Google QueryData adds control

Google Cloud put QueryData into preview on April 10, saying the service turns plain-English questions into database queries for AlloyDB, Cloud SQL, and Spanner with “near-100%” accuracy. (cloud.google.com) A database query is the line of code that asks a database for records, totals, or trends. QueryData is meant to help an artificial intelligence agent write that code without guessing at table names, joins, or business rules. (cloud.google.com) Google’s system does not rely on the model alone. It uses “context sets,” which are curated files and database-stored instructions that describe schemas, common query patterns, and the meaning of fields before the model generates Structured Query Language, or SQL. (docs.cloud.google.com) Google’s documentation says teams can define that context for AlloyDB, GoogleSQL for Spanner, Cloud SQL for MySQL, and Cloud SQL for PostgreSQL, then reference it in a QueryData call. The company also published tutorials that walk developers through building a context file with Gemini Command Line Interface and Model Context Protocol Toolbox. (docs.cloud.google.com; docs.cloud.google.com) That design shifts work to the start of a project. Instead of letting an agent infer how a database works, teams have to map the data model, write templates and facets in JavaScript Object Notation, and upload the context before the system is likely to perform well. (docs.cloud.google.com; docs.cloud.google.com) Google argues that extra setup is the point. In a second April 10 post, the company said enterprise database agents need accuracy and explainability because bad queries can create remediation costs, lost trust, and legal risk. (cloud.google.com) InfoWorld reported on April 13 that analysts see the tradeoff clearly: safer, more deterministic query generation in exchange for more implementation and configuration work up front. The publication said that requirement could slow early adoption for teams that want fast proofs of concept. (infoworld.com) The launch also fits into Google Cloud’s broader push around data agents, which are software agents that answer questions or take actions over business data. Google’s Conversational Analytics architecture now lists QueryData as the method agents use to query supported operational databases. (docs.cloud.google.com) For customers, the practical question is not whether an agent can write SQL once. It is whether the agent can keep writing the right SQL against production databases after a team has encoded enough context to constrain it. (cloud.google.com; infoworld.com)