Cube vs Docker Cloud Sandboxes
Docker Cloud Sandboxes move an agent’s sandbox from your laptop to the cloud with one command. Cube gives your agents a whole computer to live on.
Published
The short version
Docker Cloud Sandboxes, launched in September 2026, run an agent in an isolated microVM in Docker’s cloud. You can start a sandbox on your laptop and move it to the cloud with the sbx CLI. They’re metered per second and suit isolated, task-sized runs.
Cube is a persistent Linux computer in the cloud for your coding agents, with a Mac app, your GitHub repos, a worktree per task and phone check-ins.
Side by side
| Docker Cloud Sandboxes | Cube | |
|---|---|---|
| Unit | A sandbox per task (microVM) | A computer for all your agents |
| Session length | Up to 24 hours (1 hour default) | Always on |
| Billing | Per second, $0.07–$1.12/hour by size | Cube plan |
| Interface | sbx CLI | Mac app, phone, and SSH |
| Your repos | You bring them into each sandbox | Synced from GitHub, a worktree per task |
| Requires | Docker account with pay-as-you-go | A Mac |
When to choose Docker Cloud Sandboxes
- You already use Docker Sandboxes on your laptop and like the workflow.
- You want each agent run in a fresh, isolated environment you throw away.
- Your usage is bursty and per-second billing fits it.
When to choose Cube
- You want one place where all your repos, tools and agent sessions live.
- Your agents run most of the day, and you don’t want a meter on each run.
- You want a desktop app and a sidebar of every session, not only a CLI.
Questions
How long can a Docker Cloud Sandbox run?
As of September 2026, a session runs for up to 24 hours (one hour by default). Paused sandboxes keep their state and cost nothing.
How are Docker Cloud Sandboxes billed?
Pay as you go, per second, by size: from $0.07 an hour for 1 vCPU and 2 GiB up to $1.12 an hour for 16 vCPU and 32 GiB. New accounts get $250 of credit.
When should I choose Docker Cloud Sandboxes?
When you already use Docker Sandboxes locally and want the same isolation and CLI in the cloud, or you want strictly isolated, per-task environments.