madarco/agentbox

What it solves

AgentBox provides a sandboxed environment for running multiple AI agents (such as Claude Code, Codex, and Open Code) in parallel. It eliminates the need to manually set up separate environments for each agent, allowing users to quickly deploy agents into dedicated virtual machines (VMs) or Docker containers, either locally or in the cloud, while maintaining persistent shells and IDE integration.

How it works

AgentBox uses a CLI to "teleport" a project into a dedicated VM or container. It creates a sandboxed "box" for each agent, which includes a dedicated browser, screen sharing via webVNC, and persistent shells. It supports various cloud providers (Hetzner, Vercel, E2B, DigitalOcean, Daytona) and local Docker setups. The system uses FUSE overlays for state management and allows users to create checkpoints to start new boxes from a warm state in under a second.

Who it’s for

Developers who use AI coding agents and want to isolated, scalable, and reproducible environments for their projects without compromising their local machine's security or configuration.

Highlights

• Multi-Agent Parallelism: Run multiple agents in parallel with a single command.
• Cloud & Local Flexibility: Deploy to local Docker, remote Docker, or various cloud providers.
• Cloud-Native Tooling: Integrated support for VS Code/Cursor IDE, web apps via URL tunnels, and VNC screen sharing.
• Fast State Management: Sub-second startup from checkpoints and auto-pause functionality to save resources.
• Safe Credentials: Keeps git credentials on the local machine, requesting permission to push to remote repositories.