AI Web Agents with BrowseFleet

The Problem

AI agents need to interact with real websites: clicking buttons, filling forms, reading content, and navigating multi-step workflows. Local browsers crash, get detected as bots, and cannot handle the concurrency needed for production agent deployments. Existing tools were not designed for the screenshot-to-action loop that vision-based agents require.

The Solution

BrowseFleet's Computer API was built specifically for AI agents. Every action (click, type, scroll) returns a screenshot that you can pass directly to Claude, GPT-4o, or Gemini for the next decision. Sessions start in under a second, stealth mode prevents detection, and the Agent API provides a higher-level interface for common agent patterns.

import { BrowseFleet } from 'browsefleet';
import Anthropic from '@anthropic-ai/sdk';

const bf = new BrowseFleet({ apiKey: 'bf_...' });
const anthropic = new Anthropic();

// Create a session for the agent
const session = await bf.sessions.create({
  stealth: 'full',
  viewport: { width: 1920, height: 1080 },
});

// Navigate and get initial screenshot
let screenshot = await bf.computer.navigate(
  session.id,
  'https://example.com'
);

// Agent loop: screenshot → LLM → action → screenshot
while (!done) {
  const response = await anthropic.messages.create({
    model: 'claude-sonnet-4-20250514',
    messages: [{
      role: 'user',
      content: [
        { type: 'image', source: { type: 'base64', data: screenshot } },
        { type: 'text', text: 'Click the search button and type "cloud browsers"' },
      ],
    }],
  });

  // Execute the action, get next screenshot
  screenshot = await bf.computer.execute(
    session.id,
    parseAction(response)
  );
}

Features Used

Benefits

• Computer API returns screenshots after every action
• Compatible with Claude, GPT-4o, and Gemini vision models
• Sub-second session startup for responsive agents
• Stealth mode prevents bot detection during agent workflows
• Cookie persistence lets agents resume authenticated sessions
• Scale to hundreds of concurrent agent sessions

Learn More