Google’s QueryData Launch

Databases answer questions by running structured commands, usually Structured Query Language, or SQL, against tables of records. On April 10, Google Cloud said its new QueryData tool can turn plain-English requests into those commands for AlloyDB, Cloud SQL and Spanner. (cloud.google.com) Google said QueryData launched in preview and is aimed at “agentic” software that takes actions after pulling live business data. The company said the tool works with AlloyDB, Cloud SQL for MySQL and PostgreSQL, and Spanner for GoogleSQL. (cloud.google.com) The pitch is accuracy. Google said QueryData delivers “near-100% accuracy” on natural-language database work and builds on Google Cloud’s top ranking on the BiRD benchmark, a public test for turning text into SQL. (cloud.google.com) The underlying problem is that large language models can write valid SQL without understanding a company’s tables, naming shortcuts or business rules. Google’s documentation says QueryData uses “context sets,” which are files of examples, templates and rules that describe how a specific database should be queried. (docs.cloud.google.com) Google says developers create those context files with Gemini Command Line Interface help, then upload them in Google Cloud so QueryData can map a user’s question to the right tables and filters. The documentation lists customer service, shopping assistants and field operations as target uses. (docs.cloud.google.com) Security is part of the design. Google said developers can use Parameterized Secure Views, which pass fixed values such as a user identifier or region separately from the generated query so the agent only sees approved rows. (cloud.google.com) Google had already been moving its database products toward conversational querying before this launch. AlloyDB release notes on April 6 said QueryData entered preview there, and March 30 release notes added conversational analytics powered by the Conversational Analytics Application Programming Interface. (docs.cloud.google.com) One limitation is in the fine print. Spanner documentation says the current examples are for GoogleSQL databases and that QueryData does not support Spanner’s PostgreSQL interface. (docs.cloud.google.com) Google also pointed to an early production user. In the launch post, Hughes Network Systems said QueryData sits “at the heart” of its support system and has delivered “near-100% accuracy in production,” though Google did not publish benchmark details for that deployment. (cloud.google.com) The launch gives Google Cloud a narrower tool for one of the hardest jobs in enterprise artificial intelligence: letting software ask operational databases precise questions without handing the whole task to a general chatbot. For customers, the next test is whether the preview holds up outside Google’s own benchmarks and early user stories. (cloud.google.com)