Run AI agents safely in local sandboxes. Docker Sandboxes provides disposable, isolated micro VMs for agents like Claude Code, Gemini CLI, Copilot CLI, Codex, Kiro, and Open Code. Installation is quick: on macOS `brew trust docker/tap && brew install docker/tap/sbx` or on Windows `winget install Docker.sbx`.

Each agent runs in its own micro VM with your dev environment and project workspace mounted, allowing package installation, config changes, and even Docker container launches while keeping the host untouched. The default mode `--dangerously-skip-permissions` (YOLO) gives agents autonomy without prompts, but the isolation wall makes it safe. Organizations can enforce policies via Docker AI Governance, adding network restrictions, filesystem controls, and MCP governance.

Nano Claw’s creator Gavriel Cohen notes that Sandboxes let agents perform long‑running tasks safely, and Warp’s engineering lead Ben Navetta highlights integration for consistent local or cloud agent execution. Common questions address sandbox definition, supported agents, safety of YOLO mode, differences from traditional VMs, and configurable controls. For advanced admin needs, Docker AI Governance offers centralized network policies, filesystem rules, and team‑wide MCP governance.

Talk to an expert to tailor network access, filesystem restrictions, and admin configurations for your team.