AWS adds retrieval modes that reduce model calls in Bedrock

AWS has added single-search answers and passage-only retrieval to Amazon Bedrock Managed Knowledge Base's AgenticRetrieveStream API. Developers can now use the same API for direct lookups and questions requiring multiple searches, choosing how much model processing each request receives. Setting maxAgentIteration to 1 retrieves passages once and uses one foundation model call to generate an answer. Setting it to 0 returns passages without any foundation model calls; response generation must be disabled. Existing settings from 2 to 5 retain their behavior for searches that involve planning and refinement. Applications must select the mode themselves; the parameter does not automatically route questions.