About
A small team that shows its work
Hydrafetch is built by a small team led by Akash Rajpurohit, an engineer and tinkerer since 2018 who has worked at large companies and at small startups. We build the product, answer the support email and take the calls ourselves.
Founder-ledSupport from the people who build itPublic benchmarks
[ 01 / Why it exists ]
Our own products needed it.
The need
We were building products that read the web for a living. Every one of them started with the same step: give me this page as clean text a model can use.
What we found
The tools we tried returned the page with its menus, cookie notices and footers still attached, or half of it, or nothing and called it a success. The bill for a page depended on things we could not know before asking for it.
What we did
We built the reader we wanted. Then we built a way to measure it, because we did not trust our own impression of whether it was any good.
[ 02 / How claims are checked ]
Every number here traces to a source.
If you find one on this site that you cannot trace, tell us and we will correct the page.
Extraction benchmark
First on 3 of 7 page types
Scored against 6 other extractors on 1,497 labelled pages, 0.838 word-level F1 overall. The corpus, the method and every score are public.
Read the benchmarkToken index
40 sites
What well-known websites cost a model to read, measured page by page, with the method and the data next to the chart.
Read the indexComparisons
Complete tables
When we compare Hydrafetch with another product, every page we tested is in the table, with the method and the list of pages so you can run it again.
Read one[ 03 / Reach us ]
You reach the people who build it.
[ Start ]
Clean web data is one call away.
500 free credits, no card required. Failures are never billed.
Success rate
Median scrape
ms
