Docker Introduces Cloud Sandboxes for AI Agent Execution and Workflow Portability

Introducing Docker Cloud Sandboxes

Docker has released Cloud Sandboxes, which are secure, hosted execution environments designed for running AI coding agents on Docker-managed infrastructure. These new sandboxes build upon the existing Docker Sandboxes, which previously offered local microVM environments for autonomous agent operations.

Addressing Scalability and Persistence

The introduction of Cloud Sandboxes responds to the increasing demand for persistent and scalable execution environments beyond local machines. Developers are now running multiple long-horizon tasks in parallel, requiring infrastructure that can support tasks lasting hours without interruption, unlike a laptop's limitations.

Cloud Sandboxes provide a consistent execution environment and unified CLI workflows, allowing developers to seamlessly move workloads between local and cloud. This enables tasks to run for extended periods or in parallel across numerous isolated cloud sandboxes.

Workflow Portability

A key feature is the ability to move a sandbox between local and cloud execution using a single command, such as "$ sbx move my-project --to cloud". This command captures the sandbox's filesystem and recreates it in the new environment, ensuring work continuity. This allows developers to start tasks locally and transfer them to the cloud for more resources or to continue running unattended.

Updated Kits Specification

Alongside Cloud Sandboxes, Docker has updated its Kits, which are pre-configured sandboxes. In the latest Kits v3 specification, Kits are no longer separate artifacts but are packaged as standard OCI images. This change allows them to be used like any other Docker image, supporting standard build and pull operations, and serving as a base for creating more complex Kits.