Amazon DynamoDB now includes vector search as a generally available feature, letting teams run similarity searches directly on data already stored in the table. There's no need to copy embeddings into a separate vector database or maintain a synchronization pipeline. AWS suggests the capability for agentic memory, retrieval augmented generation, recommendation engines, personalized experiences, and anomaly detection.

The feature is serverless and shares DynamoDB's existing pay-per-request pricing and infrastructure. AWS says vector search delivers single-digit millisecond latency with 99%+ recall and is designed for any scale, up to trillions of vectors. Vector indexes have no storage limits and scale horizontally as data grows. There are no servers to provision, patch, or manage, no software to install, and no versions or maintenance windows.

To get started, you generate embeddings with a model such as Amazon Bedrock Titan Text Embeddings, Cohere Embed, or OpenAI text embedding models, and store them as a list of floats in an attribute using a standard PutItem call. Then create a vector index on that attribute, specifying the number of dimensions, the distance function, and any non-vector attributes to use as filters. The SearchVectors API accepts a query vector, a result count of up to 100, and optional filters, and returns results ranked by similarity. DynamoDB supports up to 4,096 dimensions, Euclidean, Cosine, and Dot product distance functions, and inline filtering.

A walkthrough covers an online sporting goods store adding semantic search to a product catalog. Existing items get a descriptionEmbedding attribute, and a vector index is created with marketplace as the optional partition key and category as a filter attribute. Running a query for "lightweight running shoes for summer" with a top-K of 5 in the US marketplace and footwear category returns the five most similar products, along with normal attributes like name and price. For Cosine and Euclidean indexes, lower similarity scores mean closer matches; for Dot product, higher scores mean closer matches.

Vector search in DynamoDB is generally available in all commercial AWS Regions, including AWS GovCloud (US) Regions. Pricing details are available on the Amazon DynamoDB pricing page.

Source: AWS