Web Browser Extension for AgentOS

Browser automation capabilities for AgentOS agents - navigate pages, scrape content, click elements, and capture screenshots.

Features

• Navigate: Go to any URL and get page content
• Scrape: Extract content using CSS selectors
• Click: Interact with page elements
• Type: Fill in forms and input fields
• Screenshot: Capture visual snapshots
• Page Snapshot: Get accessibility tree for intelligent interaction

Installation

npm install @framers/agentos-ext-web-browser

Quick Start

import { createExtensionPack } from '@framers/agentos-ext-web-browser';
import { ExtensionManager } from '@framers/agentos';

const extensionManager = new ExtensionManager();

// Register the browser extension
extensionManager.register(createExtensionPack({
  options: {
    headless: true,
    timeout: 30000,
    viewport: { width: 1920, height: 1080 }
  },
  logger: console
}));

Tools

browserNavigate

Navigate to a URL and retrieve page content.

const result = await gmi.executeTool('browserNavigate', {
  url: 'https://example.com',
  waitFor: 'networkidle2',
  returnText: true
});
// Returns: { url, status, title, text, loadTime }

browserScrape

Extract content using CSS selectors.

const result = await gmi.executeTool('browserScrape', {
  selector: 'article h2',
  limit: 10
});
// Returns: { selector, count, elements: [{ tag, text, html, attributes }] }

browserClick

Click on an element.

const result = await gmi.executeTool('browserClick', {
  selector: 'button.submit',
  waitForNavigation: true
});
// Returns: { success, element, newUrl }

browserType

Type text into an input field.

const result = await gmi.executeTool('browserType', {
  selector: 'input[name="search"]',
  text: 'AgentOS documentation',
  clear: true
});
// Returns: { success, element, text }

browserScreenshot

Capture a screenshot.

const result = await gmi.executeTool('browserScreenshot', {
  fullPage: true,
  format: 'png'
});
// Returns: { data (base64), format, width, height, size }

browserSnapshot

Get accessibility tree for intelligent interaction.

const result = await gmi.executeTool('browserSnapshot', {});
// Returns: { url, title, elements, links, forms, interactable }

Configuration

| Option | Type | Default | Description | |--------|------|---------|-------------| | headless | boolean | true | Run browser in headless mode | | timeout | number | 30000 | Default timeout (ms) | | userAgent | string | - | Custom user agent | | viewport.width | number | 1920 | Viewport width | | viewport.height | number | 1080 | Viewport height | | executablePath | string | auto | Path to Chrome executable |

Use Cases

Web Research Agent

// Search and scrape information
await gmi.executeTool('browserNavigate', { url: 'https://google.com' });
await gmi.executeTool('browserType', { selector: 'input[name="q"]', text: 'AI agents 2024' });
await gmi.executeTool('browserClick', { selector: 'input[type="submit"]', waitForNavigation: true });
const results = await gmi.executeTool('browserScrape', { selector: '.g h3' });

Form Automation

await gmi.executeTool('browserNavigate', { url: 'https://signup.example.com' });
await gmi.executeTool('browserType', { selector: '#email', text: '[email protected]' });
await gmi.executeTool('browserType', { selector: '#password', text: 'securepass123' });
await gmi.executeTool('browserClick', { selector: 'button[type="submit"]' });

Visual Verification

await gmi.executeTool('browserNavigate', { url: 'https://myapp.com' });
const screenshot = await gmi.executeTool('browserScreenshot', { fullPage: true });
// Send screenshot to vision model for analysis

Dependencies

This extension requires Chrome/Chromium to be installed on the system. It uses puppeteer-core which does not bundle a browser.

License

MIT © Frame.dev