Foundry IQ in Action: Agentic Retrieval for Foundry Agents and Copilot Studio

Every new agent seems to start the same way: another index, another chunking strategy, another set of credentials, another retrieval pipeline to maintain. Foundry IQ changes that equation. It gives you a reusable, permission-aware knowledge layer that any number of agents can share, whether those agents live in Microsoft Foundry Agent Service or in Copilot Studio.

In this session, David Lorenzo and Rafsan Huseynov build the whole thing live. We start with the theory: what a knowledge base actually is, how knowledge sources differ from a classic single-shot RAG index, and why agentic retrieval; query planning, parallel sub-queries, semantic reranking, and iterative search produces meaningfully better answers on ambiguous, multi-step questions.

Then we build. We create a knowledge base backed by multiple knowledge sources, tune the agentic retrieval prompt and reasoning effort to trade latency against quality, and wire up authentication properly using managed identities and Entra ID rather than keys. Finally, we connect two very different agents, one pro-code in Foundry, one low-code in Copilot Studio, to the same knowledge base, and watch them return grounded, cited answers that respect the user’s permissions.

You’ll leave with a repeatable architecture pattern you can take back to your own tenant on Monday.

Agenda (~45–50 minutes):

3. Benefits of Attending

• Understand what Foundry IQ is, how it differs from classic RAG, and when agentic retrieval is worth the extra latency and token cost.
• Learn how to design a knowledge base that unifies multiple knowledge sources behind a single endpoint, instead of wiring retrieval logic into every agent you build.
• See how to tune the agentic retrieval prompt and reasoning effort to control the quality-versus-latency trade-off.
• Get a working security pattern: managed identities, Entra ID, and permission-aware retrieval so users only ever see what they’re entitled to see.
• Watch the same knowledge base serve both a pro-code Foundry agent and a low-code Copilot Studio agent, proof that one investment in knowledge can serve your whole agent estate.
• Take away practical guidance on preview-versus-GA status, common configuration mistakes, and what to plan for before going to production.

4. Is there a demo?

1. Knowledge base creation with multiple knowledge sources — configuring the sources, automatic indexing/chunking/vectorization, and tuning the agentic retrieval prompt and retrieval reasoning effort.
2. Permission and identity setup, securing the knowledge base with managed identities, role assignments, and permission-aware retrieval so results are trimmed per user.
3. Foundry agent connected to Foundry IQ, creating an agent in Foundry Agent Service, attaching the knowledge base, and testing grounded responses with citations.
4. Copilot Studio agent connected to the same Foundry IQ knowledge base, the low-code path, then a side-by-side comparison of both agents answering the same questions from a single shared knowledge layer.