Agent Builder by AirtopWeb agents that heal themselves.
Promoted
Launching today
Lyrenth
The AI-readable web index: clean pages, not raw HTML
0 followers
The AI-readable web index: clean pages, not raw HTML
0 followers
Most web readers fetch the page again for every caller. Lyrenth is an index: fetched once, cleaned into one AIDocument, then served to everyone who asks for that URL. Over 3.7 billion pages, growing by 50 million a day. Measured, not estimated: the Stripe API reference drops from 307,902 raw HTML tokens to 2,000, a 99.4% cut, and repeat reads come back in milliseconds. The AIDocument format is an open standard, so no lock-in. One line to install as an MCP server. Free tier, no card.
Hi Product Hunt,
I am Aleksandar, the founder. Lyrenth is bootstrapped and I have been building it since the end of 2025.
It started when I saw a short post while scrolling on social. Someone wrote that the next big thing is not only making AI smarter, it is giving AI the ability to see the web the way humans see it. After that I could not unsee the problem.
Every AI company, every agent, every RAG pipeline was solving the same task privately: fetch a page, strip the navigation, remove the cookie banner, parse broken HTML, and pay for that work again and again. The information was already on the web. What was missing was a layer that machines could read.
The biggest thing that changed while building it: I started out thinking the value was cleaning pages. It is not. Cleaning one page is easy. The value is that nobody has to do it twice. Once I moved from "a cleaner" to "an index", the rest followed. One document per URL. The first caller pays for the fetch, everyone after that is served from the index in milliseconds, where freshness mater, re-indexing is there. And the slow part, running JavaScript on pages that ship an empty shell, happens once on our side instead of on every caller's machine.
The second change was about the crawler. If you are going to read the whole public web, you have to be someone site owners can identify, verify, and switch off. So it publishes its addresses, honors robots.txt, signs its requests, and does not train foundation models on what it reads. The document format itself is an open standard, kept separate from the company, so it does not depend on us being around.
Where it is today: almost 3.7 billion pages across 166 million domains, growing by more than 50 million a day. The Stripe API reference goes from 307,902 raw HTML tokens to 2,000 in our format, and repeat reads come back in milliseconds. Free tier is 2,000 pages a month with no card.
I will try to read every comment here. Tell me if you find something needs improvement.
Aleksandar