Amazon DynamoDB Launches Vector Search With 99%+ Recall Across 4,096 Dimensions
Updated
Updated · AWS Blog · Aug 5
Amazon DynamoDB Launches Vector Search With 99%+ Recall Across 4,096 Dimensions
3 articles · Updated · AWS Blog · Aug 5
Summary
Amazon made DynamoDB vector search generally available, letting customers run similarity queries directly on operational data with single-digit millisecond latency and up to 100 results per search.
The feature is aimed at users who previously had to copy DynamoDB data into a separate vector database, adding synchronization pipelines, extra costs and latency-management overhead.
DynamoDB stores embeddings as float lists in existing tables and adds a new vector index type with optional inline filters, plus Euclidean, Cosine and Dot product distance functions.
The service supports up to 4,096 dimensions, horizontal scaling to trillions of vectors, and serverless pay-per-request pricing with no servers, versions or maintenance windows.
Amazon said the capability is available in all commercial AWS Regions, including AWS GovCloud, for uses such as RAG, agent memory, recommendations and anomaly detection.