Developers using Amazon Bedrock Knowledge Bases can now compare three customer-managed vector backends through a single workload-focused guide. AWS evaluates OpenSearch Service, Aurora PostgreSQL with pgvector and S3 Vectors against different retrieval-augmented generation needs.
Retrieval-augmented generation, or RAG, finds relevant document passages and supplies them to a language model as context. The vector store therefore affects response latency, search features, operating cost and how easily an application can use data it already holds.
AWS positions OpenSearch for high-throughput catalog search where low latency and a mix of keyword and semantic retrieval matter. Aurora combines relational data with vector search and supports the IVFFlat and HNSW index types, making it useful when an application already depends on PostgreSQL records. S3 Vectors targets very large, cost-sensitive collections; AWS claims sub-second similarity queries and storage savings of up to 90% compared with traditional vector databases.
Those descriptions are product guidance from AWS, not an independent cross-cloud benchmark. They still offer a practical decision rule: start from the application’s latency, scale, filtering and data-location requirements. Moving embeddings later can be expensive, so the cheapest storage option is not automatically the cheapest system to operate.