finbarr/yolobox

What it solves

AI coding agents often require high levels of permission to be useful, but granting them direct access to a host machine can lead to accidental data loss (e.g., running rm -rf ~). yolobox provides a secure sandbox environment that allows AI agents to run commands with full permissions without risking the user's home directory.

How it works

The tool runs AI agents inside a container. It mounts the specific project directory to the container at its real path, but keeps the home directory isolated unless explicitly requested. It provides persistent volumes to maintain tools and configurations across sessions and includes built-in guidance for agents like Claude Code, Codex, and Kimi Code so they understand their sandboxed environment. To further enhance security, users can restrict network access, environment variables, and project directory permissions.

Who it’s for

Developers who use AI coding agents (such as Claude Code, Codex, Gemini, and Copilot) and want to allow these agents to execute commands and install software autonomously without risking their primary system files.

Highlights

• Isolated Execution: Protects the host home directory from destructive AI-generated commands.
• Pre-bundled Tooling: Includes Node.js, Python, Go, Bun, Git, and various build tools out of the box.
• Agent-Specific Shortcuts: Dedicated commands for launching popular AI CLIs (e.g., yolobox claude, yolobox codex).
• Project Customization: Supports .yolobox.toml files for project-specific environment variables and package installations.
• Flexible Security: Options to disable network access or mount the project directory as read-only.