In progress
Natural language querying inside your product
Your customers ask in their own words and get an answer without leaving your application. The question resolves against the model you already maintain, not against a separate copy of your data built for the assistant.
One semantic model, whether the question is typed or clicked
The AI assistant in the SQL editor shipped in January 2026, aimed at whoever writes the query. The next step points the same assistant at your customer’s plain question. One definition of a metric, two ways to reach it, rather than a second set of numbers reachable only by asking.
Permissions resolve before the model answers
An AI answer crosses a tenant boundary exactly the way a cached number does. The assistant's query goes through the same permission resolution as every other query. Rows outside the tenant filter are not returned, so they cannot reach the answer.
Every answer traces back to a query
An answer resolves to a query you could have written yourself, against the sources Sumboard already reads. Sumboard keeps no copy of your data beyond cached results, so there is no second store for an assistant to reach. You can ask where a number came from and get a query back.
