Product Tools

What an AI can do with your product catalog through Yaplet or a connected AI app — connect, change and refresh product sources, and write, switch on and delete the claims and instructions the AI follows when it talks about your products.

Overview

Through Yaplet, the assistant inside your dashboard, or a connected AI app, an AI can connect the catalogs a brand's products come from, keep them up to date, and write a brand's product rules: its claims (what a product may be said to do) and its instructions (what the AI must follow whenever it talks about a product). It looks the actions below up with find_actions and runs each one with the run tool for its kind — see MCP Server. In find_actions these actions sit in the chatbot area, next to the Chatbot Tools.

Permission required:Content sources, not AI agent. A product catalog is shared content, read by your website's AI agent, by phone agents, by inbox reply suggestions and by AI social posting alike. An app connected by browser login also needs the Chatbot area ticked on the login screen.
Brand limits apply. A team member limited to some brands reaches only those brands' product sources, claims, instructions and products — to read them and to change them — in Yaplet, in apps connected by browser login, and through API keys whose AI access they switched on. Another brand's items are refused as not found or not accessible.

For what any of this means in the product, see the dashboard documentation on claims and product instructions and product sources.


Actions

ActionLevelWhat it does
product_rules.createChangeCreates one claim or instruction for a brand (type: claim or instruction, instruction by default): its keys and its text. It starts switched off (a draft) unless you confirmed the exact wording.
product_rules.bulk_createChangeCreates up to 200 claims or instructions in one go, all or none — for example to load a list of approved claims. Switched off unless you confirmed the wording.
product_rules.updateChangeChanges one claim's or instruction's keys, text, note, scope, type or status. Changing type moves it between the Claims and Instructions tabs; setting it back to draft switches it off without deleting it.
product_rules.approveChangeSwitches on claims or instructions that are off (drafts) — chosen ones, or every draft of a brand, optionally only those from one document import.
product_sources.createChangeConnects a catalog: a Shopify or WooCommerce shop (just its address), the shop's own JSON (Product API) or a product feed (Google Shopping, Heureka, Árukereső or Ceneo).
product_sources.updateChangeChanges a source's name, address, secret, extra headers, key fields, schedule, or what happens after repeated failures. Refused on a website-based list.
product_sources.fetch_nowChangeReads a source's products in again right away — one source by id, or every source of a brand by brandId — each at most once every 5 minutes. Uses AI credits. Refused on a website-based list.
product_rules.deleteDelete or sendDeletes one claim or instruction for good. The AI must repeat its name — the first 60 characters of its text.
product_sources.deleteDelete or sendDeletes one source and all of its products for good. The AI must repeat the source's name. Refused while a read or push is running. Also refused on a website-based list.

How Claims and Instructions Work

A product rule is a list of keys (the words a product and the rule share, usually an ingredient name) plus the text the AI reads, and a type:

  • claim — what a product may be said to do, be good for or be suitable for, such as an approved health claim with any condition it states. A claim also finds the products for a visitor's goal.
  • instruction (the default) — followed whenever the AI talks about the products it applies to: what to say and what never to say, warnings, and facts such as "contains gluten".
New claims and instructions start switched off (as drafts), and those never reach the AI. An AI switches one on — by creating it as active, or by approving it — only after you have read the exact wording and confirmed it. Otherwise it leaves them switched off for you to approve.
  • Matching is exact, never fuzzy. A rule applies to a product when they share a key, compared after lowercasing, removing accents and collapsing punctuation. Every spelling of the same thing goes into one rule's keys; synonyms and translations are not matched for you.
  • Scope "brand" reaches every product — for instructions only. A brand-wide instruction has no keys and applies to every product of the brand, keyless products included — for a general disclaimer or a word never to use. A claim always belongs to products through keys, so a brand-wide claim is refused.
  • Every switched-on rule counts. Every switched-on claim and instruction that reaches a product goes with it, and type says which of the two a rule is.
  • Claims switch the brand's claims mode. While a brand has at least one switched-on claim, it is in claims mode: its AI says what products do only with their claims, and recommends a product for a goal only when a claim covers it — see How Yaplet finds answers. A call that would switch on a brand's first claim, or switch off or delete its last one, is refused and nothing changes; the answer explains what the switch would mean. The AI must then tell you, ask you, and only after your yes send the same call again with confirm_claims_mode: true. In Yaplet, the Confirm card of such a call shows a Claims line saying which way it may switch.
  • The claim check. In claims mode, Yaplet works out which of the brand's claims really fit the visitor's question; a chosen claim brings its products into the answer and leaves look-alike products out (a product the visitor names is kept). Brands without a switched-on claim never run it.
  • A keyed rule needs at least one key to go live. An approval that includes such rules fails and names them; brand-wide rules are exempt.
  • Bulk create is all or nothing. If one rule is refused, or the organization's limit on rules would be passed, none are created.

