← all repositories
tinyfish-io/agentql

SQL for web pages, if queries understood English

AgentQL lets you extract and automate across live sites using natural language instead of brittle CSS selectors.

1.4k stars Python AgentsCoding Assistants
agentql
Velocity · 7d
+1.7
★ / day
Trend
steady
star history

What it does AgentQL wraps Playwright with an AI-powered query language. You write plain-English requests like “get the product prices” and it returns structured data, even on authenticated or JavaScript-heavy pages. The project ships Python and JavaScript SDKs, a REST API, a browser debugger extension, and integrations with LangChain and Zapier.

The interesting bit The pitch is “self-healing” selectors: the same natural-language query supposedly works across similar sites and survives UI redesigns because it reasons about page content rather than DOM structure. That is the hard problem in scraping, and the entire value proposition rests on it actually working.

Key highlights

  • Natural language queries define both extraction shape and light transforms in one go
  • Deep Playwright integration: headless, stealth mode, session persistence, remote browsers
  • Browser extension debugger for live query tuning on real sites
  • REST API endpoint for use without the SDKs
  • Examples cover auth flows, infinite scroll, popups, and e-commerce pricing

Caveats

  • README is heavy on feature lists and light on how the AI query resolution actually works under the hood
  • “Self-healing” and cross-site compatibility are claimed but no benchmarks or failure rates are provided
  • The project appears to be a hosted service with API keys; self-hosting feasibility is unclear

Verdict Worth a look if you maintain brittle scrapers that break weekly when target sites redesign. Skip if you need fully offline operation or cannot tolerate opaque AI middlemen between you and the DOM.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.