Hi btwstl,
Yes, agentic retrieval in Azure AI Search does support querying multiple indexes, and this is a key part of its design.
According to the official Microsoft documentation and announcements from Build 2025:
- Agentic retrieval uses a multiturn query engine that can break down complex queries into subqueries.
- These subqueries are then dispatched in parallel across both traditional text fields and vector embeddings within Azure AI Search.
- While the demos may have shown only one index in use, the architecture is designed to support parallel searches across multiple data sources, which can include multiple indexes if configured appropriately.
- The retrieval engine orchestrates the entire process: it invokes the LLM for query planning, dispatches subqueries (potentially to different indexes), collects and reranks results, and returns a unified response.
This means that if you have different indexes for different domains (e.g., product data, support documents, user manuals), agentic retrieval can intelligently route subqueries to the appropriate index and synthesize the results into a coherent answer.
If you have any further assistant, do let me know.