Google’s QueryData launch

Asking a database a question in plain English is easy; getting an artificial intelligence agent to write the right query is harder. Google Cloud said on April 10 that its new QueryData service is meant to close that gap. (cloud.google.com) QueryData is in preview, and Google says it translates natural-language requests into database queries with “near-100% accuracy.” The launch covers AlloyDB, Cloud SQL for MySQL and PostgreSQL, and Spanner for GoogleSQL. (cloud.google.com) The basic problem is that database queries are brittle: one wrong table, join, or filter can return the wrong answer. QueryData adds a layer between the user’s request and the database by using “context sets,” which are code-based instructions about how data should be accessed and queried. (cloud.google.com) (docs.cloud.google.com) Google is pitching that control layer at companies that want agents to do more than chat. In BigQuery, Google’s data platform, data agents already let users ask questions in natural language, and Google now documents how teams can define context and instructions for those agents. (docs.cloud.google.com) (cloud.google.com) That matters because companies are trying to connect large language models to live business data without letting them improvise against production systems. Google said QueryData can run inside BigQuery data agents or through application programming interface calls in custom workflows, giving enterprises a way to keep query generation and execution inside governed systems. (infoworld.com) (cloud.google.com) Google tied the product to a broader push around “data agents,” software that can answer questions, run analysis, and trigger actions across enterprise data. In recent months, the company has also promoted a Conversational Analytics application programming interface for BigQuery and a BigQuery AI package that groups machine learning, generative artificial intelligence functions, vector search, and agent tools. (cloud.google.com 1) (cloud.google.com 2) Google said Hughes Network Systems has already deployed QueryData in production. The company also said the tool builds on Google Cloud’s top ranking in the BiRD benchmark, a widely watched test for turning natural language into structured query language. (cloud.google.com) Analysts quoted by InfoWorld said the approach could help in regulated or mission-critical settings, but they also said the deterministic controls require more upfront design work from enterprise teams. That trade-off is the point of the launch: less freedom for the model, more structure around the data. (infoworld.com)