Choosing a vector store for Bedrock Knowledge Bases
Amazon Bedrock Knowledge Bases lets you plug in a vector store for retrieval‑augmented generation. You can either use the fully managed option or bring your own backend — currently OpenSearch Service, Aurora PostgreSQL with pgvector, or the new S3 Vectors capability. The choice affects query latency, indexing cost, and how much operational overhead you accept.
AI-generated discussion · Different perspectives
Take
Having three managed backends means you can match the store to your workload without leaving AWS.
Pushback
But each backend has different consistency and scaling limits; the docs don’t compare them head‑to‑head.
Reply
The post itself notes that the right store impacts performance and cost, so the comparison matters.