Star 历史趋势
数据来源: GitHub API · 生成自 Stargazers.cn
README.md

Quicksand

PyPI Changelog Docs License: MIT

Quicksand

Quicksand is an async Python API to launch, control, and snapshot QEMU virtual machines with a particular focus on sandboxing AI agents. Quicksand provides pre-built Linux VMs for Ubuntu and Alpine distros and supports x86_64 and ARM64 across macOS, Linux, and Windows. Running sandboxes needs no root privileges or Docker; install the QEMU and image extras for a bundled runtime on supported platforms.

Installation

pip install 'quick-sandbox[qemu,alpine]'

Or install the core package and add QEMU/images separately:

pip install quick-sandbox
quicksand install qemu alpine

Usage

Hello, World!

import asyncio
from quicksand import Sandbox

async def main():
    async with Sandbox(image="ubuntu") as sb:
        result = await sb.execute("echo 'Hello from the sandbox!'")
        print(result.stdout)

asyncio.run(main())

Run commands

result = await sb.execute("apt update && apt install -y python3")
print(result.stdout, result.exit_code)

Mount host directories

Share host directories into the VM at boot or on the fly.

# At boot
async with Sandbox(
    image="ubuntu",
    mounts=[Mount("./workspace", "/mnt/workspace")],
) as sb:
    ...

# Or dynamically on a running sandbox
handle = await sb.mount("/tmp/data", "/mnt/data")
await sb.execute("ls /mnt/data")
await sb.unmount(handle)

Configure networking

Sandboxes are network-isolated by default. Opt in to internet access with NetworkMode.FULL. Port forwards work in either direction: Forward(Host(...), Guest(...)) exposes a guest port on the host, and Forward(Guest(...), Host(...)) exposes a host service inside the guest, even when the sandbox is otherwise offline.

async with Sandbox(
    image="ubuntu",
    network_mode=NetworkMode.FULL,
    port_forwards=[
        Forward(Host(8080), Guest(80)),    # host:8080 -> guest:80
        Forward(Guest(3128), Host(8080)),  # guest 10.0.2.101:3128 -> host:8080
    ],
) as sb:
    ...

Multiple Linux users

Give multiple agents independent Linux user accounts in a single sandbox VM.

async with Sandbox(image="ubuntu") as sb:
    alice = await sb.create_user("alice")
    bob = await sb.create_user("bob")

    await alice.execute("cat > hello.txt", stdin="Hello from Alice!\n")
    await bob.execute("cat > hello.txt", stdin="Hello from Bob!\n")

    for user in (alice, bob):
        result = await user.execute("whoami && pwd && cat hello.txt")
        print(result.stdout)
        await sb.delete_user(user.name)

Save and load

Save the VM's disk state to a directory. Load it later, even on a different machine.

await sb.execute("pip install numpy pandas")
await sb.save("my-env")  # VM keeps running

# Load later
async with Sandbox(image="my-env") as sb:
    await sb.execute("python3 -c 'import numpy; print(numpy.__version__)'")

Checkpoint and revert

Capture the full VM state and roll back if something goes wrong.

await sb.checkpoint("before-experiment")
await sb.execute("apt install -y something-risky")
await sb.revert("before-experiment")  # the VM snaps back to the checkpoint

Control a desktop

Desktop images provide a full Xfce4 graphical environment with a browser. Install one with quicksand install ubuntu-desktop or quicksand install alpine-desktop.

async with Sandbox(image="ubuntu-desktop", enable_display=True) as sb:
    await sb.screenshot("screen.png")
    await sb.type_text("hello world")
    await sb.press_key(Key.RET)
    await sb.mouse_move(500, 300)
    await sb.mouse_click("left")

Configuration

Here are all of the Sandbox configuration options:

Sandbox(
    # Image or save name to boot
    image="ubuntu",
    # Guest RAM (default: "512M")
    memory="2G",
    # Virtual CPU cores (default: 1)
    cpus=4,
    # Host directories shared into the VM at boot
    mounts=[Mount("/host", "/guest")],
    # NONE, MOUNTS_ONLY (default), or FULL internet access
    network_mode=NetworkMode.FULL,
    # Forward TCP ports in either direction (host -> guest, guest -> host)
    port_forwards=[Forward(Host(8080), Guest(80)), Forward(Guest(3128), Host(8080))],
    # Expand the guest filesystem on boot
    disk_size="10G",
    # Attach virtual GPU, keyboard, and mouse for screenshot/type_text/mouse control
    enable_display=True,
    # Auto-save VM state on stop
    save="my-save-name",
)

Available images

ImageTypeWheel sizeInstall commandWhat is it
ubuntuBase~309-357 MBquicksand install ubuntuUbuntu 24.04 headless
alpineBase~82-85 MBquicksand install alpineAlpine 3.23 headless (faster boot)
ubuntu-desktopOverlay (ubuntu)~273-281 MBquicksand install ubuntu-desktopUbuntu 24.04 + Xfce4 + Firefox
alpine-desktopOverlay (alpine)~346-363 MBquicksand install alpine-desktopAlpine 3.23 + Xfce4 + Chromium
quicksand-agentOverlay (ubuntu)~306-335 MBquicksand install quicksand-agentUbuntu + Python 3.12, uv, build-essential, requests, pyyaml, ddgs, markitdown
quicksand-cuaOverlay (quicksand-agent)~489-500 MBquicksand install quicksand-cuaAgent Sandbox + Xvfb, x11vnc, noVNC, Playwright, Chromium

Sizes are approximate full image-wheel downloads for the September 14, 2026 releases and vary by architecture. Overlay sizes exclude their base images.

Building from source

git clone https://github.com/microsoft/quicksand.git
cd quicksand
uv sync
uv run uvr build --all-packages

Documentation

TopicGuideUnder the Hood
InstallationInstalling packagesQEMU binaries, kernels, qcow2 disks
Sandbox LifecycleCreating and configuring sandboxes-m, -smp, -accel, machine types
Running Commandsexecute(), streaming, exit codesKernel boot, agent tokens, hostfwd
File ExchangeMounts, hot-mounts, getting data in/outCIFS via guestfwd, 9p via -fsdev
Save and RollbackCheckpoints, reverts, persistent savesqcow2 overlays, savevm, blockdev-snapshot-sync
Desktop ControlScreenshots, keyboard, mouseVNC, GPU, USB tablet, QMP input injection
Network and IsolationNetwork modes, port forwardingSLIRP NAT, restrict=on, guestfwd
PerformanceWhat makes it fastio_uring, IOThreads, TCG vs KVM

Contributing

GuideWhen to use
Creating ImagesBuild a new base or overlay image package
Extending the SandboxAdd a method, OS, architecture, or QEMU flag
TestingRun or write tests
ReleasingCut a release

Full guides: User Guide | Under the Hood | Contributor Guide

关于 About

Quickly sandbox your AI agent.

语言 Languages

Python86.4%
Rust11.7%
Dockerfile0.8%
Bicep0.8%
Jinja0.3%

提交活跃度 Commit Activity

代码提交热力图
过去 52 周的开发活跃度
330
Total Commits
峰值: 94次/周
Less
More

核心贡献者 Contributors