The assistant writes the draft, you keep it or undo it
Drafting a description costs no credits and saves nothing by itself. Here is
In the product editor, beside the description field, there is a generate button. Press it and it writes you an Arabic description from the product's name, its category and its options. This post is not about pressing the button. It is about what happens afterwards, and where it stops.
A draft is a draft. The text that comes back is placed in the field and nowhere else. It is not saved to the product, not published to your storefront, and it touches nothing else. You get three choices: keep, regenerate, or undo. Undo restores the previous text exactly as it was, even after you regenerated twice. And if generation stops before it starts, the field restores itself.
After that you still save the product by hand, like any other edit. So the assistant does not write to your store. It writes into a field in front of you, and saving is your decision.
Why it is built that way. Because the model can invent product facts. A draft you review solves that; text that saves itself lets the problem into your catalogue. The safety here is in the shape of the screen rather than in the wording of the instructions.
What does it cost? Drafting a description and asking the assistant take no credits at all. They carry a limit on how many times per hour per store, and if you reach it you are told so, with the minutes remaining. Images are the part that uses credits, and the credit range shows beside the button before you press. A range, not a fixed price, because the real cost is worked out after generation.
Images, plainly. A generated image is saved to your media library first. The review you have is not whether it enters the library, it is whether it goes onto the product. Delete what you do not like from the library; nothing appears in your storefront until you put it on a product. That is what happens today, and we would rather describe it as it is.
What the assistant cannot do. The dashboard assistant has no permission to create or edit a product. It has exactly two capabilities that write: setting a stock quantity, and turning a discount on or off. Before either one, a confirmation card appears naming the operation, the records affected, and the old and new value, built from a real read rather than from the model's words. Decline, and nothing happens.
And the record. Every time you ask the assistant, a line is recorded in the audit log with your name and the time. The line records that the turn happened and what it cost, not the text of the conversation. And your store's data stays yours: the assistant does not take a store id from the model, it takes it from your session.
Where does it get its information? From the fields in front of you in the product editor and nothing else: the name, the category, the current description, and the options. Not your whole catalogue, and not another store. The request is assembled on our server from those fields rather than from text the browser sends, so there is no opening to stuff extra instructions through.
And the permission it needs. The same permission as products. There is no separate grant called artificial intelligence: whoever can edit a product can generate a description, and whoever cannot does not see the button at all.
What to do with this. Use it for the first draft and keep the review with yourself. If you like the text, save. If not, undo, and you have lost nothing.