AlloyDB AI natural language overview

This page describes AlloyDB AI natural language, which lets you create user-facing generative AI applications using natural language to query databases.

AlloyDB AI natural language generate context-aware SQL queries for application end users' natural language questions, and it provides a security API to execute these queries. Natural language streamlines the development of generative AI applications by transferring the complexity of converting natural language to SQL from the application layer to the database layer. Natural language-to-SQL (NL2SQL) orchestration is embedded in the database. You can query structured and unstructured data.

AlloyDB AI natural language also provides an intuitive interface to SQL developers that translates natural language to SQL, as shown in the following example:

SELECT
  alloydb_ai_nl.get_sql(
    ...,
    'What is the sum that client number 4''s account has following transaction 851?'
  );

AlloyDB AI natural language support is provided through alloydb_ai_nl, which is an AlloyDB for PostgreSQL extension. The extension has a dependency on PostgreSQL extensions like google_ml_integration, vector, and pg_trgm.

Use cases

AlloyDB AI natural language is well-suited for developers who want to do the following:

  • Rapidly build a customer chat interface by using query generation from natural language.
  • Refine automatically generated queries and perform thorough testing to provide accurate responses.
  • Tune generated results by adding templates and context information.
  • Deploy an intuitive, accurate chat experience that lets customers ask natural language questions about their relational data.

Key capabilities

Key capability Description
Semantic layer
  • Maintains rich schema and application context for enhanced accuracy.
  • Stores the meaning of your database—schemas, tables, columns, and relationships.
  • Enables AlloyDB to understand the context of user questions.
Concept search
  • Intelligently parses user questions to identify key value phrases.
  • Maps these values to your data with disambiguation based on concept types, improving accuracy.
Query store
  • Supports accurate query generation for a wide variety of common natural language questions.
  • Uses templates and query fragments (manual and auto generated) for faster, more accurate SQL construction.

Limitations

You must enable the AlloyDB AI natural language flag on every instance of AlloyDB for PostgreSQL. AlloyDB AI natural language objects created on the primary instance are propagated to read-only replicas and cross-region replicas. However, the alloydb_ai_nl.enabled flag setting isn't replicated automatically, so you must replicate it manually on each instance. For more information, see Generate SQL queries that answer natural language questions.

What's next