How Product Sources Work

  • Three kinds. A Shopify or WooCommerce shop source is created with just the shop's address: Yaplet reads the shop's public product list on a schedule of whole days. It has no secret, extra headers, key fields or pushes, so only instructions set to every product reach its products, and a shop without a public product list fails its first read with a message saying so (use a Product API source, or a product feed source, for that shop). A Product API source is the shop's own JSON, read on a schedule of whole days and pushable by the shop at any time; only this kind carries rule keys, and its address must be https. A product feed source reads the feed link a shop system gives — Google Shopping (XML, or CSV with commas, semicolons or tabs), Heureka, Árukereső or Ceneo — with no keys and no rules. Rows that share a group id become one product with versions, and the feed's parameters (color, size, material and similar) become product facts. The kind cannot be changed later.
  • A website-based list follows its site. The source called Your website (kind website) is made by Yaplet from the pages of your website that carry product data, and it follows the site's Pages with product data switch. product_sources.update, product_sources.fetch_now and product_sources.delete refuse it. To stop it, an AI switches the site to learn those pages as pages with knowledge.set_product_pages_mode (see Chatbot Tools); removing the site removes the list too. Given a brand, product_sources.fetch_now skips it and says why.
  • The secret is shown once. Creating an API source returns its secret, the push address and how the shop signs in — the organization's API key in the Y-API-Key header. Later answers only say whether a secret exists; the dashboard shows it again. A CSV, XML feed or shop source has no secret and cannot be pushed to, so its answer carries neither a secret nor a push address. The full JSON shape is described in The Product API.
  • A moved source gets a new secret. When an API source's address moves to another host, or a push-only source gets its first address, and the AI sends no new secret, the source gets a freshly generated secret and its extra headers are cleared — a secret your shop knows never follows the source to someone else's server. The answer says so; give your shop the new secret from the brand's Products → Sources page.
  • Key fields take effect at once. Changing which fields become the products' keys re-derives the keys of every product of that source, with no new read.
  • Fetch now runs in the background. Inside the five-minute wait the AI is told how many minutes are left — an answer, not an error. Otherwise it gets a run id and checks the outcome by reading the source again. Given a brand instead of one source, it refreshes every source of that brand, each at most once every five minutes, and lists the sources it skipped with the reason: the products arrive by push so there is no address to read from, a read is already running, minutes are left, or the list follows the website's pages. A brand with no sources is refused.
  • Repeated failures. After three failed refreshes in a row a source is switched off and the owner is emailed. The source's failure setting decides whether its products stay visible ("keep", the default) or are hidden until it works again ("hide").

Reading Products, Sources and Rules

Three resources are available through the Data Access tools, gated by the same Content sources permission:

ResourceWhat it holds
product_sourcesThe catalogs products come from, with their schedule and last run.
product_rulesClaims and instructions, each with its keys, its type (claim or instruction) and its status.
productsThe catalog itself: base products with their versions, plus each product's keys and its own rule sentences.

Read one rule with get_record and it also says how many products it reaches today; read one product and it lists every claim and instruction that applies to it, switched-off ones included, each with its type. To list only claims or only instructions, filter product_rules with ruleType (claim or instruction). Together they are the fastest way to check a set of claims before switching them on.


What an AI Cannot Do Here

  • Delete several claims or instructions at once, or every draft of a brand. An AI deletes one named claim or instruction per call. To delete many, tick them in the table on the brand's Products → Claims or Products → Instructions tab and delete them together.
  • Show a source's secret again after creation. It is shown on the brand's Products → Sources page.
  • Create the API key your shop pushes with. API keys are human-only in every connection: create one under Settings → Connections → MCP & API keys.

Example Prompts

  • "Create a product API source called 'Shop feed' on my brand, reading https://shop.example.com/yaplet/products every day, and turn the Ingredients field into keys."
  • "Create these approved claims switched off, one claim per substance, keyed on the substance name and its common synonyms."
  • "Show me every product on my brand that has no keys, so I know which ones no claim or instruction can reach."
  • "How many products does the vitamin C claim reach today?"
  • "Fetch the products of my Shop feed now."

© 2026 Yaplet. All rights reserved.