LYRENTH
DocsPricingBenchmarksIndex statsAboutBlogFor site ownersStatusContact

Give your AI agent
clean web pages.

Send any public URL and get consistent Markdown and structured page data from a shared index. Use up to 99.4% fewer input tokens on measured pages.

POST
from the cached index · site not contacted2,722 tokens · 89.0% saved
Web indexing - Wikipedia
en.wikipedia.org/wiki/Web_indexing

From Wikipedia, the free encyclopedia Methods for indexing the Internet **Web indexing**, or **Internet indexing**, comprises methods for indexing the contents of a [website](https://en.wikipedia.org/wiki/Website) or of the [Internet](https://en.wikipedia.org/wiki/Internet) as a whole. Individual websites or [intranet

lang entype Article1,556 words8 min readhas JSON-LD
Tokens saved
89.0%
24,7342,722
AIDocument tokens
2,722
what the model reads
Origin fetches
0
shared, cross-caller cache
Served from
cached index
instant · fresh copy on demand
No account, one command
curl "https://api.lyrenth.com/v1/public/aidocument?url=https://en.wikipedia.org/wiki/Web_indexing"
Or add it to your agent
claude mcp add --transport http lyrenth https://api.lyrenth.com/mcp \
  --header "Authorization: Bearer $LYRENTH_API_KEY"
  • 25 reads an hour with no account.
  • 2,000 a month with a free key.
  • The first 1,000 to claim it get Starter, 20,000 a month, free for 90 days.
  • No card.
AI-readable pages indexed
   
   
Indexed domains
  M
Pages audited
24h to 90d
Freshness, by plan
Real data, from the production API

What famous pages cost,
raw vs indexed.

Every row is a real page read through the index, with token economics reported by the API itself. Reproducible with one call.

PageRaw HTML → AIDocumentRaw tokensAIDocumentSaved
Stripe API reference
154× smaller
307,9022,00099.4%
Vercel Functions docs
110× smaller
237,3902,14999.1%
Cloudflare Workers docs
69× smaller
94,9631,37798.5%
GitHub REST API quickstart
18× smaller
88,6144,87694.5%
Kubernetes Pods concepts
18× smaller
131,9777,47594.3%
Measured 2026-07-05 · cost basis $3.00 / 1M input tokens · grey is the raw page, teal is what the model readsAll 10 benchmarks
One request, one document

The same shape,
for every URL.

Send any public URL, get one AIDocument back: the same grouped JSON envelope every time, versioned forever. Validate against the public JSON Schema.

Full AIDocument schema in the docs
POST/v1/aidocumentresolve any URL
GET/v1/read?url=…markdown, one GET
POST/v1/submitqueue for indexing
GET/v1/statspublic counter
GET/aidocument.schema.jsonthe contract
Quickstart: first call in a minute
Check your own site

How does AI
actually read you?

AI cannot recommend what it cannot read. Paste your URL and see exactly what an agent receives from your page: how much of it is real content, what is missing, and what to fix. Most sites have never been checked.

Check your site Free with an account. No card.
The adapter layer

An index,
not another scraper.

Most pages you ask for have already been read. Someone else's request warmed the index, so yours comes back in milliseconds and the site is never touched again. The more people read through Lyrenth, the more of the web is already waiting when you arrive. When freshness matters, force_refresh re-indexes on demand.

01 · Origins
one crawl →
02 · The standing index
→ every read
03 · Readers
crawl
crawl
crawl
Lyrenthindexing · live
AIDocuments in the index
POST/v1/aidocumentHIT

One canonical AIDocument per URL. Written once, read by everyone: shared infrastructure, not a per-customer scrape.

Freshness, per request
cache_firstforce_refreshyou choose
Autonomous agents
per-url · mcp
AI assistants
retrieve · cite
AI search & discovery
search · rank
RAG systems
embed · index
Model labs
corpus · bulk
Enterprise & research
pipeline · api

The open web, crawled on our schedule. Readers ask for the latest; Lyrenth fetches and serves it from the index.

A standing index sits between the two. Rendered, cleaned, normalized, and held.

Every reader resolves against the index and is answered in milliseconds.

01 · Origins
crawl
crawl

The open web, crawled on our schedule. Readers ask for the latest; Lyrenth fetches and serves it from the index.

one crawl
02 · The standing index
Lyrenthindexing · live
AIDocuments in the index
POST/v1/aidocumentHIT

One canonical AIDocument per URL. Written once, read by everyone: shared infrastructure, not a per-customer scrape.

Freshness, per request
cache_firstforce_refreshyou choose

A standing index sits between the two. Rendered, cleaned, normalized, and held.

every read
03 · Readers
Autonomous agents
per-url · mcp
AI assistants
retrieve · cite
AI search & discovery
search · rank
RAG systems
embed · index
Model labs
corpus · bulk
Enterprise & research
pipeline · api

Every reader resolves against the index and is answered in milliseconds.

Is notRaw crawlingScraping-as-a-serviceA normal search engineAnother data broker
IsThe AI-readable web index
The transform

Signal,
not markup.

Every crawl of raw HTML drags in navs, ads, cookie walls, scripts, and trackers. Models pay, in tokens, latency, and dollars, to parse junk before reaching a single useful sentence.

18× to 154× smaller
on the five pages above
Tokens to read one pagesignal · 9%
Raw HTML page~14,800 tok
Actual signal~1,330 tok
Lyrenth AIDocument~1,840 tok
Built for

Built for agents
that read the web.

If it consumes web data, it runs better on clean AIDocuments than on raw HTML.

01Autonomous agentsBrowse and act with structured pages instead of burning context on markup.PER-URL · MCP
02AI assistantsAnswer with fresh, normalized web content and clean citations.RETRIEVE · CITE
03AI search & discoveryBuild on a corpus already structured for ranking and recall.SEARCH · RANK
04RAG systemsSkip the scrape-and-clean pipeline: embed AIDocuments directly.EMBED · INDEX
05Model labsGround, retrieve, and evaluate on a clean, deduplicated web corpus.CORPUS · BULK
06Enterprise & researchMonitor, extract, and analyze the live web as structured data.PIPELINE · API
Get started

Point it at a URL and see.

Your first read takes about a minute. 2,000 a month free, no card, and nothing to install.