Choosing a database for your generative AI applications

This page helps you choose AWS vector databases and vector search options for your generative AI applications. The internal databases, data lakes, or unstructured data or document stores that you use contain a wealth of domain-specific data (such as financial or health records, or supply chain information). These data stores are generically called knowledge bases. For generative AI, you can encode your data as a set of elements, each expressed internally as a vector. Vector databases enable Retrieval Augmented Generation (RAG), the process for retrieving facts from knowledge bases to fortify large language models (LLMs) with up-to-date and accurate data.

Amazon OpenSearch Service

Amazon OpenSearch Service is a fully-managed service for running OpenSearch, an open source search engine and analytics suite. OpenSearch supports vector search through its vector engine for both managed clusters and serverless collections. OpenSearch also supports vector search with its Amazon OpenSearch Serverless deployment option. If your database type is a search engine, OpenSearch is an ideal choice for use cases where your vector indexes must scale horizontally, letting you handle more throughput for storing embeddings and performing similarity searches.

Amazon MemoryDB

Vector search for Amazon MemoryDB is an in-memory database providing the fastest vector search performance at the highest recall rates among popular vector databases on Amazon Web Services. MemoryDB offers single-digit millisecond vector search and update latencies at even the highest levels of recall. MemoryDB is ideal for in-memory database workloads requiring single-digit millisecond latencies at the highest levels of recall, or if you are already familiar with the MemoryDB, Valkey, or OSS Redis APIs.