On this page

• Why use this
• How it works
• Product cards in chat
• Step-by-step setup
• Why prices stay correct (currency-lock and geo-proxy)
• Limits and behaviour
• Common errors and troubleshooting

Why use this

• You run a PrestaShop store and want your assistant to answer with exact price and stock, not just the product description text.
• You want the chat to render product cards with image, name and price, linking straight to the product page.
• You want to import the whole store in one step — the Webshop source processes every page in the sitemap, not just products.
• On an unsupported engine? Use the Website source instead — it also crawls the catalog, but without structured price/stock or product cards.

How it works

The crawl behaves like the Website source — it starts from the sitemap — but it processes each page based on whether it's a product page or not.

1. Discover. EchoAI reads the store's sitemap and collects every URL. CDATA-wrapped URLs (common on PrestaShop) are parsed correctly.
2. Classify. Depending on the detected platform, each URL is classified as a product page or another page (category, blog, static). Shopify and WooCommerce use the public product listing (JSON API); the other platforms use the sitemap's product links; on UNAS the product links are collected from each category page's product list.
3. Process a product page. EchoAI fetches the page, extracts the product data (name, price, currency, stock, SKU, brand, category, attributes, images), and writes two things:
   • a structured record with the exact price and stock — this powers the product cards and exact pricing in chat;
   • a content item (RAG corpus) linked 1:1 to that structured record — this counts toward your content limit and makes the product searchable by text (attributes like "Colour: Natural" are included in the searchable body).
4. Process other pages. Non-product pages become a plain content item only, without price/stock — exactly like the Website source.

The crawl runs in the background and processes pages concurrently. It is idempotent: re-syncing skips already-imported pages, so no duplicates are created and your content limit isn't wasted.

Product cards in chat

When a visitor asks about products, the assistant doesn't just reply with text: it renders product cards in a two-column grid (up to 6 cards) inside the conversation. Each card is clickable and opens the product page in a new tab.

A card shows:

• Product image (if available) — dimmed for out-of-stock products.
• Product name.
• Price — formatted for the conversation language (thousands separator, currency).
• Stock status — a coloured dot plus a label (green = in stock, gray = out of stock). The label is localized (e.g. "In stock" / "Out of stock"); the exact quantity is never shown, only the binary state.

Card order is gently re-ranked so the most useful products surface first:

• first the in-stock products,
• then those that have a price,
• finally those that have an image.

The search relevance order is preserved within products of equal score. Out-of-stock products don't disappear — they're just shown more dimly and in gray, so they stay informative without being alarming.

Step-by-step setup

1. Open Integrations (content sources) in the dashboard.
2. Find the Webshop product data card.
3. Enter your webshop URL. For a multi-locale store you can narrow to one language with a path prefix (e.g. https://yourshop.com/en).
4. Select which assistant should receive the imported products.
5. Click Connect webshop. EchoAI starts the crawl, and you can track status on the card: connecting → crawling → active.

When the crawl finishes, the card shows how many products were synced. You can refresh prices and stock later by re-syncing — the idempotent crawl only picks up what's new.

Why prices stay correct (currency-lock and geo-proxy)

PrestaShop can decide which currency to render based on the visitor's IP country. That means if the crawler "looks at" the store from another country, the price can come back in the wrong currency — and the correct one is physically absent from the fetched bytes. EchoAI guards against this with two mechanisms.

Currency-lock

On the first crawl, EchoAI detects the currency from the store homepage and locks it on the source. From then on, the crawler validates every product against that locked currency:

• A product in the correct currency → imported.
• A product in the wrong currency → rejected, written neither as a structured record nor as content. Any existing, correct data is left intact — "stale but correct" always beats "fresh but wrong".

Geo-proxy

So the store renders the correct currency in the first place, EchoAI crawls geo-gated stores through a residential proxy in the store's country. EchoAI learns the connecting admin's country from the connection and stores it on the source, so the crawl "looks into" the store from there.

Limits and behaviour

Common errors and troubleshooting