DynamoDB adds real-time vector search for scalable similarity searches

Amazon DynamoDB now offers general availability for vector search, enabling users to store vector embeddings and run similarity searches directly within the database. This eliminates the need for a separate vector store, simplifying architecture and reducing operational overhead and data movement costs.

Users can store vector embeddings as a list of floats using the existing List data type and a standard PutItem call. A new vector index type is created on the vector attribute, allowing configuration of dimensions, distance functions, and non-vector filter attributes.

DynamoDB supports Euclidean, Cosine, and Dot product distance functions, with recommendations to match the function to the embedding model's training. Inline filtering on non-vector attributes allows narrowing search results at query time.

This feature enables the development of applications requiring semantic retrieval on agentic memory, retrieval augmented generation, recommendation engines, personalized experiences, and anomaly detection, directly within DynamoDB.

The new vector search capability offers single-digit millisecond latency with high recall, designed to scale to trillions of vectors. It operates on DynamoDB's fully serverless infrastructure, requiring no provisioning or management of servers, and offers zero-downtime maintenance.

Vector search in Amazon DynamoDB is generally available in all commercial AWS Regions and AWS GovCloud (US) Regions. A walkthrough is provided to demonstrate adding vector search to an existing DynamoDB table using the console.