Foundry IQ gives AI agents a shared knowledge layer for retrieving information across enterprise content, rather than tying every retrieval setup to one purpose-built agent. It matters to developers building agents and to organizations that want knowledge bases reused across more than one experience. Microsoft Events demonstrates how Foundry IQ knowledge bases connect to agents and Microsoft IQ in a YouTube session.
The shift is from building retrieval-augmented generation (RAG) separately for each agent toward creating a knowledge base that agents can discover and use. RAG is a pattern in which a system retrieves relevant information and supplies it to a model to help ground an answer. Foundry IQ is presented as an evolution of Azure AI Search, expanding the emphasis from search alone to a broader retrieval system for agents.
Foundry IQ supports lexical and vector search, hybrid retrieval and reranking, alongside an agentic retrieval layer for more involved questions. Lexical search matches terms; vector search looks for meaning-based similarity; hybrid retrieval combines approaches. The system can also represent content through knowledge graphs, which add entities and relationships to the information available for retrieval.
That matters because not every question is answered by retrieving one highly relevant passage. A count across many documents, or a question about how events developed over time, may require identifying patterns and relationships across the corpus. Microsoft’s presentation describes knowledge graphs as a native retrieval primitive, working alongside the indexing system rather than as a separate system.
Foundry IQ is aimed at technical users creating knowledge bases and agents. The broader Microsoft IQ approach is intended to let agents and experiences discover appropriate knowledge bases, while allowing an individual agent to use a narrower subset of organizational data when that is the right fit.
Teams can start in Azure or Foundry, including by using an existing search resource. The demonstrated managed option starts with a Foundry project: Foundry manages the knowledge base and uses the serverless version of Foundry IQ. The presentation describes that option as requiring no separate resource provisioning or permission setup for the demonstrated creation flow.
Foundry IQ Serverless was described as coming to general availability in December 2026. Microsoft also described higher limits with that transition: indexes up to 10 GB and up to 10,000 indexes per service. Those figures apply to the serverless offering described in the session.
A knowledge base is the reusable unit that agents connect to. It brings together content and the retrieval capabilities needed to find useful information, so an agent can use the same knowledge resource as another compatible experience instead of requiring a separate retrieval setup for every task.
The session also describes a knowledge base as an MCP server, which lets compatible agent frameworks connect to it for retrieval. This provides another route besides wiring the knowledge base directly to an agent in Foundry.
Basic RAG remains a suitable option when a straightforward search and lookup returns the needed information. More advanced retrieval is useful when a question requires a count across documents, synthesis across time, or an understanding of relationships in a corpus. The goal is not to replace simple retrieval everywhere, but to provide additional approaches for cases it does not handle well.
Serverless is presented as a fit for dynamic workloads, including services that can become idle and scale to zero. Dedicated services remain an alternative for very large-scale scenarios or when dedicated compute is needed. The session describes the two service types as compatible, so organizations can choose based on workload needs.
Before choosing an approach, consider whether the agent needs a focused knowledge base or access to a broader organizational context, and whether the workload favors serverless or dedicated capacity. The presentation also describes Microsoft IQ and Foundry IQ as parts of a wider context approach; they are not interchangeable labels for a single agent or knowledge base.
Foundry IQ is a retrieval system, described as the evolution of Azure AI Search, for creating knowledge bases and connecting them to agents. It includes search, reranking and agentic retrieval capabilities.
In Foundry, add the knowledge base as knowledge for the agent. Microsoft IQ can also discover available knowledge bases and use their descriptions to determine when they are relevant.
Foundry IQ Serverless was described as coming to general availability in December 2026.
Serverless is suited to dynamic scenarios where services may become idle and scale to zero. Dedicated services remain an option for very large-scale workloads or when dedicated compute is needed.