AI product Open source
Strands Box is an open-source sandbox engine for running AI agents. Written in Rust, it combines operating-system isolation with Dogwood semantic policies: direct file, program, and network access is configured in box.toml, while operations handled through Strands Shell, Monty for Python, the egress gateway, and the MCP broker are evaluated by an embedded Dogwood Local Engine. Operations are denied by default unless a permit rule allows them, and forbid rules override permits; rules can inspect arguments, earlier actions, and elapsed time, allowing activity in one interpreter or tool path to affect later requests. Box also supports credential injection for permitted requests, keeping configured API credentials or AWS SigV4 signing secrets outside the agent, and records policy decisions as OTLP JSON. The repository describes the current preview as local execution on macOS with Apple silicon, with Linux support planned, and distributes the project under the Apache License 2.0.
1 use taken from transcripts — each links to the moment in the video.
Strands Box is a sandbox for running AI agents with two layers of control: OS walls around files and programs, plus a policy language called Dogwood that decides every shell command, Python call, network request and MCP tool call, denying everything unless a rule permits it.
1 in the library.