AI Shopping Assistant

STOAR includes a built-in AI chat assistant on the storefront. A floating chat button sits on every page; shoppers open it, ask a question in plain language, and get an answer grounded in your live store data — your catalogue, your policies, your prices. It never makes up products, prices, stock, or policies.

The assistant runs on Grok (xAI) by default, or OpenAI — switch with a single setting. Both are used through the same interface, so swapping providers changes nothing else.

What It Can Answer #

The assistant looks up real data on demand rather than guessing, so it handles the everyday questions a shopper actually asks:

• Find products — "Do you have a waterproof jacket?", "Show me blue t-shirts under $30." Matches are pulled from your catalogue via Meilisearch and rendered as clickable links straight into the chat.
• Catalogue questions — "How many t-shirts do you have?", "What categories do you sell?" Answered from live counts, not estimates.
• Shipping — options, rates, and zones, read from your configured shipping methods.
• Returns — your return window and policy text.
• Payments — the payment methods you actually have enabled.
• Contact — your store's email, phone, and address.
• Order status — signed-in customers can check their own orders, including tracking, without leaving the chat.

If nothing relevant is found, the assistant says so plainly. Off-topic, general-knowledge questions are politely declined — it stays a shopping assistant.

How It Stays Accurate #

The assistant answers using tool calling: instead of relying on a language model's memory, it calls live queries against your store (product search, catalogue counts, category lists, store policy, order lookup) and answers only from those results. This is what keeps replies factual — a question like "how many t-shirts" returns a real count, and a shipping question returns your real shipping options.

Product matches are grounded in the same Meilisearch index that powers storefront search, so the assistant's suggestions track your live catalogue as it changes.

Privacy & Data #

• Order status is strictly personal. Only a signed-in customer can see their own orders; the assistant never accepts another shopper's details and guests are asked to sign in.
• Payment details stay private. The assistant lists the names of enabled payment methods only — never gateway keys, bank details, or credentials.
• Conversations are saved for review. Every chat is stored and viewable in the admin (Admin → Chat Conversations), so you can see what shoppers are asking. Signed-in chats are attached to the customer; guest chats are anonymous.
• Retention is enforced. Old conversations are purged automatically on a schedule — guests on a shorter window than customers — with both windows configurable, so nothing lingers longer than you want.

Managing It #

• Chat Conversations live under Admin → Chat Conversations — a read-only list with full transcripts and a delete action.
• The assistant can be turned off entirely with one setting, which hides the widget and disables the endpoint.
• Rate limiting protects the endpoint from abuse, and each reply is capped so costs stay predictable.

Good to Know #

• The widget appears across all storefront themes.
• Replies include Markdown-style links, rendered as clickable product links in the chat.
• Because answers are grounded in live data, keeping your catalogue, shipping, returns, and payment settings up to date is all that's needed to keep the assistant accurate — there is no separate knowledge base to maintain.