Grounding Copilot Studio Agents with Azure AI Search and Foundry IQ

2026-07-22 · Source: Microsoft Foundry Blog articles · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Software Development & Engineering · Depth: Intermediate, long

Summary

Microsoft's latest guidance details five retrieval patterns for grounding Copilot Studio agents using Azure AI Search and Foundry IQ, exemplified by an "Ask HR" agent. These patterns, available in the `foundry-copilot-hr-policy-knowledge` sample repo, address varying user needs, from direct document links to synthesized, grounded answers. Patterns include direct index query (A, ~1-2s latency), agentic retrieval via Foundry IQ (A2, ~2-4s), forced-grounding synthesis with Foundry Agent Service (B, ~10-14s), deterministic document locators (C, ~1-2s), and a self-hosted agent (Hosted Agent, ~10-14s). All patterns leverage a single `hr-policy-index` in Azure AI Search, allowing for flexible layering without re-indexing. The choice depends on whether users need an answer or a document, if an LLM agent is required, and the preferred runtime environment.

Key takeaway

For AI Architects designing enterprise knowledge agents, you must strategically select retrieval patterns based on user intent and performance needs. Start with simpler patterns like direct index search (Pattern A) for basic document location, then layer agentic retrieval (Pattern A2) for improved quality without agent overhead. For synthesized, grounded answers, consider Foundry Agent Service (Pattern B), accepting higher latency. Implement Pattern C for deterministic document links, routing high-volume locator queries efficiently.

Key insights

Grounding enterprise agents requires choosing from diverse retrieval patterns based on user needs and technical constraints.

Principles

• Retrieval patterns exist on a spectrum.
• Separate retrieval assets from orchestration.
• User intent dictates the retrieval path.

Method

A decision tree guides pattern selection based on three questions: 1) user need (answer vs. document), 2) LLM agent necessity, and 3) runtime hosting preference.

In practice

• Use Pattern A for simple, native citation needs.
• Implement Pattern A2 for agentic quality without agent overhead.
• Add Pattern C for fast, deterministic document links.

Topics

• Copilot Studio
• Azure AI Search
• Foundry IQ
• Retrieval-Augmented Generation
• Enterprise Agents
• Knowledge Bases
• Agentic Retrieval

Code references

• honestypugh2/foundry-copilot-hr-policy-knowledge

Best for: AI Engineer, AI Architect, MLOps Engineer

Related on AIssential

• Answers You Can Trust: Grounding Enterprise Agents with Foundry IQ — Enterprise AI trust requires grounding answers in internal data, providing citations, and enforcing user permissions at retrieval.
• Turn Enterprise Knowledge into Answers with Copilot Studio and Azure AI Search — Integrate Copilot Studio with Azure AI Search for grounded, secure enterprise answers using existing knowledge bases.
• Building retrieval harness for enterprise agents — Effective enterprise agents require a retrieval harness combining LLM capabilities with robust indexing and diverse data access tools.
• On Accelerating Grounded Code Development for Research — Lexical search and structured workflows can effectively ground coding agents in dynamic, domain-specific research.
• Introducing Web Search on Amazon Bedrock AgentCore — Amazon Bedrock AgentCore's Web Search provides agents with current, private web knowledge via a managed connector, overcoming static training data limitations.
• Prompt the agent - Perplexity — Distinguish agent "instructions" (persistent rules) from "input" (turn-specific task) for precise control and grounded retrieval.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Microsoft Foundry Blog articles.