LangChain Meets Amazon Bedrock Agentic Retrieval

Standard RAG retrieval uses a single query vector, which struggles with multi-part questions by returning only a fraction of relevant information. Amazon Bedrock Managed Knowledge Bases offers an agentic alternative via the AgenticRetrieveStream API, which breaks complex questions into sub-queries, runs them in a planning loop, and searches again if evidence is insufficient. The langchain-aws package exposes both standard and agentic retrieval paths. This walkthrough covers knowledge base setup on Amazon S3, required IAM permissions, API differences, trace event reading, and cost tradeoffs between the two retrieval methods.