
    rzv sandboxes ssh <sandbox-id>
  4. 04

    Stop or delete

    Stop to keep files within the lifetime, without a running charge. Delete when finished to remove the VM and its files.

    Terminal · command example
    rzv sandboxes stop <sandbox-id>
    rzv sandboxes delete <sandbox-id>

Sign in and select a workspace first. Owners and admins can create and change sandboxes. Availability, sizes and the running limit come from your environment. Install and sign in with the CLI →

What to do with a sandbox

From a single command to an investigation.

Give an agent a place to run commands

Call exec from your own agent workflow, read output and check the exit status before deciding what to do next. Your integration controls the commands and credentials.

Read the exec API →

Validate a code change

Put the code and tools your checks need in the sandbox, then run Linux CPU checks. Inspect failures before carrying the change back to your repository.

Start with the CLI →

Investigate interactively

Open a shell, inspect files and rerun a command in the same workspace. Stop and start within the eight-hour lifetime to return to those files.

Explore the terminal →

Size and lifetime

Choose the space. Know the boundary.

Choose 2 vCPU / 8 GB, 4 vCPU / 16 GB, 8 vCPU / 32 GB or 16 vCPU / 64 GB. Available sizes depend on current host capacity; the list response shows what you can create now. Workspace disks range from 5 to 64 GiB, with 20 GiB by default.

Create any offered size with the rzv CLI, API or portal. Run rzv sandboxes list to check current availability.

Files in /workspace survive a stop and a start. Eight hours after the first start, the sandbox expires and its VM and files are deleted. Stopping does not pause the clock, and starting does not reset it.

All sizes and limits →

Billing

Pay for the time it runs.

Running sandboxes use your workspace allowance or prepaid credit at the runner rate for their size. 2 vCPU counts 1×; 4 vCPU counts 2×; 8 vCPU counts 4×; 16 vCPU counts 8×. A stopped sandbox is not billed.

Each start reserves 15 minutes ahead. When a run ends, only the seconds it ran are charged. Without credit, create and start are refused.

See unit-minute pricing

Take the next step.

Ask Docs

Ask a question and open it in your AI assistant. The prompt tells it to answer from the Runzivo docs and link the pages it used.

Try asking

Your question opens in the assistant you pick, under its own terms. Runzivo does not see or store it.