finbarr/yolobox

What it solves

AI coding agents often require unrestricted command execution to be effective, but granting them full access to a local machine poses a significant security risk, such as the accidental deletion of the home directory. yolobox provides a secure sandbox environment where agents can operate with high privileges without risking the host system's critical files.

How it works

The tool runs AI agents (like Claude Code, Codex, or Gemini) inside a container. It mounts the specific project directory to the agent's environment while keeping the home directory isolated unless explicitly opted in. It uses persistent volumes to maintain tools and configurations across sessions and provides built-in guidance to help agents understand their sandboxed state. 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 and want to allow them to run commands autonomously ("yolo mode") without fearing destructive accidents on their primary workstation.

Highlights

• Multi-Agent Support: Built-in shortcuts for various AI CLIs including Claude, Codex, Kimi Code, and Copilot.
• Project-Specific Customization: Ability to add extra tools, environment variables, or Dockerfile fragments via a .yolobox.toml config file.
• Pre-configured Environment: The base image includes essential development tools like Node.js, Python, Go, Bun, Git, and ripgrep.
• Granular Security Controls: Options to run agents with no network access, read-only project mounts, or excluded secret files.
• Persistent State: Named volumes ensure that tools installed by the agent via sudo are preserved across runs.