Google Cloud adds QueryData

Google Cloud has launched QueryData, a preview tool that turns plain-language requests into database queries for artificial intelligence agents with what the company says is “near-100% accuracy.” (cloud.google.com) The product went live on April 10, 2026, and Google says it works with AlloyDB, Cloud SQL, and Spanner, including Cloud SQL for MySQL and PostgreSQL deployments. (cloud.google.com) A database query is the instruction that tells a database what to fetch or change, and QueryData is meant to stop large language models from improvising those instructions on their own. Google says standard model-generated Structured Query Language can fail because models do not know a company’s schema, business rules, or access controls. (cloud.google.com) Google’s pitch is that agents now do more than answer questions: they trigger business actions tied to pricing, inventory, balances, and transactions. In a second April 10 post, Google argued that even 90% accuracy breaks down in multi-step workflows, where five dependent steps would succeed only about 59% of the time. (cloud.google.com) QueryData works by making developers prewrite “context” for the model, which functions like a map and rulebook for a specific database. Google’s documentation says that context sets contain code and JSON objects describing schema, business logic, query templates, facets, and value searches. (docs.cloud.google.com) Google says developers can build those context files with Gemini Command Line Interface tools, upload them in Google Cloud, and then call QueryData through an application or through Model Context Protocol Toolbox integrations. A Cloud SQL tutorial also says users need specific Identity and Access Management roles and read-only database privileges before testing the feature. (docs.cloud.google.com) InfoWorld reported on April 13 that analysts see a tradeoff in that design. Pareekh Jain of Pareekh Consulting said QueryData can reduce prompt engineering and improve runtime reliability, but it also creates “a new workload category” because teams must keep schema definitions and deterministic instructions updated as databases change. (infoworld.com) That tradeoff fits Google Cloud’s broader database-and-agents push. In February 2025, Google introduced its open-source Generative Artificial Intelligence Toolbox for Databases as middleware between agent frameworks and databases; QueryData now adds a more tightly controlled query-generation layer on top of that stack. (infoworld.com) For now, Google is selling QueryData as a way to make agents safer around production data without leaving query generation entirely to a model’s guesswork. The catch is that the more deterministic the system becomes, the more design work shifts to the developers wiring it into real databases. (cloud.google.com)