Leash is a tool designed to manage AI coding agents by wrapping them in containers, allowing for strict policy enforcement through Cedar. It supports Docker, Podman, and OrbStack, and runs on macOS and Linux, including WSL. You can install Leash using npm or by downloading a pre-built binary. For macOS users, there's a specific installation method via Homebrew that also enables experimental features. After installation, you can launch coding agents like Claude and Codex, and even automate the opening of their interfaces in a web browser. The tool captures detailed telemetry by monitoring filesystem access and network connections made by the agents, ensuring that Cedar policies are applied effectively. When using Leash for the first time, it prompts you to mount your agent's configuration directory, allowing you to choose whether to remember this decision. The telemetry includes system call monitoring and exposes a Control UI at http://localhost:18080. Environment variables for API keys are automatically mapped for different agents, which simplifies the setup process. You can customize the Docker image used by Leash by editing configuration files or using command-line flags. The default image is stored on AWS ECR, but you can extend it by adding project-specific packages or reusing existing images. The configuration options allow for detailed control over how the agents operate, including setting bind mounts and forwarding environment variables. For those looking to dive deeper, there are specific markdown files provided for additional configuration, telemetry details, and development insights.