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 for faster response times and more dynamic filtering. We hit rate limits constantly, so we spread our work over multiple cloud providers and AI models.

You can search things like “outdoor grill, less than $1000”, or “sweat-resistant, wireless running earbuds”, or "women's jeans from Paige that look like [https://www.gap.com/webcontent/0020/666/799/cn20666799.jpg]”. Products come back as JSON with title, brand, images, price, specs, etc.

Developers earn commission on sales they drive (averaging 5%). Channel3 takes a cut. We want you to earn way more money from Channel3 than you spend on it. We win when you win.

We provide an API, SDK (Typescript and Python), and MCP. We offer 1000 free searches, and charge $7/1000 searches after that. You can view expected commissions per-brand on our dashboard.

So far, products are US-only (sorry! we will expand). We’re live with millions of products and hundreds of developers.

To get started, make a free account at https://trychannel3.com, then select which brands you’d like to sell (choose from 50k+ or request your own), generate an API key, and start selling and earning.

We’d really appreciate feedback from this community. If you’ve built product search before, what did we miss in the schema? If you’ve tried to add commerce to an app, what blocked you? If you tried to build this yourself, what did you learn? Are there endpoints you wish existed (e.g. price alerts, back-in-stock webhooks, product feed)? For any suggestions, we’re all ears.

We’ll be in the thread all day to answer questions, share more technical detail, and hear whatever would make this most useful to you. Comment away!

user:	ThePhysicist
created:	February 28, 2013
karma:	12.8k
about:	Physicist, data scientist, programmer. Founder @ KIProtect http://kiprotect.com https://github.com/adewes

1.	WhatsApp is getting ads using personal data from Instagram and Facebook(noyb.eu) 23 points by ThePhysicist 336 days ago \| 2 comments
2.	Ask HN: Is the River reverse proxy project doing well? 1 points by ThePhysicist 1 year ago \| 0 comments