Skip to content
AskVault Docs
Try Free →

The recommend_product skill

Product questions are the ones that decide a sale: "do you sell this?", "how much is it?", "is it in stock?", "what would you recommend under $50?". The recommend_product skill lets your agent answer them with real products from your catalog instead of prose.

Off by default for every workspace. Turn on Product Recommendations under the Skills page for the workspace you want it on: that is also how you pilot it on one store before the rest. (In preview: if it is not in that list, the capability is not switched on for your account yet.)

What it does

When a visitor asks anything about products, pricing, availability or recommendations, the agent calls recommend_product. The skill searches only your synced product catalog and replies with:

  • a product card when there is one good match, or a row of cards when there are several. A carousel on the web widget, Block Kit cards in Slack, and the same details as text on WhatsApp and Telegram;
  • the title, price, currency and availability from your synced catalog;
  • a View product button linking to the product page, when the catalog has a URL for it.

If nothing in your catalog matches, the agent says so. It does not invent a product, a price, a currency or a stock status: not once, not "helpfully".

Where the products come from

The skill reads the catalog AskVault already holds for you. Three things fill it, and the Source column in Knowledge › Products tells you which one each product came from:

  • Shopify (Shopify): connect your store (see Shopify setup). The full catalog is pulled in when you connect, then kept current by Shopify's product webhooks.
  • Shopify or WooCommerce public feed (Shopify feed / WooCommerce): picked up automatically when you crawl your store, with no connection at all. This is how WooCommerce stores are read today: there is no WooCommerce product connection to make.
  • Product schema on crawled pages (Crawled pages). Any store whose product pages carry JSON-LD, microdata or Open Graph product data.

A connected store and a public feed each stop at 2,000 products per workspace; connecting the store does not raise that limit. Products read from crawled pages have no per-workspace ceiling, but each page contributes at most 50. If your catalog is larger than 2,000, talk to us before you rely on this.

Open Knowledge › Products to see exactly which products your agent can see, what each one costs, whether it is in stock, and which source it came from. If a product is missing there, the agent cannot recommend it.

Keeping it up to date

  • Connected Shopify store: the full catalog is pulled in when you connect, and Shopify's product webhooks keep it current from then on. There is no Resync button today: if you ever need a full re-read (for example after the webhooks were interrupted), reconnect the store from the App Directory and the catalog is pulled again.
  • Public feed or product schema: refreshed the next time AskVault crawls your site.

Example

An illustration of the shape of the answer, not a real store:

Visitor: do you have anything like a travel mug under 30 dollars?

Agent: Yes: two options under $30:

[card] Commuter Mug 350ml: USD 24.00, In stock, View product [card] Trail Tumbler: USD 28.50, In stock, View product

Limits and behaviour

BehaviourDetail
Results per answer1–6 (3 by default)
Budget filterThe agent passes the visitor's stated budget; hides products whose lowest price is above it
DuplicatesCollapsed when two chunks describe the same product
No matchThe agent says the catalog has no match: no card is shown
PricesShown in your store's default currency; a product with no price shows no price
ImagesShown when your catalog has one; otherwise the card renders without a thumbnail
ChannelsCarousel on the web widget, Block Kit cards in Slack, text on WhatsApp and Telegram
Catalog ceiling2,000 products per workspace from a connected store or public feed; crawled pages are uncapped per workspace (50 per page)

Turning it off

Open the Skills page and switch Product Recommendations off for that workspace. The agent falls back to answering product questions from your regular knowledge base content. Each workspace is independent, so turning it off in one leaves the others alone.

Was this page helpful?