microsoft/quicksand

What it solves

Quicksand provides a secure, isolated environment for running AI agents that need to execute code, browse the web, or interact with desktop applications. It solves the problem of safely running untrusted or experimental agent tasks by wrapping them in a controlled virtual machine that prevents them from accessing your host system.

How it works

It uses an asynchronous Python API to manage QEMU virtual machines. You can launch pre-built Linux environments (Ubuntu or Alpine) without needing root privileges or Docker. The library allows you to execute commands, mount host directories, manage network access, and even control a graphical desktop interface. You can also save the state of your virtual machine to disk or create checkpoints to easily roll back if an agent's actions cause issues.

Who it’s for

This tool is designed for developers building AI agents who need a reliable, cross-platform sandbox to test agent behavior, perform web scraping, or run code in a clean, reproducible environment.

Highlights

• Cross-platform support: Works on macOS, Linux, and Windows using x86_64 and ARM64 architectures.
• No root required: Runs without needing elevated system permissions or Docker.
• Flexible networking: Supports isolated environments or full internet access, including bidirectional port forwarding.
• State management: Features disk snapshots, persistent saves, and instant checkpoint/revert capabilities.
• Desktop automation: Includes desktop images with graphical environments, allowing for screenshots, mouse clicks, and keyboard input.
• Multi-user support: Allows running multiple agents as separate Linux users within a single virtual machine.