Docker released Cloud Sandboxes, moving an agent's microVM between your laptop and Docker's compute with one sbx move command

A coding agent that runs for six hours has nowhere sensible to live on a laptop that sleeps when the lid closes. Docker released Cloud Sandboxes on 24 September 2026, extending its local microVM sandbox onto Docker-managed compute, so an agent can start on a developer's machine, close the laptop, and keep going in the cloud with the same command line and the same isolation model.

Docker product managers Timir Karia and Srini Sekaran wrote the launch post as answering a second question about coding agents. The first, "is it safe to run unattended", was answered by the local Sandboxes released earlier in 2026: an agent gets a microVM with its own kernel and Docker daemon, isolated from the host and from its files, network and secrets. The new question is "how do I run a dozen agents at once, for five, ten, or 21 hours each, without watching any of them". Cloud Sandboxes is the same microVM, running on always-on machines.

The one command that matters

A sandbox moves between laptop and cloud with sbx move my-project --to cloud. Docker says the move "captures the sandbox's filesystem and recreates it on the other side" and that it works in both directions. The web console offers the same actions, and Docker ships pre-built "kits" for the leading coding agents, including Claude Code, Codex, Copilot, Antigravity, Open Code and Hermes. Kits v3 packages a kit as an ordinary OCI image, so a team can use docker build and docker pull on it and can build derived kits on top.

Pricing, and the free credit

Compute is metered by the second and a paused sandbox costs nothing. Docker lists five sizes at fixed hourly rates: Micro at $0.07, Small at $0.14, Medium at $0.28, Large at $0.56 and XL at $1.12. A session runs for one hour by default and up to 24 hours. Volumes, egress, and hosting public images and Kits are free, and new accounts get $250 in Cloud Sandboxes credit for a limited time. The plan is on Docker Personal and Pro; you keep your own model provider through your own API key.

The isolation is only part of it

On the InfoQ story, one Reddit reader argued the isolation only addresses part of the risk. "Any remotely useful agent workload is going to need to connect their sandboxed agents to limited external services", the commenter wrote, and those services (PyPI, Docker Hub, Hugging Face) are places the agent can call from inside the sandbox and be tricked by. Docker's answer is three controls it ships alongside Sandboxes: an MCP gateway that fronts Jira, Linear or Grafana connections once, a proxy that injects secrets per request so the agent never sees the key, and network policies that name the endpoints an agent may reach.

A team that already runs local coding agents will notice what does not change: the CLI, the microVM boundary, the kit format, the model provider. What changes is where the compute runs, and how many agents can run at once. "How much you can trust your agent shouldn't depend on where it happens to be running", the launch post argues, so local and cloud are one product with a --cloud flag. The pattern Docker is designing for is a developer who iterates on the code with an agent in front of them and hands off the long jobs before closing the laptop.

Source

Primary source: Introducing Cloud Sandboxes on the Docker blog.

Reporting: Docker Cloud Sandboxes Provide a Consistent Sandbox Abstraction Across Laptop and Cloud, by Sergio De Simone at InfoQ.