An ultra-light micro-VM sandbox for running untrusted workloads with hardware-enforced isolation, built on OpenVMM with Linux as the guest.
78/tenbox
A cross-platform VMM that runs AI agents such as OpenClaw inside isolated Linux virtual machines with controlled file access.
- Stars
- 262
- Forks
- 62
- Open issues
- 17
- Last commit
- last month
What is it?
TenBox is a virtual machine monitor for running AI agents on a personal computer inside an isolated VM. The agent can only reach the files the user explicitly shares, which limits what it can touch on the host.
The project is built on a shared C++ runtime with platform-specific hypervisor backends: WHVP on Windows, Hypervisor Framework on macOS (Apple Silicon and Intel), and KVM on Linux (x86_64 and arm64). Windows and macOS hosts get native GUI managers, while Linux runs a systemd daemon, tenboxd, controlled through a local CLI.
Guests are full Linux desktops with virtio-gpu display over SPICE, audio, virtiofs shared folders, clipboard sharing, NAT networking with port forwarding, and qcow2 or raw disk images. On Linux, the daemon also offers browser-based remote desktop and a built-in OpenAI-compatible LLM proxy.
Key features
- 01Hardware-accelerated VMs on Windows (WHVP), macOS (Hypervisor Framework), and Linux (KVM).
- 02Native managers on Windows (Win32) and macOS (SwiftUI/AppKit), plus the tenboxd daemon and CLI on Linux.
- 03Shared folders via virtiofs with optional read-only mode.
- 04Bidirectional clipboard sharing and audio output from the guest.
- 05NAT networking with host and guest port forwarding.
- 06Browser-based remote desktop on Linux via libdatachannel and H.264/Opus streaming.
Use cases
- Running an AI agent like OpenClaw that should only see a chosen set of host files.
- Isolating experimental agent workloads from personal data on a desktop machine.
- Hosting a headless Linux agent on a Raspberry Pi 5 managed by systemd.
Quick start
- 1
Run the official install script for Linux from the README (requires /dev/kvm and glibc 2.31+)
- 2
Open the pairing URL printed after install, or visit my.tenbox.ai, to claim the host
- 3
On Windows or macOS, download the latest installer from tenbox.ai or GitHub Releases
Things to know
- Linux install requires Debian 11+, Ubuntu 20.04+, or Raspberry Pi OS 11+, and /dev/kvm.
- Other arm64 boards beyond x86_64 PCs and Raspberry Pi 5 are not yet validated.
- Licensed under GPL v3, a copyleft license that applies to derivative works.
In collections
Topics
FAQ
- What is tenbox?
- A cross-platform VMM that runs AI agents such as OpenClaw inside isolated Linux virtual machines with controlled file access. TenBox is a virtual machine monitor for running AI agents on a personal computer inside an isolated VM. The agent can only reach the files the user explicitly shares, which limits what it can touch on the host.
- Is tenbox free to use? What license does it have?
- tenbox is open source under the Other license. Check the license terms before using it in commercial products.
- Is tenbox still maintained?
- Updated within 6 months. The last commit to tenbox was on Sep 17, 2026.
- What can I use tenbox for?
- Running an AI agent like OpenClaw that should only see a chosen set of host files. Isolating experimental agent workloads from personal data on a desktop machine. Hosting a headless Linux agent on a Raspberry Pi 5 managed by systemd.
- How do I install tenbox?
- Run the official install script for Linux from the README (requires /dev/kvm and glibc 2.31+) Open the pairing URL printed after install, or visit my.tenbox.ai, to claim the host On Windows or macOS, download the latest installer from tenbox.ai or GitHub Releases
- Where can I get tenbox?
- The source code and installation instructions are on GitHub: https://github.com/78/tenbox
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Runs Windows inside a Docker container with automatic download and unattended installation, accessed through a browser or RDP.
A minimalist KVM-based virtual machine monitor that runs lightweight microVMs for secure, multi-tenant serverless and container workloads.
Open-source sandbox service for AI agents that runs each workload in a hardware-isolated MicroVM, with E2B SDK compatibility and sub-60ms startup.
Fork-style microVM sandboxes for AI agents: children fork copy-on-write from a warmed Firecracker parent, KVM-isolated, with live BRANCH of running VMs.
Docker images that run full Linux desktop environments (XFCE, KDE, MATE, i3, LXQt) in any modern web browser.