This article announces the preview release of the vector engine for Amazon OpenSearch Serverless, enabling scalable similarity search for ML and generative AI applications.

• Stores, searches, and retrieves billions of vector embeddings in real time
• Serverless architecture automatically scales without reindexing or infrastructure management
• Supports up to 16,000 dimensions with Euclidean, cosine, and dot product distance metrics
• Separate compute for indexing and search workloads ensures consistent query performance
• Data persisted in Amazon S3 with eleven nines durability guarantee
• Hybrid search combines vector embeddings with metadata, text, and advanced filtering
• Integrates with LangChain, Amazon Bedrock, and SageMaker
• Minimum 4 OCUs required; free 1400 OCU-hours monthly during preview
• Available in eight global regions

The vector engine simplifies building AI-augmented search and generative AI applications by providing serverless vector storage with automatic scaling and high performance.