Amazon DynamoDB Now Supports Real-Time Vector Search!

Hi, I'm Shiichan! I've got exciting news from AWS today: DynamoDB can now do real-time vector search! I bet a lot of you have been waiting for this one.

AWS What's New aws.amazon.com

What was announced?

AWS's What's New page announced the general availability of vector search for Amazon DynamoDB. It's a new capability that lets you store vector embeddings alongside your other attributes and index and search them in real time.

It's designed to deliver single-digit millisecond latency with 99%+ recall, and to scale to trillions of vectors. Since this is AWS's official launch announcement, I'll keep things tight here and focus on what you can actually do with it.

Why it matters

As vector datasets grow into the billions or trillions, it's traditionally hard to have search speed, scale, and accuracy all at once. Latency tends to climb as the vector count grows, and avoiding that usually means giving up recall or throughput instead.

DynamoDB has always been strong as a serverless key-value/document database with zero infrastructure to manage. Adding native vector search means you can now do semantic search right alongside your existing DynamoDB data, in the same place.

What changes

The announcement calls out quite a few use cases that get a lot easier now:

• Semantic retrieval over AI agent memory (if you're already storing agent memory in DynamoDB, you can now add semantic search on top of it)
• Product similarity search
• Personalized advertising
• Retrieval augmented generation (RAG)
• Recommendation systems

And true to DynamoDB form, you still get the serverless benefits: zero infrastructure management, zero downtime, zero maintenance windows, and you only pay for what you use. You don't need to stand up a separate vector database anymore — vector search is just an extension of the DynamoDB table you're already using.

Dive Deep

The basic flow is simple: store your vector embeddings alongside your other attributes, create a vector index, and run approximate nearest neighbor (ANN) searches. You can generate the embeddings with any model you like, including models available on Amazon Bedrock.

You can also pick the vector index's partition key to control scaling, and filter on attributes to narrow down your search results.

It's available in all commercial AWS Regions and the AWS GovCloud (US) Region, so it's ready to use in a really wide footprint from day one. For pricing details and setup steps, AWS points to the AWS News Blog, the Amazon DynamoDB product page, and the Amazon DynamoDB Developer Guide, so check those out if you want to go deeper.

Wrap-up

• Vector search for Amazon DynamoDB is now generally available
• Single-digit millisecond latency, 99%+ recall, and scaling to trillions of vectors
• Supports use cases like agent memory retrieval, product similarity search, advertising, RAG, and recommendations
• Serverless with no infrastructure management and pay-as-you-go pricing
• Available in all commercial AWS Regions and AWS GovCloud (US)

You no longer need a separate vector database to get semantic search — DynamoDB alone can now handle it, which makes this a really compelling update if you're building AI agents or RAG pipelines.