Latest

Microsoft's Two RAG Approaches in Azure AI Search: What the Documentation Says

Microsoft describes agentic retrieval—a complete RAG pipeline with LLM-based query planning, currently in preview—and the classic hybrid search pattern for teams that need generally available features. Before committing, Chooscope flags unverified questions around preview stability, access control in practice, and data source compatibility.

Microsoft's official documentation for Azure AI Search describes two RAG approaches. The first, called agentic retrieval, is described as a complete RAG pipeline with LLM-based query planning—currently in preview—that handles multi-source access and returns structured responses optimized for agent consumption. The second is the classic RAG pattern: hybrid search plus semantic ranking, intended for simpler requirements or when generally available features are required. Microsoft also states that Azure AI Search underpins Foundry IQ, described as a managed knowledge layer that transforms enterprise content into reusable, permission-aware knowledge bases for agents in the Microsoft Foundry portal. According to the same documentation, opening private content to LLMs requires granular access control, and users and agents must only retrieve authorized content.

What to verify before committing to a scope

The documentation names several real constraints worth treating as evaluation questions rather than solved facts. The agentic retrieval pipeline—including LLM-based query planning—is described as preview-stage, meaning its feature scope or availability may change. Microsoft says the classic pattern suits teams that need generally available features now; that framing implies a concrete trade-off between capability and stability worth testing against your own timeline. On access control, the documentation states that granular permission-awareness is a design goal, but independent verification of how that works in practice across specific enterprise data sources—SharePoint, databases, blob storage—has not been reviewed by Chooscope. Regional availability and pricing are not addressed in the reviewed excerpt. Before defining the knowledge scope your RAG deployment will cover, the practical questions to resolve are: which data sources your team can connect without disrupting existing governance; whether preview-stage features are acceptable for your production timeline; and how access permissions from each source map to the search index. The vendor documentation gives a starting frame; the answers depend on your environment.

Sources

  1. Microsoft — 2026-10-18

Change history

  • rev.1Published

Display language

Related