Amazon DynamoDB Supports Real-Time Vector Search at Any Scale

Amazon DynamoDB has officially launched its vector search feature. It allows storing operational data and vector embeddings in the same table and performing similarity searches without replicating data to a separate vector database.

Native vector search targets over 99% recall and single-digit millisecond latency, scaling to trillions of vectors. It requires no server provisioning, patching, or maintenance, and vector indexes scale horizontally as data grows.

Existing DynamoDB users previously had to operate separate dedicated vector databases and synchronization pipelines, but now vectors and operational data share the same serverless infrastructure and pay-per-use model. This reduces the burden of data movement, licensing, operations, and latency management.

Users can store vectors generated by models such as Amazon Bedrock Titan Text Embeddings, Cohere Embed, and OpenAI embedding models in a List type attribute in DynamoDB, then create a vector index on that attribute. When creating the index, users specify the number of dimensions and the distance function, and can apply filters based on non-vector attributes during search.

Key supported features include:

• Vectors up to 4096 dimensions
• Euclidean, Cosine, and Dot product distance functions
• Returning up to 100 search results
• Similarity search via the SearchVectors API
• Inline filtering at query time

AWS explained that DynamoDB vector search can be utilized for applications that need to handle operational data and semantic search together, such as product catalogs, agent memory, RAG, recommendation engines, personalization services, and anomaly detection.