Solutions

Run AI tools safely.

Launch Claude Code, Codex, Gemini CLI, or OpenClaw in a secure sandbox with the SDK.

Tools

One sandbox per agent session.

Each tool runs in its own microVM, started from a template with the SDK.

Claude Managed Agents

Anthropic runs the agent loop. Each session's tool calls execute in an isolated microVM on your own infrastructure.

ant beta:worker poll --on-work spawn.sh

Claude Code

Your AI coding assistant in a secure cloud sandbox. Full VM isolation for every session.

Sandbox.create('claude-code')

Codex CLI

OpenAI's coding agent, safely contained in its own microVM. No shared state.

Sandbox.create('codex')

Gemini CLI

Google's CLI agent in an isolated VM. Sandboxed execution with full filesystem access.

Sandbox.create('gemini-cli')

OpenClaw

AI agent gateway with WhatsApp, Telegram, and more. Sandboxed with built-in LLM proxy.

Sandbox.create('openclaw')

Getting started

Three steps. That's it.

Install the SDK

$npm install @omnirun/sdk

Create a sandbox

await Sandbox.create('agent')

Run commands

await sandbox.commands.run('...')

Programmatic access

Or build your own.

Use the SDK to create sandboxes programmatically. TypeScript and Python SDKs available, with full API access for any language.

Read the SDK docs

import { Sandbox } from '@omnirun/sdk'

const sandbox = await Sandbox.create({
  template: 'agent',
  memory: 1024,
})

const result = await sandbox.commands.run('python3 agent.py')
console.log(result.stdout)

await sandbox.kill()

from omnirun import Sandbox

sandbox = Sandbox.create(
  template="agent",
  memory=1024,
)

result = sandbox.commands.run("python3 agent.py")
print(result.stdout)

sandbox.kill()

Run it on your own box.

Self-host OmniRun on one KVM host, or have a14a run it for your team.

Self-host in 10 minutes Managed by a14a