Introduction

OpenSteer is a browser automation framework for AI agents with deterministic replay.

Install

# npm
npm install opensteer

# pnpm
pnpm add opensteer

# bun
bun add opensteer

Requirements

• Node.js >= 20
• Playwright-supported browser (Chromium, Firefox, WebKit)
• API key for LLM features (CUA agent, structured extraction)

What it does

• Unified local/cloud execution — same API surface, swap with a flag.
• Descriptor-aware actions with replay — interact by description, cache selectors for deterministic re-runs.
• Structured data extraction — pull typed data from pages using schemas.
• CUA agent support — delegate multi-step browser tasks to an LLM agent.

Core workflow

import { Opensteer } from 'opensteer'

const opensteer = new Opensteer({ name: 'demo' })

try {
  // Launch a browser session
  await opensteer.launch()

  // Navigate
  await opensteer.goto('https://example.com')

  // Snapshot for action context
  await opensteer.snapshot({ mode: 'action' })

  // Interact by description
  await opensteer.click({ description: 'more-info-link' })

  // Snapshot for extraction context
  await opensteer.snapshot({ mode: 'extraction' })

  // Extract structured data
  const data = await opensteer.extract({
    description: 'page-heading',
    schema: { title: 'string', href: 'string' },
  })
  console.log(data)
} finally {
  await opensteer.close()
}

Runs locally by default. Pass cloud: true (or set OPENSTEER_MODE=cloud) for cloud execution.