BAND and Docker Integrate Docker Sandboxes with BAND Rooms

BAND and Docker described an integration that connects BAND’s collaboration layer with Docker Sandboxes, so developers can run coding agents inside isolated microVM environments and connect those agents to a broader BAND room.

The companies framed the work around enabling agents to collaborate and eventually execute tools, including scenarios where multiple agents do not run in the same environment. The integration also included controls for how agents can access networks and manage credentials.

BAND provided collaboration infrastructure including persistent agent identities, shared rooms, @mention-based routing, message history, and tool execution events. Docker Sandboxes were described as fully isolated microVM environments with their own kernel, filesystem, Docker daemon, and network.

The integration connected the systems through an outbound WebSocket connection in which a remote agent established connectivity to BAND and behaved like a participant in a BAND room. The sandbox included a security boundary with deny-by-default network access, credential handling via a host-side proxy, and the ability for developers to inspect network requests. “Agents shouldn’t have to live in the same environment to work together,” said Arick Goomanovsky, CEO and co-founder of BAND. “Sooner than many think, we will see a reality where some agents are hosted, some run locally, and each has access to the tools and data it actually needs. BAND provides the collaboration layer to enable this, while Docker Sandboxes provides an isolated execution environment. Together, they make it much easier to compose these agents into efficiently working teams.” “Docker Sandboxes give agents a place to run freely without giving up security or control: an isolated, managed environment with clear boundaries,” said Eli Aleyner, VP of Strategy at Docker, Inc. “BAND's Sandbox kit, enables multiple agents to collaborate with each other under those same principles, regardless of where each one is deployed. Developers can build teams of agents the way they already build applications, with runtime isolation, governance, and portability.”