Amazon OpenSearch Service

Amazon OpenSearch Service is an AWS-managed search, log analytics and vector database, run as sized clusters or as usage-billed OpenSearch Serverless.

Amazon OpenSearch Service runs the open-source OpenSearch engine so that teams can search text, logs and embeddings without operating the servers underneath. Amazon Web Services builds and sells it, and it carried the name Amazon Elasticsearch Service before AWS renamed it. OpenSearch itself is a community-driven project at the Linux Foundation, so the managed service runs the same engine a team could host on its own. AWS adds the infrastructure, scaling and snapshots around that engine. In its first mode, managed clusters called domains run on general purpose, compute optimized, memory optimized or NVMe SSD instances, and can add UltraWarm managed storage. The second mode, OpenSearch Serverless, swaps the cluster for collections that add shard replicas under query load and scale to zero when idle. Its k-NN engine builds HNSW indexes through Faiss and measures cosine, dot product or Euclidean distance. Scalar and binary quantization shrink those vectors, and disk-optimized storage serves indexes larger than memory. Neural search embeds text at ingest through connectors to Amazon Bedrock and Amazon SageMaker, and a hybrid search pipeline merges keyword and semantic scores into one ranking. Engineering groups that host their workloads on AWS can add search without leaving that account. Developers building semantic search for generative AI applications use collections as their vector store, with embeddings from Amazon Bedrock models. IAM data access policies grant each role read or write rights per collection and index. Operations teams keep application logs in domains for analytics and monitoring. Serverless fits traffic that swings or goes quiet, while domains suit teams that want to choose their own instance sizes. Teams run interactive log analytics and real-time application monitoring on domains, while vector ingestion jobs load Parquet or JSONL files from Amazon S3 in bulk. Website search indexes records that applications keep in Amazon DynamoDB, Amazon DocumentDB or Amazon Aurora. For semantic search, metadata filters sit inside the k-NN query, so numeric ranges or geographic areas narrow results before ranking. It fits a team that wants logs, site search and vectors in one engine inside the AWS account it already runs.