Gemini Enterprise adds live queries across cloud databases
Google previews database connectors with broad user access and separate controls for stored results and write actions
Google announced preview federated queries in Gemini Enterprise on October 2. The connectors cover BigQuery, Spanner, Cloud SQL and AlloyDB for PostgreSQL. They let the assistant query databases directly through each user’s existing access.
Access follows the person asking
The connector guide describes requests through Model Context Protocol servers without copying the source data into an index. Access extends across resources the user can query. Attaching a connector does not restrict it to one table or database.
Knowledge Catalog is enabled automatically when a federated connector is attached. Its read only tools find accessible assets and retrieve schemas and business context. Administrators can control that integration separately from the connector.
Results still need protection
Google’s security documentation says results return to the agent in the Gemini Enterprise app’s location and can be stored in conversations. Keeping source tables in place therefore leaves teams with a separate decision about conversation access and retention.
If connector write actions are enabled, the agent can also modify the underlying data within the user’s permissions. Google documents encryption keys, regional controls and service perimeters, but these require appropriate configuration. Catalog discovery being read only does not make every connector operation read only.
Test accuracy and costs before expanding access
The Cloud SQL guide warns that broad queries can struggle with complex transactional schemas. For tightly scoped, high accuracy work, Google recommends a curated data agent with selected tables and verified SQL. The preview may have limited support.
The BigQuery setup guide also makes billing explicit. Queries consume BigQuery compute, and administrators can specify a billing project. Avoiding a copied index does not remove the cost of running queries.
Google’s implementation guidance emphasizes business definitions and curated metadata alongside built in analytical skills. A familiar term such as revenue needs an agreed calculation and an authoritative dataset before an answer can be useful.
Start with questions whose answers your team already knows. Inspect the generated SQL, compare the results and measure query spending. Keep write actions disabled during that first evaluation and review who can open the resulting conversations.
This adds a database workflow inside Gemini Enterprise. Our earlier Data Agent Kit report covers the separate toolkit for coding assistants, including its setup and permission requirements.
ByteForward reviewed the public documentation. We have not run these connectors or independently verified their accuracy, access enforcement or costs.
Archival NOIRLab server racks photograph credited to NOIRLab/NSF/AURA/P. Horálek under Creative Commons Attribution 4.0. The unchanged image illustrates database infrastructure and does not depict Google facilities or this software.
