CurioSetupInboxInsightsFeedbackTrustCatalog gapsAssistantCampaignsBillingProductsKnowledgeTest chatDemo storefront

Sample data — this is not a real store

No Shopify store is connected to this session. Every number on this page — conversations, product counts, ratings — comes from a seeded demo catalog of 65 snowboards, not from a merchant's shop. Open the app from your Shopify admin to see your own store. Treat this store's chat as public: don't enter real personal details.

Curio Test Store

demo.curio.local

Test the assistantOpen inbox

Estimated OpenAI cost (not a bill)

Mid-case reference from the published cost model: about $2.30 per 100 engaged sessions (gpt-5.4-mini defaults). This week (directional): $1.01 (~220 AI turns). These are estimates for planning, not an invoice, and no API keys are shown here.

Waiting on you

69

Conversations

279

Products the assistant can see

59

Trust

Open trust

Curio answers from your catalog and policies. It won't invent prices, stock, or promises. Score uses thumbs and grounded source signals from the last 7 days.

Trust score

35

Thumbs-up

n/a

Source hits

30%

Last 7 days

Demo sample data

Opens

0

Messages

0

Card clicks

0

Handoffs

11

Thumbs down

0

Open feedbackOpen inbox

Setup

demo.curio.local

Done

Catalog synced

59 in-stock products

Products
Done

Indexed for search

76 chunks searchable

Reindex
Done

Policies added

10 knowledge documents

Knowledge
Done

Live on the storefront

Demo store — nothing to install

Setup

Human handoff

When a shopper asks for a person, the assistant stops answering that conversation and hands it to you. It does not resume until you hand it back.

51 waiting for a replyNeeds you

What shoppers rated badly

Every thumbs-down a shopper leaves is a real answer that missed. It is the cheapest list of things to fix you will get.

Review 2 flagged replies2

How the assistant answers

Every reply is grounded in this store's own data. Products and knowledge documents are indexed with both vector and keyword search, the closest matches are retrieved per question, and the model is given those and told to invent nothing outside them. Edit the catalog or the knowledge base and the answers change with it.

Try it against this storeTone, greeting and branding