The web,
rebuilt for AI.
Lyrenth turns messy web pages into clean, structured AIDocuments: smaller, cheaper, and faster for agents, assistants, and model companies to read than raw HTML. One index. The open web, made machine-readable.
From layout noise
to structured intelligence.
An indexed page is not a scraped page. It is parsed, cleaned, normalized, and rewritten into a single canonical document a model can read in one pass.
cookie consent · gdpr banner
<div id="ad-slot-top-970x250">
<script src="analytics.js">
newsletter modal · share bar
<article>
The actual content lives here,
buried under boilerplate.
</article>
related links · 38 footer nav items
tracking pixels · web fonts · css
</body>
"url": "example.com/article",
"title": "The actual content",
"lang": "en",
"content": [ 14 clean blocks ],
"entities": [ … ],
"links": [ canonical only ],
"tokens": 1840,
"noise_removed": 0.91,
"freshness": "2m ago"
}
AI reads the web the hard way.
Every crawl drags in navs, ads, cookie walls, scripts, trackers, and layout markup. Models pay, in tokens, latency, and dollars, to parse junk before they reach a single useful sentence.
A new layer between
the open web and AI.
Lyrenth sits in the middle of the stack, continuously turning the live web into a single machine-readable format every AI system can consume.
The open web
HTML · JS · noiseBillions of messy, inconsistent, render-heavy pages.
Parse · clean · normalize
render → extract → structureRender, strip boilerplate, extract entities, normalize to schema.
AIDocument JSON
clean · canonical · smallOne structured document per URL, versioned and fresh.
AI systems
agents · RAG · labsRead per-URL, or license the corpus in bulk.
One request.
One clean document.
Point Lyrenth at any URL and receive a structured AIDocument, content, entities, links, and metadata, ready to drop into a context window or a vector store.
Every read resolves against the shared index, not the origin. When a thousand agents request the same URL, the origin sees one fetch and everyone else is served from cache in milliseconds. And most of the web doesn't change between reads: for the typical page, the cached copy is the page. No other fetcher amortizes like this: a per-call scraper re-fetches the same unchanged page for every customer, every time. When freshness actually matters, force_refresh re-crawls on demand.
Every system that
needs to read the web.
If it consumes web data, it runs better on clean AIDocuments than on raw HTML.
Lyrenth is the
AI-readable
web index.
The web, transformed into clean structured data that AI systems can read, search, and use directly.
Read the web
like a machine.
Get an API key and pull your first AIDocument in under a minute. Web-scale, structured, and live.