Google launches QueryData

Google Cloud has launched QueryData in preview, a service that sits between artificial intelligence agents and databases to make generated queries more predictable. (cloud.google.com) The product translates plain-language questions into database queries for 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 agents that need live business data such as pricing, inventory, balances, or transaction records. (cloud.google.com) A database query is the instruction that asks a system for specific records, and QueryData tries to keep that instruction tied to fixed rules instead of letting a model improvise. Google’s documentation says developers feed the service “context sets,” files that describe schema, relationships, and business logic so the system can map a question to an approved query pattern. (docs.cloud.google.com) Google said the service is meant to solve three problems at once: accuracy, security, and setup. In its launch post, the company said ordinary large language models can write query code but still miss table meaning, shorthand field names, or access rules inside enterprise databases. (cloud.google.com) The release lands as cloud vendors push “agents” from chat interfaces into systems that can look up records and trigger actions. Google’s separate Conversational Analytics application programming interface now supports AlloyDB, Spanner, and Cloud SQL through a new QueryData method, extending the same agent stack beyond BigQuery and Looker tools. (docs.cloud.google.com) Google is also threading the feature into its database products. Cloud SQL release notes on April 6 said context sets, previously called data agents, are in preview for MySQL and PostgreSQL and now improve structured query language generation with automatic “value search” that matches values and their context inside a database. (docs.cloud.google.com) Google’s pitch is reliability. The company said QueryData can deliver “near-100% accuracy,” and cited Hughes Network Systems as a production user that said the tool sits at the center of its support-operations agent system. (cloud.google.com) Analysts say the tradeoff is more engineering work before an agent ever runs. InfoWorld reported that teams must explicitly define schema details, access patterns, and deterministic instructions, and analyst Pareekh Jain said that creates an ongoing maintenance job as databases change. (infoworld.com) Google’s own documentation adds another limit: QueryData is still a Pre-General Availability preview, which Google says is offered “as is” and may have limited support. That leaves the product positioned less as a plug-and-play chatbot and more as a controlled execution layer for companies willing to tune how agents touch live data. (docs.cloud.google.com)