Products

Connect your shop's catalogue to a brand and your AI can talk about what you sell — recommending products in conversation, showing product cards in chat, quoting prices and stock read live, and following the claims and instructions you set for each product.

Overview

A brand's catalogue lives at Brand → (your brand) → Products. You connect the place your products come from, we read them, and everything your AI learned about them is kept in step with that source.

The catalogue belongs to the brand, not to the AI agent. Replace or delete the agent and your products stay exactly where they are.

Prices and stock are never stored as something the AI "learned" — they are read at the moment an answer is written, so the number in the AI's sentence and the number on the product card beside it can never disagree.

The five tabs

The page is split into five screens, each with its own address, so you can link a colleague straight to the one you mean.

Sources

Where your products come from — a Shopify or WooCommerce shop, a shop's own JSON API or a CSV feed — with its refresh schedule, its security settings and the report from its last read.

Products

Everything your AI can talk about. Open a product to see the keys it carries and which claims and instructions apply to it.

Claims and Instructions

Two tabs. Claims: what your AI may say a product does, is good for or suits. Instructions: wording your AI always follows when it talks about the products they apply to.

API

The developer reference: the JSON your shop serves, and how it pushes changes to us as they happen.

Three kinds of source

You choose the kind when you create a source, and it cannot be changed afterwards — switching means adding a new source.

A Shopify or WooCommerce shop source is the quickest way in: give us your shop's address and we read its public product list, with versions, prices and stock. Like a CSV feed, it carries no keys and takes no pushes, so only instructions set to every product reach its products. The table compares the other two.

Product APICSV feed
What it isYour shop serves its products as JSON at an address we readA Google Shopping style CSV file we download
Your shop can push changes to usYes, at any momentNo
Can carry keys for claims and instructionsYes — this is the only kind that canNo
Custom fields (ingredients, dosage, anything)YesNo
Versions (flavors, sizes, colors under one product)YesNo — every row is its own product
Nested categoriesYesYes, from the product_type column
AddressMust be httpshttp is allowed
CSV feeds cannot carry keys. Products read from a CSV file get no keys, so no claim can ever apply to them, and the only instructions that reach them are the ones set to every product of the brand. If your products need claims or instructions of their own — because you sell supplements, medicines, or anything else where the wording is regulated — you need a Product API source.
A CSV feed cannot group versions either. Every row in a CSV becomes its own product, so eight flavors of one protein powder arrive as eight separate products. If your catalogue has flavors, sizes or colors that belong together, that is a second reason to use a Product API source.

A product is a base product with versions

A product is the thing you talk about — "Gold Whey". A version is the thing a customer buys — "Gold Whey, pistachio, 1 kg" — and it carries its own price, stock, link and id.

A product you sell in one form only has exactly one version and you never think about it. A product sold in eight flavors and three sizes is one product with up to twenty-four versions, which is what lets the AI say "we have it in eight flavors" instead of listing the same thing eight times.

Identity, search, claims and instructions live on the product. Price, stock, the buy link and the sellable id live on the version.

Categories are a path, not a label

A product's category can have up to six levels — Protein > Whey > Isolate — rather than one flat name. The AI uses that shape to answer questions about your catalogue as a whole, like "what kinds of protein do you have", by asking your catalogue for its structure instead of guessing from whichever products a search happened to return.

How products reach your AI

  1. A source is read, on its schedule or because your shop pushed a change.
  2. Each product becomes a search card — its title, category, brand, every option value, its search terms and its description — split into pieces so nothing is cut. Alongside it, we write overview pages: one listing your categories, and one per category listing its products. Phone calls use these lists.
  3. When a visitor asks something in the chat, the AI searches your help content — knowledge bases, documentation, files, website pages, question-and-answer entries — and the catalogue separately, at the same time, so neither kind can crowd the other out. It can also ask the catalogue directly for its categories and what is in them.
  4. If a product ends up in the answer — at most 5 products per answer — its full sheet is built at that moment from live prices and stock, and the claims and instructions that apply to it are handed to the AI with it.

Price, sale price and availability are not part of what the AI learned. Changing a price costs nothing and needs no refresh — the price was already right. You refresh a source to add and remove products, not to correct prices.

Who can open this page

The permission is Content sources — the same one that gates the AI sources on Brand → Knowledge. A product catalogue is shared content: it is read by Yaplet, by your phone agents, by the reply suggestions your team sees in the inbox, and by AI social posting alike.

Where products do and do not appear

A product source does not appear in the sources list on Brand → Knowledge, so the same thing cannot be deleted from two different screens. Its refresh status, its last error and its product count live only on this page.

Limits

LimitValue
Products per source20,000
Versions per product500
Category depth6 levels
Refresh scheduleEvery 1 to 30 days, or off

Your organization also has overall limits on products, versions, sources, claims and instructions. If you reach one, write to us and we can raise it.

What Yaplet may learn follows who can pay: your free 14 days, bought credit or a paid plan (see How much Yaplet can learn). Only some older free workspaces keep a limit of 250 searchable pieces. There, products share that one pool with your articles, pages and files, and a product draws one to a few pieces from it depending on how much it carries. The catalogue overview pages are written for you and are not counted against it. If a read fails because the pool is full, upgrade or remove other sources to make room, then choose Fetch now.

An organization with no subscription and no credit balance also gets smaller feed limits — 5 MB and 1,000 products per read or push, six pushes a minute. Product sources has the full comparison.

© 2026 Yaplet. All rights reserved.