Launch HN: Channel3 (YC S25) – A database of every product on the internet

https://news.ycombinator.com/rss Hits: 8
Summary

Hi HN — we’re George and Alex, building Channel3 (https://trychannel3.com/), a database of every product on the internet, searchable via text/image, and with built-in affiliate monetization. Here’s a demo: https://www.youtube.com/watch?v=Mx8FyP7KvJg.It’s surprisingly hard to find good product data. If you want your software to recommend products and deep-link to merchants, you’ll quickly discover that the data you need—clean titles, normalized attributes, deduped listings, current prices and inventory, variant options, images, and brand info—is not just messy; it’s also spread across a long, long tail of retailers, and often lives behind advanced bot-detection systems.We ran into this problem while building an AI teacher that could recommend relevant supplies. We asked Exa for products, but got back articles, not structured data. Same for Tavily and Bing (deprecated as of 8/13/25). And we got rejected from affiliate programs, who suggested we come back with 1000s of TikTok followers. Channel3 is the API we wished we had.Product detail pages (PDPs) usually present the main item alongside recommendations. We use computer vision to isolate the main product and find its attributes, like title and price. We apply the same logic to the rest of the PDPs on the domain.Products are often sold across multiple retailers, with no guarantee they’ll be labeled consistently. We collapse products across the web into a canonicalized set by using LLMs and multimodal embeddings to actually understand each product.To normalize everything into a schema that tries to be both minimal and extensible, we have to be opinionated. Are a $50 10” and $60 12” skillet the same product? Probably not, but the S/M/L variants of a T-shirt are. Our goal is that any product you’d search for specifically is treated as its own product.We process a massive amount of data. We quickly ran out of room on our Cloudflare Vectorize indices and moved to the brand-new AWS S3 Vectors platform, syncing to OpenSearch...

First seen: 2025-08-20 17:22

Last seen: 2025-08-21 00:29