docker/sbx-releases

Docker Sandboxes is a tool for running AI coding agents in isolated environments on your machine or Docker-managed cloud infrastructure.

The tool addresses the security risk of letting AI agents operate autonomously by providing sandboxed execution with controlled access to your filesystem, network, and tools. It uses Docker's containerization technology to isolate agent workloads, allowing them to work without permission prompts while preventing access to resources outside the sandbox boundary. File access, network connectivity, and tool availability are all configurable per sandbox.

Docker Sandboxes suits developers who want to run AI coding agents locally or in the cloud without compromising machine security or data privacy. It works vendor-neutrally with multiple AI models and tools including Claude Code, Codex, Gemini CLI, and OpenCode. The tool is available on macOS, Windows, Ubuntu, and Rocky Linux through package managers or manual installation from release artifacts.

The project maintains three release channels: stable releases distributed through Homebrew, WinGet, and package repositories; release candidates published as prerelease tags that represent builds promoted to stable; and nightly builds from the main branch. Issue reports and feedback are actively solicited through the repository's issue tracker.