Sandboxes for AI agents.
One command away.

Noid is a lightweight micro-VM orchestration tool for isolating AI agents — with checkpoint and restore built in. Run it on your infrastructure. No vendor lock-in. No per-minute billing.

Install Noid Read the docs →

$ noid create my-vm
VM created: my-vm (running)

$ noid exec my-vm -- echo "Hello from Noid"
Hello from Noid

$ noid console my-vm
noid@noid:~$ exit

$ noid checkpoint my-vm --label before-deploy
Checkpoint created: ckpt_a1b2c3d4

The Problem

Hosted sandboxes scale costs, not control

Linear cost scaling.

Hosted VM platforms charge per CPU-hour and per GB-hour. At 100 VMs, you're spending thousands per month for compute you could own.

No infrastructure control.

Your code runs on someone else's machines. No control over data residency, networking, kernel versions, or storage backends.

Vendor lock-in by design.

Proprietary APIs, managed runtimes, opaque billing. Migrating away means rewriting your integration layer.

The Solution

Your VMs. Your hardware. Your rules.

Noid runs isolated sandboxes on any Linux host you control. A single binary manages VM lifecycle, and the CLI talks to it from Linux or macOS. Checkpoint, clone, and restore VMs instantly — with zero-copy snapshots on btrfs.

One-command VMs

noid create spins up a sandbox with networking, serial console, and exec in seconds. No YAML. No manifests. No orchestrator.

Instant checkpoints

Freeze VM state to disk and restore it later — or clone it into a new VM. On btrfs, snapshots are zero-copy.

Run from anywhere

The CLI runs on Linux x86_64, macOS Intel, and macOS Apple Silicon. Manage remote VMs from your laptop over authenticated HTTP.

Built for automation

Multi-tenant with token auth, a REST + WebSocket API, and a TypeScript SDK (@noid/sdk). Build on top of Noid programmatically.

How It Works

Three commands. Full lifecycle.

Create

$ noid create my-vm --cpus 2 --mem 4096
VM created: my-vm (running)

$ noid exec my-vm -- uname -a
Linux noid 6.12.71 #1 SMP x86_64 GNU/Linux

Create a VM and run commands inside it. Networking, console, and exec are ready immediately.

Checkpoint

$ noid checkpoint my-vm --label before-deploy
Checkpoint created: ckpt_a1b2c3d4

$ noid restore my-vm ckpt_a1b2c3d4 --as my-vm-v2
VM restored: my-vm-v2 (running)

Freeze VM state at any point. Restore it or clone it into a new VM. On btrfs, this is instant.

Automate

import { NoidClient } from '@noid/sdk';

const client = new NoidClient({
  baseUrl: 'https://noid.example.com',
  token: process.env.NOID_TOKEN!,
});

const vm = await client.createVm('agent-sandbox', { cpus: 2, memMib: 4096 });
const result = await vm.exec(['python3', 'run.py']);
const checkpoint = await vm.createCheckpoint({ label: 'done' });

Use the TypeScript SDK to integrate Noid into your platform. Every CLI operation is available programmatically.

Comparison

Noid vs. hosted platforms

Feature	Noid	Hosted Platforms
Runs on your hardware	Yes	No
Pricing model	Free	Per CPU-hour + per GB-hour
Data residency	You choose	Provider-controlled
Checkpoint/restore	Instant (btrfs zero-copy)	Varies / limited
Self-hosted	Yes, single binary	No
CLI + SDK	Linux, macOS, TypeScript	Platform-specific
Open source	Source available	Partial or closed
Vendor lock-in	None	Ecosystem-dependent

Use Cases

Built for workloads that need isolation

AI Agent Sandboxes

Give each agent its own VM. Execute untrusted code, install packages, checkpoint state between turns. Restore to a known-good state if anything goes wrong.

Coding Sandboxes

Spin up isolated dev environments per user or per session. Instant restore means you can reset to a clean state in milliseconds, not minutes.

Long-Running Workloads

Run background jobs, CI pipelines, or data processing inside VMs with full kernel isolation. Checkpoint progress and resume later.

Dev/Test Environments

Clone a VM at any checkpoint to create parallel test environments. Test against real system state, not mocked containers.

Start in 30 seconds

curl -fsSL https://noid.one/install | sh

Or download binaries directly from GitHub.

CLI works on Linux, macOS, and Windows WSL. Server requires Linux with KVM support.

Sandboxes for AI agents.One command away.