jamesstringer90/appsandbox

Easily create full virtual machines that are sandboxed for development or computer use models.

What it solves

App Sandbox provides a high-performance, easy-to-use virtual machine (VM) manager for Windows and macOS. It allows users to run isolated desktop environments (Windows 11, Ubuntu, and macOS) with full GPU acceleration, which is typically difficult to achieve in standard VM software. It specifically addresses the need for a clean, isolated environment for software testing, driver development, and running AI agents that require a graphical interface.

How it works

The project uses native system APIs to create VMs rather than traditional hypervisors like Hyper-V Manager. On Windows, it leverages the Windows Host Compute System (HCS) and Host Compute Network (HCN) APIs (the same foundation as WSL2), utilizing GPU Paravirtualization (GPU-PV) to share the host GPU with the guest OS. On macOS, it uses Apple's Virtualization.framework. Both platforms include a headless daemon and a Python SDK (asb.py) that allows users to programmatically manage the VM lifecycle—including provisioning, snapshotting, and branching—via a local HTTP/JSON API.

Who it’s for

  • AI Developers: Those running "computer-use" AI agents (like Claude's computer use model) that need an isolated, GPU-accelerated desktop to interact with.
  • Software Testers: Developers who need clean, reproducible Windows or Ubuntu installs for testing software or drivers (including a built-in Test Mode for driver signing).
  • Security Researchers: Users who want to run potentially untrusted programs in a separate environment from their main OS.
  • Windows Home Users: Windows 11 Home users who can run VMs without needing a Pro license for Hyper-V.

Highlights

  • GPU Acceleration: Supports DirectX 12, OpenGL, Vulkan, CUDA, and OpenCL on Windows; Metal on macOS.
  • Headless API & SDK: A dependency-free Python SDK for automating VM creation, snapshots, and management.
  • SSH Access: Built-in SSH support via Hyper-V sockets or virtio-vsock, requiring no network configuration.
  • Broad OS Support: Supports Windows 11, Ubuntu 26.04 LTS, and macOS guests.
  • Nested Virtualization: Enables running Docker or other VMs inside the guest VM.

Related

  • Project
    dondai44423/donsetch

    A high-performance web research tool for AI agents that bypasses bot detection using native Chrome TLS and provides keyless search and token-optimized content extraction.

  • Project
    agarwalvishal/claude-chat-exporter

    A privacy-first browser tool that exports Claude.ai conversations to high-fidelity Markdown, optimized for Obsidian and RAG pipelines.

  • Project
    jdtoscano94/NABLA-SciML

    A unified framework for Scientific Machine Learning (SciML) providing implementations of PINNs, DeepONets, and KANs for studying complex physical systems.

  • Project
    m1ckc3s/claude-status-bar

    A macOS menu bar app that provides live status indicators for Claude Code, showing whether the AI is working, thinking, or awaiting user permission.