Skip to content

CIQRA AI

CIQRA AI answers questions about your store in plain language. The screen states its own scope in one line: “Ask about your store in your own words. It reads your real data and changes nothing on its own.”

Both halves of that sentence matter, and the second half is the one worth reading twice.

This is not a chatbot that has been told about e-commerce in general. It looks at your catalog, your orders, your stock. That is why the suggested questions are the ones the screen offers under “Try one of these”:

  • “Which products are about to run out?”
  • “How do my most recent orders look?”
  • “Give me a quick picture of my store.”
  • “Which products are still drafts?”

What it can look at and do is shown on the same screen, in the What I can do card. The card asks you to pick an example (“Pick an example to put it in the message box. Nothing is sent until you press Ask.”); each example is marked Read only, Needs approval, Prepares a draft or I prepare, you run. The What I can do in this store section counts the operations open to you as I look up, I change with your approval and I prepare, you run.

After each answer it shows what it consulted, under the label “Looked at:” — or, when it answered from general knowledge alone, “Answered without looking anything up”.

🔑 That second label is the useful one. An answer that looked nothing up is an answer about e-commerce, not about your store, and it should be read with exactly that much weight.

When a question implies a change, CIQRA AI does not make it on its own. It shows what it would do, under the heading “It wants to change something”, with the reassurance spelled out: “Nothing changes until you confirm a card below (unless your store lets low-risk changes apply on their own - those say so).”

The change is described concretely — the target, the old value, the new value — and applying it is a separate, deliberate action (Apply this change). An applied change can be reversed with Undo.

The exception is yours to switch on: the Changes without a card card on the screen has Apply low-risk changes without asking. It is off by default and covers only low-risk changes that can be undone with one click. Payments, refunds, team and roles, domains and DNS, keys and credentials, deletions, billing, backups and messages to customers are never applied on their own, whatever this says. The Changes made by CIQRA AI card lists what it changed, with the value it replaced.

This is a deliberate boundary rather than a limitation to be worked around. An assistant that edits your catalog while you are still phrasing the question is an assistant you have to supervise; one that proposes is one you can use quickly.

Questions are asked inside a conversation, and a follow-up is answered with what was said before it. The hint under the question box says which case you are in:

  • “This starts a new conversation.”
  • “Continuing this conversation — it remembers what was said above.”

So “Which products are about to run out?” followed by “and which of those sold best last month?” works the way it would in a chat.

What it remembers is the recent part of the conversation, not all of it: each new question is sent with at most the last 12 messages of the conversation, and at most about 12,000 characters of them. In a long conversation the earliest questions drop out first. When a follow-up depends on something said long ago, repeat it — or start a new conversation.

One exception: on a store where the What I can do card shows “No tools are available to CIQRA AI on this store.”, each question is answered on its own, without the earlier messages.

  • Enter sends; Shift + Enter starts a new line. The hint says so: “Enter to send, Shift + Enter for a new line.” While an input method for Japanese, Chinese or Korean is composing a word, Enter confirms the word and does not send.
  • The open conversation is part of the page address. Reloading the page, using the back button or opening a saved link brings you back to the same conversation.
  • Opening CIQRA AI without a conversation in the address reopens your most recent one, so a reload never hides your history behind an empty box.
  • New chat, above the question box, starts a blank conversation. New conversation in the Conversations list does the same.

The Conversations list keeps every conversation until you delete it. Deleting is permanent — the confirmation says “This cannot be undone.” Conversations belong to the staff account that started them: a colleague signed in to the same store does not see yours, even from a link.

Answers consume AI credits. The number of credits an answer used is shown under it.

A follow-up costs more than the same question asked fresh, because the earlier messages of the conversation are sent with it and are part of what is billed — up to the limit above. Starting a new conversation for an unrelated question keeps that cost down.

The screen also has a label for an answer that was reused rather than generated, “No credits used (already answered)”. Read from the source on 2026-10-10, an answer that consulted your store’s data is always generated and charged; asking the same question again is charged again.

Credits are bought and tracked under Settings → AI, which sits next to Billing on purpose: it is spending, and the admin groups it as spending. See Your account settings.

If the service is unavailable the screen says “CIQRA AI could not be reached just now.” That is a temporary condition on our side, not something wrong with your store or your question — your data is untouched, and nothing was charged.

  • It does not list every kind of question it can answer, or every change it can propose. Both are properties of the model and the tools available to it, and a list printed here would be stale the next time either changes. The What I can do card is the current list for your store.
  • It does not state how many credits a given question costs, or what a credit costs you. The number is shown per answer in the admin and the price belongs to your plan; a figure here would be wrong for someone.
  • It does not say whether the AI’s answers are used to train anything. That is a real and reasonable question, and it has not been measured for this page, so it is not answered here rather than guessed. Ask us and we will tell you.