Create a view with Ask AI

On a project or workspace canvas you can start a view from a sentence instead of a right-click. Ask AI resolves the tables from your catalog, asks the clarifying questions that matter, and proposes a view. Nothing is created until you open the proposal in the view designer and choose Save.

Ask for the view

Open Ask AI while you are on the canvas and describe the view you need, for example: “I need a view for the customer shipping address combining Address and CustomerAddress where the address type is Shipping, as of now.”

The assistant resolves each table name against your catalog. If more than one table matches a name, it asks which one you meant before going further. It never invents tables.

Answer the clarifications

When your organization’s methodology includes a Data Vault layer and the request describes attributes of a business concept, the assistant asks whether to shape the view like a satellite of the matching hub. Accepting adds the hub’s business key column and a derived hash key column whose expression is computed on the server with the same hashing settings your staging pipeline uses, so it matches the real hub byte for byte.

If no business key can be found for the concept, the assistant says so and offers a plain view instead. It never guesses a business key.

Ask for a dimension or a fact

When your organization models with Data Vault, describe the dimension or fact you want and Ask AI grounds the proposal on the vault instead of guessing at a shape.

Ask for something like “a customer dimension” or “an orders fact”. The assistant resolves the business concept against your hubs and links: a dimension anchors on the matching hub, a fact anchors on the matching link, and every satellite carrying that concept’s attributes is added automatically. You don’t pick the anchor or the satellites by hand.

The proposal takes its name from your organization’s dimension-view and fact-view naming pattern, for example dim_customer for a customer dimension or fact_orders for an orders fact, and sets the target layer to Data Mart. It also derives the hash-key point-in-time joins that pull each satellite’s current attributes as of now. See Point in time tables for what those joins emit and how ghost records keep them matching as inner joins.

With the anchor, the satellites, the name, the layer, and the joins already decided, the remaining work is column selection. Choose Open in view designer and you land in the same designer described in Create a view from the canvas, pre-populated end to end: pick the columns you want from the anchor and its satellites, then continue as normal.

Review the proposal

The proposal appears as a card in the chat thread showing the view name, target layer, the tables, the join summary, the filter, the as-of setting, any added key columns, and the assistant’s reasoning. Two buttons:

• Open in view designer validates the proposal and opens the designer pre-populated.
• Discard cancels the proposal.

If the tables could not be joined automatically, the card says so; the designer still opens and you connect the tables on the canvas there.

Refine and create

The view designer opens with the name, layer, as-of value, filter conditions, and derived key columns already in place. Everything from here is the normal designer flow described in Create a view from the canvas: pick more columns, adjust joins, preview the SQL, and choose Save. The AI never saves anything on its own.

Troubleshooting

• The Open in view designer button is disabled. The proposal was not made on a canvas, so there is nowhere to open it: ask again from a project or workspace canvas. A proposal made on a canvas you have since left stays enabled, and choosing it takes you back to that canvas first.
• The proposal fails when you open it. The tables were validated when you accepted; if something changed in between (for example a table was removed on your branch), the card shows the validation error and you can ask the assistant to propose again.
• Ask AI does not offer view creation. The tools are only available while you are on a project or workspace canvas; other pages do not offer them.