Expand Fetch is a web scraping and crawling API aimed at developers building AI agents that need to read live web pages without losing structure. A single API call runs Iris, a vision model that sees the rendered page and writes Markdown while keeping tables, icons, and navigation intact, and X-Ray, which pulls hidden JSON out of the page such as coordinates or flight gates. Every fetch is stored as a snapshot with lines that link back to their location on the page, and a Highlights mode returns only the passages relevant to a question. Crawl handles whole sites and WebSync tracks changes over time.
The launch matters because agents built on standard Markdown converters routinely drop the parts of a page that answer the question, then respond with confidence anyway. Expand built MDBench with its customers, 1,094 questions across 326 hard pages, and reports that agents using Iris answer 74% correctly against 55% for the next best API, with Iris scoring 92.7% F1 versus 69.6% on element classification. Fetch is live today at $1 per 1,000 requests.
Expand describes itself as an applied research lab working toward making the web queryable like a database, with seeing the page as the first step. The company was founded by Tim Suchanek, previously co-founder of Stellate and a founding engineer at Prisma , and is part of Y Combinator's Summer 2024 batch.
Comments (10)
Sign in to join the discussion.
N
Nadia V.17h ago
The voiceover pacing in that demo clip is doing a lot of heavy lifting. Whoever edited this understood that nobody watches past 20 seconds unless you hook them with a crawl failure.
O
Obi Adeleke17h ago
Naming the vision model Iris is cute until you realize there are already four other scraping tools named Iris. Pour one out for the SEO team.
D
Dilnoza R.16h ago
Two system-one models for parsing a webpage feels like a lot of compute per fetch. Curious how the per-page pricing shakes out versus the Markdown APIs you're dunking on.
P
Prathamesh K.17h ago
Rate limits? Webhook support when a long crawl finishes? Also please tell me the SDK isn't just a thin wrapper around requests with vibes.
S
Sofía Carrillo16h ago
How are you handling robots.txt and sites that explicitly disallow AI crawlers? Asking because our legal team will absolutely ask me this in 20 minutes.
L
Liang Wei17h ago
Tweet thread cuts off mid-sentence on Iris and never finishes the thought. Bold strategy to launch a scraping product with a scrape-worthy broken thread.
M
mckenzie17h ago
The thumbnail is a wall of grey with one accent color and it kind of slaps. Finally an AI infra launch that didn't use that same purple gradient everyone else is using.
Y
Yusuf A.17h ago
Try: 'Your agent is browsing blind. Expand gives it eyes.' You're welcome, invoice in DMs.
R
raven17h ago
Every scraping demo looks magical until you point it at a Cloudflare-protected enterprise portal with a session cookie. Would love to see that in the next clip.
H
Henrik Østergaard16h ago
The best crawlers are the ones you forget are running. Everything else is just a browser with a résumé.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Expand
A new way to scrape and crawl the web
A new way to scrape and crawl the web
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Expand Fetch is a web scraping and crawling API aimed at developers building AI agents that need to read live web pages without losing structure. A single API call runs Iris, a vision model that sees the rendered page and writes Markdown while keeping tables, icons, and navigation intact, and X-Ray, which pulls hidden JSON out of the page such as coordinates or flight gates. Every fetch is stored as a snapshot with lines that link back to their location on the page, and a Highlights mode returns only the passages relevant to a question. Crawl handles whole sites and WebSync tracks changes over time. The launch matters because agents built on standard Markdown converters routinely drop the parts of a page that answer the question, then respond with confidence anyway. Expand built MDBench with its customers, 1,094 questions across 326 hard pages, and reports that agents using Iris answer 74% correctly against 55% for the next best API, with Iris scoring 92.7% F1 versus 69.6% on element classification. Fetch is live today at $1 per 1,000 requests. Expand describes itself as an applied research lab working toward making the web queryable like a database, with seeing the page as the first step. The company was founded by Tim Suchanek, previously co-founder of Stellate and a founding engineer at Prisma , and is part of Y Combinator's Summer 2024 batch.
Comments (10)
The voiceover pacing in that demo clip is doing a lot of heavy lifting. Whoever edited this understood that nobody watches past 20 seconds unless you hook them with a crawl failure.
Naming the vision model Iris is cute until you realize there are already four other scraping tools named Iris. Pour one out for the SEO team.
Two system-one models for parsing a webpage feels like a lot of compute per fetch. Curious how the per-page pricing shakes out versus the Markdown APIs you're dunking on.
Rate limits? Webhook support when a long crawl finishes? Also please tell me the SDK isn't just a thin wrapper around requests with vibes.
How are you handling robots.txt and sites that explicitly disallow AI crawlers? Asking because our legal team will absolutely ask me this in 20 minutes.
Tweet thread cuts off mid-sentence on Iris and never finishes the thought. Bold strategy to launch a scraping product with a scrape-worthy broken thread.
The thumbnail is a wall of grey with one accent color and it kind of slaps. Finally an AI infra launch that didn't use that same purple gradient everyone else is using.
Try: 'Your agent is browsing blind. Expand gives it eyes.' You're welcome, invoice in DMs.
Every scraping demo looks magical until you point it at a Cloudflare-protected enterprise portal with a session cookie. Would love to see that in the next clip.
The best crawlers are the ones you forget are running. Everything else is just a browser with a résumé.