Google Cloud has put Data Cloud connectors for Gemini for business into preview, letting users ask questions of connected enterprise data. Google says queries use the person's existing permissions; the scope of those permissions matters.
The big change
- Gemini uses catalog context to identify data and generate a query under the user's identity.
- Federated mode queries data in place instead of copying it to the Gemini data store.
- A connector is not automatically restricted to one table; it can reach resources the user can query.
From a question to data
In Google's November 6 example, a user asks about monthly sales of cold-weather products, including coats. Knowledge Catalog maps business terms to database structures before Gemini executes SQL as the user.
Google says BigQuery, Cloud SQL and Looker are in preview. Administrators configure connections, users authorize them, and row- and column-level permissions apply.
The scope of access
The developer guide distinguishes ingestion, which copies data into a Gemini data store, from federated queries against the source. In federated mode, the connector authenticates as the user and retrieves accessible catalog context.
A connector can query all user-permitted resources in its source, not one table.
For deterministic restriction to selected data, the guide points to a scoped conversational analytics agent. It also warns accuracy can vary on complex relational schemas without table scoping or verified queries. Organizations would need to check answers against their own definitions and data before relying on them; BIG CHANGE has not tested the preview.
