So we’re starting with the ones you need.

Our index skews towards fashion, then beauty, then furniture. Each has their own quirks and challenges. Our enrichment system bootstraps and auto-learns subtrees for each vertical.

We started with fashion sites for two reasons:

1. It is what our first customers needed.
2. And the data is notoriously difficult to comprehend and structure. There is no standard way for representing variants. Some sites represent variants as canonical URLs. Others make them subsets of root products. This provided a fertile training ground for our Voyager agent.

The tldr: based on API demand.

Two ways:

1. API demand /products/lookup works for both known/voyaged and unknown sites. For the latter, it uses a just-in-time LLM parser. We analyze the request counts for "unknown sites" as part of demand signals then fire the Voyager agent to discover those new sites.
2. Search quality evals We periodically run our own search queries against other agentic indexes (ie, Exa, Parallel etc) and compare it to our own. We add new catalogs that would have been made our searchresults "better".

Best case scenario: you'd get very basic data. You'd still have to do enrichment, deduping and per-store schema transformations.

Affiliate "catalog feeds" are ductaped together with antiquated feed management systems and other cruft and transported mostly over FTP. So if the is clean is clean, the latency -- and thus accuracy -- will kill you.

Hence we source our data from primary sources.

Some on our team spent many years building software products for ecommerce. The bodies are buried everywhere.

Shopify is the GOAT-ecommerce platform.

If you're reading this --Tobi and Harley and team-- we're fans 🫡.

Shopify provides programmatic access to their customers' catalogs. Which really simplifies our ability to index Shopify stores!

But:

1. Shopify's market share is ~30% of the ecommerce market in the US.
2. Stores bend Shopify's schema in all sorts of weird ways: Stuffing semantics into tags or other untyped key-value pairs. and the variant-ambiguity problem is rampant.
3. Shopify's cataog data is often upstream from what stores show their human customers. Many stores have CMS systems, review platforms and other things that patch product data temporally at the presentation layer.

Some on our team spent many years building technology for ecommerce. The bodies are buried everywhere.

No. Currently, only US-sites that sell products and serviced in USD are indexed. Supporting multiple currencies will take some work. See more on demand signals above.

Many smart teams are working on the checkout problems: the card networks, the agentic-protocol providers and the payment processors (Stripe et al).

Not sure we have an ace to serve there.

We try to keep them so fresh and so clean-clean based on demand signals. Every product response has a timestamp of when it was last updated or seen.

The /lookup endpoint allows you to explicitly force-refresh -- or force refresh if data is older than a threshold.

Dynamic data such as: price and availability-per-sku are refreshed very often.

Yes.

The dashboard provides a simplified UI for our Voyager agent. Submit a domain, and cross your fingers (Voyager's success rate is ~95% but steadily improving: A beefier-agent loop tweaks the failing ones daily).

Yes. Blueprint for how Voyager works is below.