Full desktop environments your agents can see, click, type, and run commands in — designed around how agents actually work.
No spam. Early users get free compute credits.
// quickstart
import { AutopilotComputer } from '@autopilot/computer'
import googleSearchWorkflow from './googleSearchWorkflow.json'
// Agent gets a full dev environment
const pc = await AutopilotComputer.start()
// Open a terminal — it stays alive as a persistent process
const bash = await pc.openBash()
await bash.run('npm create vite@latest myapp -- --template react')
await bash.run('cd myapp && npm install && npm run dev')
// Check logs on demand — agent reads output whenever it needs
const logs = await bash.logs() // e.g. "ready in 320ms"
// Open a browser and navigate to the app
const browser = await pc.openBrowser()
await browser.navigate('http://localhost:5173')
// Mouse actions
await browser.leftClick(500, 300)
await browser.rightClick(500, 300)
await browser.doubleClick(500, 300)
await browser.drag([100, 200], [400, 500])
// Keyboard actions
await browser.type('Hello, world!')
await browser.press('Enter')
const screen = await pc.screenshot() // agent sees the UI
// Or run a reusable workflow
const { results } = await browser.runWorkflow({
workflow: googleSearchWorkflow,
inputs: { searchQuery: 'vite react starter' }
})
pc.openBrowser() gives your agent a live browser handle — navigate, click, and screenshot without any extra tooling.
pc.openBash() keeps a terminal alive across steps. Run commands, check logs on demand — the process never dies between agent calls.
browser.runWorkflow() executes predefined browser automations with typed inputs and outputs — composable building blocks for your agent.