I built this to explore how online shopping can change with emerging frameworks like the Agent Commerce Protocol (ACP). This prototype illustrates what the ACP can unlock with a set of unnecessary funny shopping assistants with different personas selecting products for you. Technically, you don't need such protocols to build these experiences, but it is going to set up the shopping ecosystem to build such experiences faster. Note: the checkout functionality is illustrative.
Products are fetched live via a catalog API for each query. Results are ranked by a simple popularity score: rating × log(1 + review_count). This rewards items that are both highly rated and widely reviewed — a product with 4.8★ from 3 reviews ranks below one with 4.6★ from 800 reviews.
Stack: Catalog API → rating × log(1 + reviews) → ranked list
The same product catalog is fetched live for each query and passed in full to a Claude model (50 products, with ratings, reviews, and delivery info). Claude reasons over it and selects the top 5 matches. Clicking Buy Now on any pick opens an ACP-compliant checkout session — no redirect required.
Stack: catalog API → LLM catalog reasoning → ACP-like Stripe checkout (test)