Generative AI platform
Amazon Bedrock Knowledge Bases
Managed Retrieval-Augmented Generation over private data sources and vector stores.
Key points
- Connects foundation models to private data through managed retrieval, embeddings, and vector search.
- Ingests content from supported data sources and stores vector representations in a supported vector store.
- Returns retrieved context to the model so answers can be grounded in enterprise data.
- Can cite retrieved sources, which helps users inspect where an answer came from.
- Works with Bedrock Agents and direct application calls for RAG-based question answering.
When to use it
- Choose Knowledge Bases when users need natural language answers from private documents.
- Use it for support knowledge, policy search, internal documentation, and product manuals.
- Use it when you want managed RAG instead of building your own embedding and retrieval pipeline.
Exam tips
- RAG retrieves external context at inference time; fine-tuning changes model behavior through training data.
- Knowledge Bases are not the same as Amazon Kendra, which is enterprise search rather than a Bedrock-native RAG feature.
- Good RAG depends on source quality, chunking, embeddings, and access controls.
- Use Guardrails with Knowledge Bases when retrieved context or generated answers need safety filtering.