When you approve a semantic model design in a Warehouse Builder chat, the flow continues with two cards in the chat: first the access check for the model's owner role, then the created model and its handoff to the Model AI. This page covers both.
The design card itself is on The semantic model design card, and the builder chat around it is on Warehouse Builder and Lake Builder.
After the design is approved, the check starts in the same chat. A card titled Grant role access lists every object the model needs, with the instruction:
"Open each object below, add role under Permission Groups, then Save. Tell me when you are done."
The card counts the progress as "n of m granted", and every object carries a state:
When everything is granted, the card finishes: the helper text disappears and the title reads "role can read every object". For the admin role the card finishes straight away, and the done title reads "admin reaches every object".
The card's corner Open details button opens the full picture in the chat's Tool History panel: one row per object with its state, the missing columns, the personal data it would show in clear, and its own row-level security.
An object can carry warning lines that name exactly what the grant would expose:
When a column is not marked as personal data, the model cannot be created until the Warehouse Builder marks it. Send the builder a message in the same chat, then select I've granted them, check again.
When the check passes, the model is created and its card arrives in the chat, headed Semantic model created, with chips for the model, its Owner role and its data mode (tokenized or clear values).
When the Model Builder sends the brief itself, the send button and the whole What the Model AI will receive section are hidden, and the card notes: "The Model Builder sends the brief and builds this model with its Model AI." The builder's progress cards appear in the chat, and when its report from the Model AI finishes, that card refreshes itself.
If the personal data check flags the brief, the Model AI panel shows what it found before anything is sent: "If its personal data check flags the brief, the Model AI panel shows what it found before anything is sent."
The model itself is described in Semantic models and BI tables, and the assistant panel beside it is Model AI.
The model reads the objects its design lists. Before it is created, the platform checks that its owner role can read every one of them, and the card shows what is missing and how to grant it. The check follows the design you approve on The semantic model design card.
Select Open permissions on an object that is not granted, add the owner role under Permission Groups in the Warehouse popup and save it, then select I've granted them, check again so the check runs again. Repeat until the card counts everything as granted.
The chat is busy: an answer is streaming, or you are typing. Wait for the chat to be free, then select it again.
The role can read the object, but not every column the model needs. The row names the columns, as "Columns missing: columns". Grant the missing columns on the object, then select I've granted them, check again.
The check could not use that object: it is not deployed in the Warehouse yet, or it could not be found. Only an object that was found and is missing access offers Open permissions; once the object is in place, select I've granted them, check again.
No. The admin role reaches every object, so the card finishes straight away, and its done title reads "admin reaches every object".
A column the model needs is not marked as personal data in the Warehouse. The warning reads: "column is not marked as personal data: reports would show it in clear too. The Warehouse Builder must mark it before the model can be created." Once the builder marks the column, select I've granted them, check again.
Empty, in the requester's personal workspace: "Created empty in the requester's personal workspace; move it later with Move."
It sends the build brief into the model's Model AI chat, and the model opens in DataPrism, so the Model AI can work from the brief. Open Model AI chat opens the model without sending the brief.
The Model Builder is sending the brief itself. The send button and the What the Model AI will receive expander are hidden then, and the card says "The Model Builder sends the brief and builds this model with its Model AI."
Yes. Open What the Model AI will receive on the card to read it, and Copy the brief to keep a copy. When the Model Builder sends the brief itself, the expander is not shown.