Newly launched Managed KB

As to my understanding, the reason you don't see an option to manually select a vector database in the newly launched Managed Knowledge Base is because abstracting the infrastructure is the primary purpose of this feature.

In a traditional (or "Custom") Bedrock Knowledge Base, you are required to provision, connect, and manage your own vector store (such as Amazon OpenSearch Serverless, Pinecone, or Aurora PostgreSQL).

However, with the new Managed Knowledge Base (released in June 2026), AWS fully manages the underlying vector storage, indexing, and embedding model selection on your behalf. Under the hood, this managed offering leverages Amazon S3 Vectors (S3's native vector storage capabilities) rather than spinning up a traditional, dedicated database cluster. This architectural shift is exactly why the manual selection UI is omitted. It allows Bedrock to offer a fully managed, serverless RAG pipeline where you only pay for ingestion and retrieval operations, completely eliminating the operational overhead and 24/7 cluster costs of maintaining a standalone vector database.

When to use what:

• Stick to Managed KB if your priority is speed, lower operational overhead, and cost-effectiveness (since it uses S3 storage underneath).
• Switch back to a Custom KB if your use case requires advanced search capabilities, sophisticated metadata filtering, custom ranking strategies, or direct API access to the underlying OpenSearch/Pinecone index.

Official Reference: You can review the architectural differences and the managed S3 capabilities directly on the official product page under the "Managed Knowledge Base" section -> https://aws.amazon.com/bedrock/knowledge-bases/