Introducing Sentience: The Perception Layer for AI Agents

The Problem with Traditional Web Automation

AI agents are getting smarter, but they're still blind. Most web automation tools scrape HTML and hope the agent can figure out what's on the page. This works for simple tasks, but breaks down when you need to:

• Train agents to click specific UI elements
• Validate visual layouts for QA testing
• Extract content from dynamically rendered pages
• Generate ground truth datasets for ML models

Traditional headless browsers like Puppeteer and Playwright are powerful, but they come with significant drawbacks:

• Slow: Full browser rendering takes 2-5 seconds per page
• Expensive: Running Chrome instances costs $0.005+ per request
• Resource-intensive: Each browser uses 200-500MB of RAM

Enter Sentience: Fast, Accurate, Affordable

Sentience is the perception layer for AI agents. We built an adaptive hybrid architecture that intelligently chooses between speed and accuracy based on your needs:

Reader Mode - For Content Extraction

When you just need clean content for RAG or analysis, Reader Mode delivers:

• Clean Markdown output in ~400ms
• 90% token reduction compared to raw HTML
• Perfect for feeding LLMs and knowledge bases
• Just 1 credit per request

Map Mode - For Visual Grounding

When your agent needs to know WHERE things are on the page:

• Performance Engine: Bounding boxes in ~500ms (2 credits)
• Precision Engine: Pixel-perfect coordinates in ~2s (10 credits)
• Train agents to click, hover, and interact
• Generate visual validation datasets

Real-World Impact

Early customers are using Sentience to:

• Reduce infrastructure costs by 90%: One customer saved $4,200/month switching from Puppeteer
• Train better AI agents: Visual grounding improved agent success rates from 73% to 94%
• Scale to millions of requests: Our Performance Engine handles high-frequency automation without breaking the bank

Get Started in 5 Minutes

Ready to give your AI agents vision? Here's how simple it is:

1curl -X POST https://api.sentienceapi.com/v1/observe \
2-H "Authorization: Bearer YOUR_API_KEY" \
3-H "Content-Type: application/json" \
4-d '{
5  "url": "https://example.com",
6  "mode": "read"
7}'

That's it. No browser setup, no Docker containers, no infrastructure headaches. Just fast, accurate web perception for your AI agents.

What's Next?

We're just getting started. Our roadmap includes:

• Python and Node.js SDKs
• WebSocket streaming for real-time updates
• Custom rendering profiles for specific sites
• Advanced visual diffing for change detection

Want to stay updated? Follow our blog for weekly deep dives on AI agents, web automation, and visual grounding.