Start building
125ms boots · bare metal · us-west-2

Serverless microVMs
for AI agents

Securely host, run, and scale autonomous AI agents in 125ms. Give your LLMs a real environment with zero-trust KVM isolation, persistent memory, and WebSocket streaming. Built on bare metal.

sandbox.ts 
import { OpenHost } from '@openhost/sdk';

const host = new OpenHost(process.env.OPENHOST_API_KEY);

// Spin up a secure, isolated agent environment in 125ms
const sandbox = await host.sandboxes.create({
  image: 'openclaw/ubuntu-python:latest',
  tier: 'micro-v1'
});

// Stream agent actions directly to your frontend via WebSockets
sandbox.on('stdout', (data) => console.log('Agent:', data));

await sandbox.commands.run('python3 agent.py --task "build a snake game"');

// Terminate immediately when done
await sandbox.destroy();
Hardware-level isolation
Don't run untrusted AI code in basic containers. openhost.sh uses Firecracker microVMs with strict KVM boundaries. If an agent goes rogue, your infrastructure stays untouched.
Sub-second cold starts
Scale from zero to thousands of concurrent agents instantly. Environments boot in less than 125 milliseconds, so your users never see a loading spinner.
Stateful suspensions
LLMs spend 90% of their time waiting for API responses. We snapshot your agent's exact memory to disk when idle and wake it instantly — saving up to 80% on compute.

How it works

1
Deploy your image
Bring your own Dockerfile or use our optimized OpenClaw and Hermes templates.
2
Execute via API
Trigger the agent via REST or our TypeScript SDK. We provision a dedicated kernel and network namespace.
3
Stream the action
Connect over our secure WebSocket bridge to stream shell commands, file edits, and agent logs to your frontend.
4
Scale to zero
When the task is done the microVM shuts down. You only pay for the milliseconds the CPU was actually spinning.

Simple, usage-based pricing

Start free. Pay for the milliseconds your agents actually spin.

Starter
Perfect for testing your agent architectures.
$0
/ month
✓100 hours of microVM compute included
✓Shared CPU & 1GB RAM per agent
✓Community Discord support
Pro
POPULAR
For production agent fleets and SaaS platforms.
$29
/ month
✓$0.05 per vCPU/hour after 500 hours
✓Persistent snapshot storage included
✓Bring your own keys (BYOK)
✓Zero-trust egress filtering

Stop managing infrastructure for your agents

Join the developers building the next generation of autonomous software on openhost.sh.