Every AI agent launches Chromium, waits 5 seconds, then asks an LLM “what am I looking at?” Browsy skips all of that. It parses HTML, computes layout, and tells your agent: “This is a login page. Fill field 19, click 34 to submit.” 200ms. 6MB binary. No browser.
Browsers compute layout, then paint pixels. Browsy keeps the layout, adds page intelligence, and skips the paint entirely.
html5ever — the same battle-tested parser behind Firefox Servo — handles real-world HTML with all its quirks. We get a clean DOM tree.
Our from-scratch CSS engine: selector matching, property parsing, var() resolution, calc() evaluation, @media queries, and style inheritance. Feeds computed styles to Taffy for Flexbox + Grid layout.
Our custom output layer: element emission, smart deduplication (34-42% reduction), landmark markers, hidden content exposure, text fallback chains, behavior detection, and delta diffing.
Benchmarked against 50+ tools. 100% detection accuracy across 39 real-world snapshots. No Chromium, no GPU, no binary dependencies.
| Tool | Approach | Speed (HN) | Chars/element | Dependencies | Page Intelligence |
|---|---|---|---|---|---|
| browsy | Zero-render (Rust) | 203ms | 58 | 6MB binary | 13 action types |
| Jina Reader | Cloud API | ~1,200ms | ~96 | Cloud API | None |
| agent-browser | Playwright wrapper | ~5,377ms | ~157 | Chromium (282MB) | None |
| Playwright MCP | Screenshot + a11y tree | ~5s | ~120 | Chromium (282MB) | None |
| Browser Use | Playwright + vision | ~5s | ~150 | Chromium + Python | LLM-only |
| Stagehand | Playwright + LLM | ~5s | ~140 | Chromium + Node | LLM-only |
Page intelligence, hidden content, and deterministic output. Everything an agent needs to understand the web — nothing it doesn't.
Automatic page type detection — Login, Search, Form, Article, List, Captcha, Dashboard, SearchResults, and more — with 13 action recipes and stable element IDs. Your agent gets “fill field 19, click 34” instead of a raw tree to interpret.
Dropdown menus, modals, accordion panels, tab content — it's all in the HTML, just hidden by CSS. Browsy includes it with a hidden: true flag. Agents see the full page without executing JavaScript.
Search DuckDuckGo or Google directly through browsy. Search and fetch the top N result pages in a single call — no separate search API needed.
Navigate, click, type, select, go back, search by text or role. Full agent action vocabulary with cookie persistence and O(1) element lookup by ID.
Real HTML is full of wrapper noise. Browsy detects and collapses redundant containers — 40% reduction on Hacker News, 42% on Wikipedia.
After first load, only changes are emitted. Filter to above-fold, below-fold, or visible-only elements. Dramatically reduces token cost for multi-step sessions.
Detects interactive patterns from HTML alone — onclick handlers, Bootstrap toggles, ARIA controls. No JS execution needed. Agents see what's behind every dropdown without clicking.
Detects reCAPTCHA, hCaptcha, Cloudflare Turnstile, and image grid challenges from HTML structure. Cookie consent banners surfaced automatically. Your agent knows when it's blocked before wasting tokens.
Distinguishes registration, contact, login, and generic forms. Extracts field names, types, and labels. Download links identified with file extensions. 13 action types total — not just “it's a form.”
Page intelligence is computed, not inferred. The same HTML always produces the same Spatial DOM, the same page type, the same action recipe. Auditable, reproducible, debuggable.
Page intelligence from raw HTML. No browser, no LLM, no guessing.
Navigate to a login page. Browsy detects the page type, identifies the form fields, and gives your agent an action recipe — no LLM interpretation needed.
page_type tells you what you're looking at. suggested_actions tells you what to do. Element IDs are stable across sessions.
Dropdown menus, modals, accordion panels — Browsy exposes everything Chrome's accessibility tree hides. Your agent sees the full page without executing JavaScript.
90% of the web pages agents interact with don't need a browser to understand. These industries are the sweet spot.
Benefits applications, permit filings, tax forms. Government portals are server-rendered HTML — browsy's sweet spot. Deterministic output means a clean audit trail.
Court filing systems like PACER, state court portals, regulatory filings. Legacy server-rendered HTML with high compliance requirements. Auditable, deterministic output.
Prior authorization forms, patient portals, insurance claim portals. HIPAA-friendly: browsy never renders PHI visually, never executes JS, never stores screenshots.
Banking portals, loan applications, KYC forms, SEC EDGAR filings. Extract mortgage rates from 50 banks hourly. No browser fingerprint reduces detection risk.
Product catalog extraction, price monitoring, inventory checking. At scale, Chromium instances cost $0.10/hour each. Browsy processes thousands of pages per minute on a single machine.
Job board data extraction, application form filling across career portals. Apply to 50 matching job postings by filling out each employer's portal automatically.
MIT licensed. 6MB binary. 200ms per page. No Chromium. No dependencies. Free forever.