Amazon advises OpenSearch, Aurora PostgreSQL or S3 Vectors for Bedrock Knowledge Base

For product search, the blog recommends OpenSearch Serverless because its hybrid search and low‑millisecond latency suit high‑traffic catalogs. For large‑scale research workloads, S3 Vectors is presented as the most cost‑effective choice, cutting storage costs by up to 90 % while keeping sub‑second query times even at a million vectors. For chatbot knowledge bases, Aurora PostgreSQL with pgvector offers a balanced mix of sub‑50 ms latency and relational join capability, allowing developers to fine‑tune indexing methods such as HNSW or IVFFlat. The guidance stresses profiling the application, matching it to a suitable store, and planning for future tiered architectures that combine stores as needs evolve.