AI沙盒 是 AI Skill Hub 本期精选Agent工作流之一。综合评分 8.0 分,整体质量较高。我们强烈推荐将其纳入你的 AI 工具库,帮助提升工作效率。
AI沙盒 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
AI沙盒 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 克隆仓库 git clone https://github.com/jamesstringer90/appsandbox cd appsandbox # 查看安装说明 cat README.md # 按 README 完成环境依赖安装后即可使用
# 查看帮助 appsandbox --help # 基本运行 appsandbox [options] <input> # 详细使用说明请查阅文档 # https://github.com/jamesstringer90/appsandbox
# appsandbox 配置说明 # 查看配置选项 appsandbox --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量(覆盖配置文件) export APPSANDBOX_CONFIG="/path/to/config.yml"
<img width="1028" height="446" alt="App Sandbox" src="https://github.com/user-attachments/assets/f66dcaff-201e-4cf5-8e23-bc480338f353" />
App Sandbox creates and runs full desktop virtual machines — Windows 11, Ubuntu, and macOS. It is free and open-source (MIT), distributed as prebuilt binaries (EV-signed on Windows, Apple-Developer-signed on macOS) on its Releases page. It is driven two ways: from a graphical UI, or programmatically through a headless daemon with a dependency-free Python SDK (asb.py). Either way you pick an OS, point it at an installer image, and App Sandbox provisions the disk, runs an unattended install, and boots the guest. You supply a Windows or Ubuntu ISO; macOS guests download their restore image automatically. It runs on a Windows 11 (x64) PC or an Apple Silicon Mac, including laptops.
Runs on Windows 11 Home, without Hyper-V. On Windows, App Sandbox does not use Hyper-V or Hyper-V Manager; it creates and runs VMs through the Windows Host Compute System (HCS) and Host Compute Network (HCN) APIs, which require only the Virtual Machine Platform feature. Hyper-V is limited to Windows 11 Pro and Enterprise, while Virtual Machine Platform is available on Windows 11 Home, so a Windows 11 Pro license is not required. Virtual Machine Platform is the same Windows feature WSL2 uses.
What you can use it for: - A full Windows or Ubuntu desktop VM — a complete graphical OS with GPU acceleration (OpenGL, Vulkan, CUDA) — for building and testing your own software on a clean install. - Running a program in a VM, kept separate from your main OS and files. - Developing and testing Windows drivers: a built-in Test Mode enables test-signing inside the guest, so the host's Secure Boot and boot configuration are left unchanged. - Running a computer-use AI agent (such as Claude's computer use model) in its own VM. App Sandbox provides the desktop and the VM; the agent's own model loop runs inside the guest. Several agents can run at once, each in its own VM, on a single laptop or desktop. - Creating GPU-accelerated (GPU-PV) VMs, from the GUI or the headless API. - Scripting the VM lifecycle from code — create, snapshot, branch, SSH into, and delete GPU-accelerated VMs — with the Python SDK.
Headless API. appsandbox.exe --headless (or sudo /Appsandbox.app/Contents/MacOS/AppSandbox --headless on macOS) starts a single-owner daemon that hosts the same core as the GUI and exposes it as a Docker-style local HTTP/JSON API on 127.0.0.1, identical on both platforms. The stdlib-only Python SDK (asb.py) wraps it: create GPU-accelerated VMs (GPU-PV is a gpuMode create option), SSH in over an auto-deployed key (Windows, Ubuntu, or macOS guests), snapshot and branch them, open the live display, and run several at once — how many run concurrently is bounded by host CPU, RAM, and GPU memory, not by a fixed limit. The daemon is single-host — it drives VMs on the machine it runs on and does not provision cloud VMs; for CI, run it on a self-hosted runner. Provisioning a VM runs a full unattended install; snapshots, branches, and templates (Windows-only) start a new VM from a provisioned state instead of reinstalling. A VM persists until deleted. A snapshot is a disk-state checkpoint taken with the VM stopped, and a branch forks a writable disk from it to run divergent actions. The full API reference and runnable examples are in tools/headless-api/.
How it works: this repo also aims to be a working example of creating full desktop VMs programmatically with the Windows HCS/HCN APIs and Apple's Virtualization.framework. On Windows, App Sandbox submits a hand-built HCS machine document to computecore.dll / computenetwork.dll — the HCS/HCN layer that also underlies WSL2 and Windows Sandbox — rather than going through Hyper-V Manager. GPU acceleration uses GPU paravirtualization (GPU-PV) to share the host's installed GPU with the guest — the same WSL2 path, not a dedicated passthrough (no VFIO/IOMMU, no second GPU). Windows guests get DirectX 12, OpenGL, Vulkan, CUDA, and OpenCL; Ubuntu guests the same minus DirectX; macOS guests Metal. A custom IddCx indirect display driver and a virtual audio device carry the screen and sound, and guest↔host clipboard, audio, input, and SSH run over Hyper-V sockets. Guest disks are built by an in-repo tool with support for ext4, squashfs, qcow2, and VHDX. On macOS, App Sandbox uses Apple's Virtualization.framework (VZVirtualMachine, VZMacOSInstaller) over virtio-vsock. On Windows, Linux (Ubuntu) guests reach the GPU through a custom DRM/KMS kernel module (asb_drm), Microsoft's WSL2 dxgkrnl, and a custom Mesa build. The apps are native C / Objective-C with an HTML/JS UI (WebView2 on Windows, WKWebView on macOS).
高质量开源AI工作流项目
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ MIT 协议 — 最宽松的开源协议之一,可自由商用、修改、分发,仅需保留版权声明。
经综合评估,AI沙盒 在Agent工作流赛道中表现稳健,质量优秀。如果你已有明确的使用需求,可以直接上手体验;如果还在评估阶段,建议对比同类工具后再做决策。
| 原始名称 | appsandbox |
| 原始描述 | 开源AI工作流:Easily create full virtual machines that are sandboxed for development or AI Age。⭐56 · C |
| Topics | AI沙盒虚拟机GPU |
| GitHub | https://github.com/jamesstringer90/appsandbox |
| License | MIT |
| 语言 | C |
收录时间:2026-06-05 · 更新时间:2026-06-06 · License:MIT · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端