AlloyDB AI provides an integrated suite of AI and machine learning (ML) capabilities in AlloyDB for PostgreSQL that lets you build generative AI applications using your operational database data. You can run high-speed vector similarity searches, translate natural language questions into SQL queries, and invoke foundation models from Gemini Enterprise Agent Platform, OpenAI, and Anthropic directly from SQL.
To simplify the process of building AI applications, AlloyDB provides the following extensions:
vector extension: the standard
pgvectorPostgreSQL extension is customized for AlloyDB, and referred to asvector. It supports storing generated embeddings in a vector column. The extension also adds support for scalar quantization features to createIVFindexes. You can also create anIVFFlatindex orHNSWindex that are available with stockpgvector.alloydb_scann extension: the
alloydb_scannextension implements a highly efficient nearest-neighbor index powered by the ScaNN algorithm.You can use the
alloydb_scannextension with PostgreSQL 14, 15, 16, 17, and 18.google_columnar_engine extension: ScaNN index can be loaded into the columnar engine for further accelerating the vector search.
google_ml_integration extension: the
google_ml_integrationextension provides AI functions for generating embeddings, semantic ranking, and implementing AI-based filters, joins and text generation/summarization. This extension also provides functions to register metadata for AI models. The registered metadata is then used to invoke predictions from these models.alloydb_ai_nl extension: the
alloydb_ai_nlextension enables developers to build applications that accurately and securely answer end user natural language questions about data in the AlloyDB database. This makes the data accessible to users who might not be proficient in writing SQL. For next-generation text-to-SQL workflows and AI agent integration, we recommend using QueryData.
The following are some use cases that AlloyDB AI and these extensions enable:
Vector search: use AlloyDB to store vector embeddings and perform highly efficient similarity searches. You can generate a highly efficient nearest-neighbor index powered by the ScaNN algorithm.
Perform intelligent SQL queries using AlloyDB AI functions: use AI directly within your SQL queries. This allows you to re-rank search results for higher relevance, integrate natural language into your SQL queries, and generate multimodal embeddings for vector search.
Call models using model endpoints: register AI models as model endpoints and call the endpoints from within AlloyDB to generate embeddings, invoke predictions, or perform similarity searches.
Generate embeddings and invoke predictions: use Gemini Enterprise Agent Platform text embedding models or registered model endpoints to generate text or multimodal embeddings.
Query data in natural language with QueryData: translate natural language questions into SQL queries using context sets (templates, facets, and value search queries) and integrate single-turn natural-language-to-SQL generation into your AI agents and applications.
Build data agents with conversational analytics: create and chat with stateful, multi-turn data agents in AlloyDB Studio to explore your AlloyDB data in natural language.
Generate SQL statements from natural language using the
alloydb_ai_nlextension: add in-database natural language capabilities to your application using thealloydb_ai_nlextension (for new text-to-SQL workflows, we recommend QueryData).