Xocoa Agent

Your catalogue,
with a sommelier.

Paste your store URL. The Agent reads every product you sell, works out how each one actually tastes, and puts a conversational sommelier on your storefront — one that recommends from your shelves and nobody else's.

Works with Shopify and WooCommerce directly, plus most storefronts that publish structured product data. You'll review the catalogue before anything goes live.

From URL to sommelier

The whole thing runs in the background. You paste a link and come back to a catalogue that has been read, tasted and quality-checked.

It reads your storefront

The Agent finds your products the way a careful shopper would — structured data first, then the page itself. No feed to export, no plugin to install, no developer time.

Shopify · WooCommerce · storefronts with structured product data

It learns how each product tastes

Every item gets a sensory profile on the scales a professional would use — sweetness, acidity, body, intensity, aroma families, what it pairs with, when you'd reach for it. Price, origin and name are never touched: those come from your catalogue, verbatim.

WSET-aligned scales · category-aware · your facts stay yours

It checks its own work before you ship

A quality gate runs before anything is published: coverage across the catalogue, sample recommendations that have to return real products, and a grounding audit. Fail any of it and the store waits for you instead of going live.

coverage · sample recommendations · grounding audit

You paste one line into your theme

One script tag. The sommelier appears on your storefront, answers in English, French or Spanish, and links recommendations back to the product page on your site.

one <script> tag · no theme rebuild · removable any time

It cannot recommend what you don't sell

The failure mode everyone fears from an AI on a storefront is the confident invention — a product that doesn't exist, a price that isn't real. That is designed out, not prompted away.

Every recommendation resolves to a real row in your catalogue. Candidates are selected by deterministic filters before any language model is involved.
Identity facts are fenced off. Name, price, currency, origin and product link are written from your storefront data and are structurally impossible for the model to overwrite.
Hard constraints are honoured. A budget or a dietary requirement filters the shelf before ranking — it is never a suggestion the model can talk past.
Answers are audited after generation. A grounding pass scans for claims that aren't supported by your catalogue and repairs them before the shopper sees anything.

“Something for a cold evening, under £30.”

→ filtered to what's in stock under £30, ranked by how well it actually matches, explained in a sentence a shopkeeper would recognise, linked to the product page.

One line, then it's live

Your console generates this for you, with your store's key already filled in.

Paste before `</body>`

Nothing else changes on your site. Remove the line and the sommelier is gone.

<script src="https://xocoa.co/widget.js"
        data-xocoa-store="your-store.com"
        data-xocoa-key="xocoa_live_…" async></script>

Try it on your own catalogue

Build it, read what it made of your products, and decide whether it goes live. Nothing publishes without you.

Open the console →

Your catalogue,with a sommelier.