Google launches QueryData

Google Cloud introduced QueryData, a service that helps AI agents generate, validate and execute SQL queries reliably across AlloyDB, Cloud SQL for MySQL, Cloud SQL for PostgreSQL and Spanner. The product separates model interpretation from deterministic execution so a constrained layer can enforce schema, permissions and validation before any query runs, aiming to reduce risky agent touches on structured data. (infoworld.com)

Google Cloud has launched QueryData in preview, a service that turns plain-language requests into database queries for artificial intelligence agents. (cloud.google.com) The product works with AlloyDB, Cloud SQL for MySQL, Cloud SQL for PostgreSQL, and Spanner for GoogleSQL, according to Google’s April 10 announcement. Google said QueryData is aimed at “agentic experiences” that need to read operational data such as pricing, inventory, and transaction records. (cloud.google.com) Text-to-SQL systems take a natural-language question and write Structured Query Language, the code databases use to fetch records. The problem is that a model can write valid code that still pulls the wrong rows, misses business rules, or asks for data a user should not see. (bird-bench.github.io) (cloud.google.com) Google’s pitch is that QueryData splits those jobs apart. A model interprets the request, then a constrained layer applies schema knowledge, validation, and access controls before anything runs against the database. (infoworld.com) Google said developers feed the system “context sets,” which are files describing database structure, business logic, and example queries. The company’s documentation says teams can author those files with the Gemini command-line interface and upload them in the Google Cloud console. (docs.cloud.google.com) The documentation lists customer service, shopping assistants, and field operations as target uses, with examples like “Where is my order?” and “Show me running shoes under $100.” Google also said a Hughes Network Systems deployment is already using QueryData in production. (docs.cloud.google.com) (cloud.google.com) Google tied the launch to its performance on the BiRD benchmark, a widely cited test for turning natural language into database queries. The benchmark’s creators describe BiRD as a large text-to-SQL dataset with 12,751 question-and-query pairs across 95 databases and 37 domains. (cloud.google.com) (arxiv.org) The release is still pre-general availability, which means Google labels it a preview and says it may have limited support under its service terms. For now, the company is selling QueryData as a way to let agents touch structured data with fewer mistakes and tighter controls. (docs.cloud.google.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.