Explore Foundry IQ
Unit 3 of 7 · Module 10: Build knowledge-enhanced AI agents with Foundry IQ
Before you start
Short explainers for ideas this lesson builds on. Watch these first if they're new to you.
Go deeper
Unlocks when you've watched the lesson.
Read this lesson
A shared knowledge platform
Building RAG yourself means a lot of plumbing: vector databases, embedding pipelines, search tuning, and the servers underneath. Now multiply that by every agent: three agents, three separate RAG setups, each built and fixed on its own. Think of a river port instead. Foundry IQ is a managed knowledge layer for agents, running on Azure AI Search. It turns retrieval into one shared service that any number of agents can call. Say your product guides sit in SharePoint, your customer policies in Blob Storage, and your training material in OneLake. The old way, each agent would need its own copy of all three, indexed separately. With Foundry IQ, the barges unload once into shared warehouses, called knowledge bases, and any agent can pull from them. So you spend your time on what your agents should know, not on how they find it.
Organized by business domain
The warehouses are labeled by business topic, not by where the cargo came from. That matches how people actually think about information. An agent doesn't go looking in SharePoint site A, or Blob container B. It asks the Product Documentation warehouse, or the HR Policies one. A Product Documentation knowledge base might combine specs from SharePoint, API guides from Blob Storage, usage data from OneLake, and past support tickets from an existing search index. For the agent, that's one place to ask. For you, it's a set of knowledge sources plugged into one knowledge base, instead of a separate retrieval system for each.
Connecting data sources
You point Foundry IQ at a SharePoint site, a Blob container, or OneLake, and the cranes do the heavy lifting: indexing, embeddings, and search tuning. Adding a source kicks off four jobs. Discovery: the location is scanned for documents. Processing: each document is split into chunks and turned into embeddings, so it can be searched by meaning. Indexing: the chunks become searchable through the knowledge base. And keeping current: indexer runs on a schedule pick up new and changed documents. You set up each source once, and every agent using that knowledge base sees the updates.
Built-in retrieval intelligence
When an agent asks a question, the harbour master does the thinking about how to find the answer. A language model plans the search, and breaks a tricky question into smaller subqueries. Asking about the return policy for damaged items calls for a different search than asking for a list of every return policy. The subqueries run side by side, using keyword, vector, or hybrid search, and semantic ranking then puts the most relevant chunks first, so the agent gets strong context in fewer tokens. You can also set how hard it thinks about retrieval, with a reasoning effort of minimal, low, or medium. And every answer comes back with citations, so people can check where it came from. No custom code needed: you decide what goes in the warehouse, and Foundry IQ works out how to fetch it.
Connect agents, share knowledge
Connecting an agent is a short job, across a bridge built on the Model Context Protocol, MCP. First, a project connection lets your project reach the knowledge base, using the project's managed identity, which needs a reader role on your search service. Then you add an MCP tool to the agent: a server label, like product docs, and a server URL made from your search endpoint, slash knowledgebases, the knowledge base name, and slash mcp. The tool allows one operation, knowledge base retrieve, and points at your project connection. Give the agent instructions like: use the knowledge base for every answer, say I don't know when it isn't there, and cite your sources. Now picture three agents: a support agent, an HR assistant for staff, and a developer agent for the AP Is. The support and developer agents share the Product Documentation knowledge base. Only the HR assistant uses HR Policies. Each agent gets just the knowledge it needs, and when you improve a knowledge base, every agent connected to it benefits at once.
Facts from Microsoft's documentation: What is Foundry IQ?; Connect Agents to Foundry IQ Knowledge Bases (CC BY 4.0), rewritten in Loop's own words and examples. Narration, animation, and quiz by Loop. Microsoft, Azure, and Microsoft Foundry are trademarks of Microsoft. Loop is independent and not affiliated with or endorsed by Microsoft.