Docker Cloud Sandboxes Run Coding Agents on Managed Compute
Docker's new Cloud Sandboxes run the same microVM isolation as its laptop sandboxes on Docker-managed compute, movable with one command and metered per second.
Docker introduced Cloud Sandboxes, which the company describes as the same microVM-based sandbox as Docker Sandboxes but running on Docker-managed compute, with a single command to move a sandbox between a laptop and the cloud. The announcement was published on the Docker blog on September 24, 2026.
The pitch is aimed at developers who hand long-running work to coding agents. According to Docker, agents have started taking on tasks that run for hours rather than minutes, while a laptop sleeps when the lid closes, slows on battery and disconnects when you move. Cloud Sandboxes are meant to answer how to run a dozen agents at once for five, ten or 21 hours each without watching any of them.
Docker says Cloud Sandboxes use the same isolation model and the same CLI as Docker Sandboxes, which the company launched earlier in the year as microVM environments where coding agents can work autonomously. Each Cloud Sandbox gets its own microVM with its own kernel and Docker daemon, its own secrets and its own network policy, according to the company.
One command, both directions
Docker says the command sbx move my-project –to cloud moves a sandbox by capturing its filesystem and recreating it on the other side, and that this works in both directions. Sandboxes can be spun up and managed from the CLI or from the web console, and Docker says sbx –cloud run codex starts a Cloud Sandbox.
Terminal setup, per Docker, uses brew install docker/tap/sbx, sbx login and sbx –cloud run claude. In the browser, the company says users sign in to the web console, choose a kit and click Run. Docker Sandboxes on the laptop remain free and standalone, with no Docker Desktop required, and local and cloud sandboxes keep separate secrets, templates and network policies.
Kits, MCP gateway and proxied secrets
Docker says kits are pre-configured, pre-built sandboxes for an agent, and that Claude Code, Codex, Copilot, Antigravity, Open Code and Hermes are ready today. MCP servers such as Jira, Linear, Grafana, incident.io or any streamable HTTP endpoint can be connected once and reached by all agents through a single gateway, whether running in the cloud, locally, or in other clients like the ChatGPT desktop app, according to Docker.
The company also says the Cloud Sandboxes proxy injects stored keys or tokens per request so agents do not see the actual secret, and that network policies on which endpoints agents can access are defined once. Centralized governance is coming soon for enterprises through Docker AI Governance, Docker says. Users can bring their own model key and keep their existing provider for inference.
Per-second pricing and five sizes
Docker lists five sizes with hourly prices: Micro at 1 vCPU and 2 GiB for $0.07; Small, the default, at 2 vCPU and 4 GiB for $0.14; Medium at 4 vCPU and 8 GiB for $0.28; Large at 8 vCPU and 16 GiB for $0.56; and XL at 16 vCPU and 32 GiB for $1.12. Sandboxes run for one hour by default and up to 24 hours per session.
Docker says Cloud Sandboxes are pay-as-you-go, metered by the second for compute and nothing else, with paused sandboxes costing nothing and volumes, egress, and hosting public images and kits being free. The company says sbx 0.45.1 or later and the pay-as-you-go plan are required, available on Docker Personal and Pro accounts, and that new accounts get $250 in free Cloud Sandboxes credit for a limited time.
Run agents on your laptop, in the cloud, and move between them with one command, all safely.
That line comes from Docker’s Timir Karia, senior principal product manager, and Srini Sekaran, principal product marketing manager for AI, who also wrote: “These aren’t two tiers of one product. Interactive work belongs on your laptop. Work that takes hours belongs in the cloud.”
Docker says it is bringing the open source Sandbox Kit Spec to the CNCF, per a related post listed on the page.
What is not yet known
Docker has not said which cloud providers or regions host the managed compute, nor whether the launch is generally available or limited. There is no stated expiration date or eligibility restriction for the $250 credit, no web console URL, and no release date for Docker AI Governance. It is also unclear whether pricing varies by region or currency, what the pay-as-you-go plan costs beyond per-second compute metering, and whether Cloud Sandboxes support non-coding agents.
Related reading
Sources
Some links on this page may be affiliate links. If you buy through them we may earn a commission at no extra cost to you. See our affiliate disclosure.
