Produce the AI overview for a shopper's search query: a short written summary of what the catalog offers for it, and the handful of products that answer it best. This is the search bar experience, so the result is deliberately compact (one set of four or five products, or several sets of a single product, each set introduced by a line of prose), small enough to sit above the traditional search results and pointed enough to draw the shopper into a full chat conversation. Not every query suits that treatment, and the index key's search bar configuration decides: a query its classification judges better served by traditional search is answered with 422 before any content is produced, and the caller renders its own search results instead. A configuration without query classification answers every query. shopper_input is the query itself. search takes no parameters, because it always runs the searchbar domain configured for the index key, and a key without that domain is answered with 400. Leave num_result_events unset unless dropping product sets is the intent: the agent chooses between one multi-product set and several single-product sets, so a cap of 1 keeps one product while the prose around it still describes the rest. Results stream back as server-sent events (text/event-stream) so the overview can be rendered as it is produced. To receive the finished overview as a single JSON document instead, send Accept: application/json.