Help

Ask it something about your own catalog: "how many products have no meta description", "which variants have a compare-at price below the price". That is a read, it needs no approval, and it tells you more about what the tool sees than any demo can.

That is the intended behaviour. A change is not written when you ask for it; it becomes a plan with one row per change, and it waits. Look for the plan in the chat, read the rows, then approve. Until you do, the store is untouched.

Scope and rule in one sentence, the way you would say it to a colleague: "raise prices by 10% for every XL variant and round to .99". No syntax and no filter builder. If the scope is ambiguous, the plan will show a count you did not expect, which is the signal to decline and say it more narrowly.

Read and write on products, plus whatever else you want it to change, such as inventory or collections. You approve the scopes at install. A field outside the scopes you granted is not offered in a plan at all, so if something you expect is missing, the scope is usually why.

The scope was read more narrowly than you meant, or rows were excluded on purpose. A row that is skipped appears in the skipped list with its reason, for example an existing value when the instruction said "that has none". Decline, widen the sentence, and look at the count again before approving.

The count in the first line of the plan is the scope, resolved. If it is bigger than the set you had in mind, the sentence matched more than you meant. Decline. Nothing is written by a plan you did not approve, so an inspection costs you nothing but the read.

Open the job in history and revert it. The value each row had before is saved with the job, so revert builds a new job that writes those values back, only to the rows the first job changed. It works for supported updates; where a step cannot be put back, the case page for that job says so.

Revert works from the history entry, not from a position in a queue, so an older job is reverted the same way as the most recent one. What it writes back are the values as they were when that job ran. If you have changed the same rows since, say so in an email before you revert and we will look at the history with you.

Skipped rows are listed with the reason, and they are not failures: usually the row did not match the rule, or the field was outside the granted scopes. The rows that did match were written and are in history as one job, so they can be reverted together.

100 AI credits a month, the preview on every change, small bulk jobs and basic history. It resets monthly and it is not a trial that expires. It is enough to run a real job on your own catalog and decide from the result.

Credits are what the AI work costs, so a bigger catalog and a longer job use more of them. The free plan resets each month; a paid plan raises both the allowance and the job size. Basic is $19/mo, Pro $49, Ultimate $99. If the shape of your usage does not fit any of them, write to us.

Through Shopify, on your existing Shopify bill. We never see or store a card. Plan changes go through the same approval screen Shopify shows for any app charge.

That is the Agency plan, from $299/mo: several stores under one agreement, with credits scoped to the catalogs you actually work on instead of a fixed tier. Every store keeps its own preview, approval and history. Tell us how many stores and roughly how many products each, and you get a number back.

Your catalog is never used to train a model, and the writes we made are Shopify changes you approved on rows you saw. What we keep, for how long, and how to ask for deletion are set out on the privacy page.