Foundry IQ: Unified knowledge layer for agents in Microsoft Foundry (Part 3)

[TallesValiatti]

Introduction

Hello everyone,

This post is a show-and-tell version of my article
“Foundry IQ: the knowledge layer for agents in Microsoft Foundry (Part 3)”.

In Part 1, I provisioned a minimal lab (Microsoft Foundry + Azure Blob Storage + Azure AI Search), ingested Contoso Software (fictitious company) SLA documents, and created a Foundry IQ Knowledge Base on top of Azure AI Search so the retrieval layer could be reused and governed centrally.

In Part 2, the goal was the next practical step: attach that Knowledge Base to a Foundry Agent, validate the end-to-end flow in the new portal, and inspect observability details using Debug and MCP calls.

In Part 3, we go one step further. The goal is to extend the same Knowledge Base with web data, using Azure Bing Custom Search, and validate how a single agent can reason over internal enterprise documents + external web content, all mediated by Foundry IQ.

---

Full article (Part 3)

https://www.azurebrasil.cloud/blog/foundry-iq-a-camada-de-conhecimento-dos-agentes-no-microsoft-foundry-parte-3/

Related

Part 2

• Discussion: https://github.com/orgs/microsoft-foundry/discussions/300
• Article: https://www.azurebrasil.cloud/blog/foundry-iq-a-camada-de-conhecimento-dos-agentes-no-microsoft-foundry-parte-2/

Part 1

• Discussion: https://github.com/orgs/microsoft-foundry/discussions/289
• Article: https://www.azurebrasil.cloud/blog/foundry-iq-a-camada-de-conhecimento-dos-agentes-no-microsoft-foundry-parte-1/

---

What you’ll see in this lab (Part 3)

• ✅ Extend an existing Foundry IQ Knowledge Base with web sources
• ✅ Configure Azure Bing Custom Search as a knowledge source
• ✅ Combine Blob-based internal docs + web content in the same Knowledge Base
• ✅ Bind the updated Knowledge Base to a Foundry Agent
• ✅ Run mixed-context prompts (internal + external)
• ✅ Inspect how Foundry IQ performs agentic retrieval with sub-queries

Features and Screenshots

Technical Details

Technical Details

• AI Platform: Microsoft Foundry
• Knowledge Layer: Foundry IQ (Knowledge Bases + Knowledge Sources)
• Search Engine: Azure AI Search
• Internal data source: Azure Blob Storage
• External data source: Azure Bing Custom Search
• Model (chat completions): gpt-4.1-mini
• Scenario: Contoso Software (fictional) — SLA policies + Azure Key Vault authentication documentation

Challenges and Solutions

Challenges and Solutions

• Unifying internal and external knowledge
  → Solved by adding web sources directly into the same Foundry IQ Knowledge Base.

• Avoiding multiple, ad-hoc RAG pipelines
  → Retrieval logic remains centralized and reusable across agents.

• Maintaining traceability and trust
  → Debug view clearly shows sub-queries, sources, and evidence used.

• Scaling agent capabilities without changing code
  → Knowledge evolution happens at the Knowledge Base level, not inside the agent.