Give your shopping agent a background check for brands.
Agents built on agentic checkout recommend products from brands nobody vetted. One API call returns whether a brand is a real company or storefront #88 of a single operator — deterministic facts with the USPTO record attached, not a score.
The recommendation is the liability
Your recommendation is an endorsement
When the agent picks the product, the user stops evaluating brands — that's the point of an agent. When the product turns out to be a disposable storefront's, the user blames the agent, not the storefront.
Nothing at recommend time says who's behind a brand
Reviews are gameable, storefronts are weeks old, and a polished listing costs nothing. The signals agents can see are exactly the signals operators optimize.
Checkout rails don't vet brands
Agentic checkout protocols verify that a merchant can take payment, not that the brand is a real company. That layer is missing, and it's the one your users assume exists.
One call before you recommend
GET /api/v1/merchants/:domain
The store itself, checked as a whole: one state to gate on over five checks that each carry their own receipts. Who owns the matching trademark, the filer behind it, the legal entity the store publishes on its own site and whether a corporate registry agrees, and when the domain was registered. A domain that merely spells a famous brand never reads as verified.
Deterministic, so you can gate on it
No model in the loop and no score to tune thresholds against. The same input returns the same facts, and every fact links its USPTO record — an answer you can log and defend.
Batch for catalogs
Screen an assortment in one POST instead of a call per product. Same facts, same receipts, metered pricing with a free tier to build against.
Questions
What does the API return for an unknown brand?
no_record — no USPTO trademark matches the name. That's a fact, not a failure: unregistered brands exist, and your agent decides what to do with one. When multiple unrelated owners could match, the state is ambiguous and no owner is claimed.
Is there a model in the lookup path?
No. Lookups are deterministic reads of a nightly-built graph over the trademark register and marketplace seller disclosures. Same input, same answer, receipts included.
How do we start?
Sign up, create a key, and the first thousand lookups are free — no card. The docs include a quickstart and a recipe for vetting a brand before recommending it.
Related: For marketplace trust & safety · How to find who owns an Amazon brand
Try it on a brand your agent would recommend
The free check is the API's exact output, rendered. If the answer would have changed a recommendation, you need the call.
Wondering if it fits your stack?Text Josh, the founder (279) 240-3865