What is Foundry IQ?
Foundry IQ is the managed knowledge and retrieval layer of Microsoft Foundry. The model powering an agent has a knowledge cutoff and cannot reach proprietary enterprise data on its own. Foundry IQ closes this gap: it combines one or more data sources into a configurable knowledge base that agents query through a single endpoint. Multiple agents and applications share the same knowledge base.
On each query, Foundry IQ uses agentic retrieval to process the question, retrieve relevant content, enforce user permissions, and return a grounded answer with citations. Foundry IQ builds on the agentic retrieval capability of Azure AI Search and can be used from Foundry Agent Service, the Microsoft Agent Framework, or any custom application through the knowledge base APIs.
Core Features
- Unified knowledge bases: A knowledge base groups multiple knowledge sources under a single endpoint. Indexed sources include Azure Blob Storage, OneLake, SharePoint, and existing search indexes; remote sources return always-current content from SharePoint (Copilot Retrieval API) and the web (Grounding with Bing). MCP is in private preview.
- Agentic retrieval: An LLM decomposes complex queries into parallel subqueries, selects sources, semantically reranks results, and aggregates them into a grounded answer with citations. Microsoft benchmarks cite roughly 36 percent higher response quality than single-shot RAG.
- Automated indexing: For indexed sources, Foundry IQ handles chunking, vector embeddings, and metadata extraction. Recurring indexer runs keep data incrementally up to date.
- Permission-aware answers: ACLs for supported sources are synchronized, Microsoft Purview sensitivity labels are honored, and permissions are enforced at query time under the caller’s Microsoft Entra identity.
Typical Use Cases
RAG knowledge base for agents: A Foundry agent answers domain questions from internal documents in Blob Storage, OneLake, and SharePoint. Agentic retrieval also covers questions with multiple asks and returns source-cited answers.
Permission-aware enterprise search: Employees receive answers only from content their identity is authorized to access. Enforcement happens at query time based on synchronized ACLs and sensitivity labels.
Shared knowledge base for multiple agents: Several agents and applications use the same knowledge base. Data sources, indexing, and retrieval parameters are maintained centrally instead of duplicated per agent.
Benefits
- One endpoint for scattered data sources instead of separate integrations per agent
- Cited answers with source references for traceability and trust
- Consistent permission enforcement through Microsoft Entra and Purview
- Free tier and free token allowance for cost-efficient proofs of concept
Integration with innFactory
As a Microsoft Solutions Partner, innFactory supports you with the adoption and operation of this service.
Typical Use Cases
Frequently Asked Questions
What is Foundry IQ?
Foundry IQ is a managed knowledge layer in Microsoft Foundry. It groups multiple data sources under a single knowledge base and endpoint so AI agents receive permission-aware, source-cited answers. Technically, Foundry IQ builds on the agentic retrieval capability of Azure AI Search.
When should I use Foundry IQ?
Use Foundry IQ when agents need to answer complex questions over scattered enterprise data such as Azure Blob Storage, SharePoint, OneLake, or the web. It fits RAG scenarios where multiple agents share one knowledge base and answers must be traceable to source documents.
How much does Foundry IQ cost?
Billing is consumption-based across two services. Azure AI Search charges for retrieval tokens; the free plan includes a monthly free token allowance, after which pay-as-you-go applies on the standard plan. Azure OpenAI additionally charges for the tokens used in query planning and answer synthesis, based on the model you assign.
How does Foundry IQ enforce permissions and data sovereignty?
Foundry IQ synchronizes access control lists for supported sources and honors Microsoft Purview sensitivity labels. Queries run under the caller's Microsoft Entra identity, so agents return only authorized content. The underlying Azure AI Search service is available in EU regions.
