Foundry IQ: Unlocking ubiquitous knowledge for agents

pablocastro

Microsoft

Nov 18, 2025

Introducing the next generation of RAG, to fuel every agent with enterprise context.

Agents have become a part of everyday work at frontier firms. As organizations scale from a handful of copilots to dozens of agents, a familiar challenge emerges: every team is rebuilding its own way to connect models with business context: homegrown RAG pipelines, one-off vector databases, access control complexity, and enterprise governance all create friction for developers to build.

With Foundry IQ in public preview, we are launching something different, a unified knowledge layer for agents. It is an agent’s single endpoint for knowledge, delivering better context with automated source routing and advanced agentic retrieval, all while respecting user permissions.

Built on Azure AI Search, we are expanding Microsoft Foundry with new preview capabilities to bring this knowledge layer to life:

Foundry IQ knowledge bases: Available directly in the new Foundry portal, knowledge bases are reusable, topic-centric collections that ground multiple agents and applications through a single API. Building agents becomes simpler, no longer requiring a tangle of data tools stitched into every project.
Automatic access to indexed and federated knowledge sources: Expand what data an agent can reach by connecting to both indexed and remote knowledge sources. For indexed sources, Foundry IQ delivers automatic indexing, vectorization, and enrichment for text, images, and complex documents.
Agentic retrieval engine in knowledge bases: A self-reflective query engine that uses AI to plan, search, and synthesize answers across sources with configurable “retrieval reasoning effort.”
Enterprise-grade security and governance – Support for document-level access control, alignment with existing permissions models, and options for both indexed and remote data.

Foundry IQ is a culmination of the intelligence of Microsoft Cloud, your custom applications, and the web. This announcement is part of Microsoft Cloud-wide initiative to power every organization with universal enterprise context. Work IQ from Microsoft 365 provides signals on how your organization operates, Fabric IQ brings business meaning to the data in Power BI, and Foundry IQ unifies and centralizes access to knowledge to ground every agent with the right context.

From custom pipelines to reusable knowledge

Traditional RAG puts a heavy tax on every new project. Every team must rebuild data connections, chunking logic, embeddings, routing and permissions from scratch. It leaves organizations with a mess of fragmented, duplicated pipelines all trying to answer the same question in a silo: what context does the model need to respond effectively?

Foundry IQ shifts that work into knowledge bases. Instead of wiring retrieval logic into every agent, you define a reusable knowledge base around a topic (such as employee policies, product documentation, or support content) and create it in Foundry portal. From there, any number of agents and applications can connect and be grounded with that same knowledge base.

Behind the scenes, Foundry IQ federates data across indexed and remote knowledge sources: M365 SharePoint, Fabric IQ, OneLake, Azure Blob Storage, Azure AI Search indexes, the web, and MCP in private preview—can all contribute to the same knowledge base. Developers do not need to manage routing or implement different retrieval strategies per source; the knowledge base presents a simple but sophisticated endpoint for agents to query.

For indexed sources, Foundry IQ automatically manages the full indexing pipeline: content is ingested, chunked, vectorized, and prepared for hybrid retrieval. When you enable Azure Content Understanding on supported sources, complex documents gain layout-aware enrichment as well—tables, figures, headers, and sections are extracted and structured for better retrieval without extra engineering steps.

The effect is that getting started feels much closer to “plug in the knowledge this agent should have” than “rebuild a RAG stack.” Teams can stand up new agents by grounding them on existing knowledge bases, rather than recreating data connections, access rules, and retrieval logic for each project.

Retrieval that plans, iterates, and respects context

Single-shot RAG, where one query hits one index once, quickly runs into limits when questions are ambiguous, multi-step, or span several systems. Foundry IQ uses an agentic retrieval engine inside knowledge bases to tackle these harder questions.

When an agent calls a knowledge base, the engine treats retrieval as a reasoning task, not just a keyword lookup. It plans how to search, rewrites and decomposes the question when useful, reaches into multiple sources, evaluates whether it has enough signal, and iterates when it does not—before synthesizing context for the model, complete with citations.

Developers guide this behavior through high-level controls rather than plumbing. A configurable retrieval reasoning effort setting lets you express what to prioritize: low effort for fast, lightweight lookups; higher effort when it is worth taking extra steps to gather better context from across the estate. At higher effort levels, the engine leans more on agentic techniques such as iterative search and richer planning over sources.

Because the engine spans the full knowledge base, it can combine content from any supported source in one pass. That means an agent answering a customer question can draw on product manuals, troubleshooting flows, past tickets, and policy documents without developers implementing bespoke orchestration code.

Crucially for enterprise scenarios, this retrieval is permission-aware. Knowledge sources can use the caller’s identity and honor granular access control where the underlying system supports it. When a user queries an agent grounded on Foundry IQ, the retrieval engine takes into account document-level permissions in SharePoint and Azure Blob / ADLS Gen2, so the context assembled for that user reflects what they are actually allowed to see.

Customers already using Azure AI Search’s retrieval capabilities have seen what this kind of engine can do at scale:

AT&T integrated Azure AI Search and retrieval-augmented generation into its multi-agent framework and reduced customer resolution times by 33 percent, cut average handle time by nearly 10 percent, and scaled 71 AI solutions to 100,000 employees—turning disconnected data into instant, trusted insights.
Ontario Power Generation (OPG) used agentic retrieval “to sift through over 40 years of nuclear operating experience, unlocking data-driven decision-making and helping new staff quickly learn from decades of institutional knowledge. Its scalable, secure vector search and powerful reranking capabilities have put critical insights at our team’s fingertips.” -Mishca de Costa, OPG, Sr. Manager Digital Innovation & Strategy

Foundry IQ brings that same retrieval strength into a managed knowledge layer, so every new agent can start from a proven foundation instead of reinventing RAG.

Governance, observability, and end-to-end trust

For agents to be truly enterprise-ready, organizations need guarantees about who can see what content, and how to trace a RAG agent’s behavior back to the content and policies that were used

Foundry IQ is built on an enterprise-ready foundation with Entra-ID based governance. It respects user permissions in the knowledge sources that Foundry IQ connects to, plus data classifications and sensitivity labels from Microsoft Purview are respected through the indexing and retrieval pipeline. Classified content remains tagged and governed as it flows into knowledge bases, and policies you have defined in Purview continue to apply when agents are grounded on that data. This closes one of the biggest gaps in DIY RAG, where retrieval stacks often have to approximate or duplicate security rules and policy in application code.

On top of this, usage of Foundry IQ is monitored through the Foundry Control Plane. Central teams can see which agents exist, what knowledge bases they draw from, and how they are being called. They can dig into traces and investigate when LLM judges assess agents to be providing ungrounded responses or when Microsoft Defender intervenes to block bad behavior or poisoning.

With today’s announcement of Foundry IQ, Microsoft is building on the work of Azure AI Search to deliver a knowledge layer that is easy to use, capable, and secure at enterprise scale—so you can scale from experiments to mission-critical workflows with confidence.