Walmart for AI Agents

The core loop is search to cart. search_products takes a query and lands on the search results page. From there your agent can sort_results, apply sidebar facets with filter_results (a filter_name plus a value), and page through the grid with next_page and prev_page. view_product opens a product detail page by a distinctive substring of the product name, and add_to_cart puts the item in the basket. Cart management is fully mapped too: view_cart, update_quantity, and remove_from_cart all work on individual line items.

Beyond shopping, the config covers the signed-in account surface: checkout, promo codes, order tracking, cancellations, returns, order history and order detail, wishlists, loyalty status and point redemption, plus saved addresses and payment methods. There are also public utility actions like browsing categories, reading reviews, and finding stores with view_store_locator.

These come straight from the mapped action behavior, and your agent should plan around them.

• **Walmart has two product card types.** Simple single-SKU products carry an Add to cart button directly on the search-result card, so add_to_cart works from the results grid. Variant products show an Options link instead: your agent must view_product first and pick options on the product detail page before adding to cart.
• **Adding and quantity are separate verbs.** add_to_cart adds an item; any quantity change goes through update_quantity, which drives the same quantity spinbutton whether the item is on the product page or a cart line. remove_from_cart deletes the line entirely.
• **Pagination is URL-driven.** Walmart paginates search results via a ?page={N} URL parameter at roughly 40 results per page, and next_page clicks the next-page button at the bottom of the grid. Budget for multiple pages on broad queries.
• **Search navigates by URL, not by typing.** search_products uses URL-based navigation for reliability, so searches land directly on the results page instead of depending on the search box.

Walmart sign-in is passwordless. The login action takes an email param: Walmart emails the account a 6-digit verification code, and the user enters that code in the Rindler Connect form. The mapped login URL is walmart.com/account/login, and Rindler handles the session from there.

You only need it for the account half of the catalog. 13 of the 29 actions require auth, including checkout, track_order, cancel_order, return_order, view_order_history, and the saved-address and payment-method actions account_addresses and account_payment_methods. Everything in the search-to-cart loop is public, so an agent can do product research and cart building anonymously and only prompt for sign-in when it is time to check out or touch the account.

The config models Walmart as four screens: landing, search_results, product_detail, and cart. Every action lands your agent on one of them, and Rindler returns each screen as structured content rather than raw page markup.

A session starts on the landing screen. search_products moves to search_results, where filtering, sorting, and pagination keep you on the same screen with updated data. view_product moves to product_detail, which is also where variant products must be handled before adding to cart. view_cart moves to the cart screen, where quantity updates and removals operate per line item. Because the screen model is explicit, your agent always knows where it is and which actions apply next.