🛡️ Security by Compartmentalization for AI Coding Agents.

Agent VM

This project provides a secure and compartmentalized environment for running AI agents. It utilizes microvm.nix to create a lightweight NixOS-based virtual machine. We would like to give a huge thanks to the microvm.nix project for their great work.

The primary goal is to enable developers to safely run AI agents without compromising the security of their host system. The agent operates within a MicroVM and only has access to its own home directory, which is created alongside the machine. Project directories from the host can be selectively mounted into this home directory, making them accessible to the agent in a controlled manner.

[!TIP] There is nothing special about this VM to make it particularly limited to AI agent use-cases. These machines can be used for compartmentalizing other applications as well.

[!WARNING] I'm not a security expert, and this is designed around my understanding of potential risks of running AI agents. Use it with caution.

Table of Contents

• Features
• Getting Started
  • Prerequisites
  • Quick Start
  • Building the VM
  • Running the VM
• Usage
  • Mounting Host Directories
  • Gemini Authentication
• Technical Details
• Documentation
• Troubleshooting
• Future Work
• Contributing

Features

• Secure Environment: Run AI agents in a lightweight, isolated MicroVM.
• Controlled Access: Selectively mount host directories into the agent's home directory.
• NixOS-based: Leverage the power and reproducibility of NixOS.
• Easy to Use: Get started quickly with a single command.

Getting Started

These instructions will guide you through building and running the MicroVM on your lo