Knowledge graphs in Foundry IQ and Microsoft Retriever 1, a new specialized model are now in preview.
Your applications already help people find business information by keyword, field and facet. Agents use that same information to get work done: checking a policy in SharePoint, finding a supplier contract, or working out which products a new regulation affects. Retrieving the right information from your business data takes expertise and the right building blocks. Without them, agents repeatedly search for and piece together the same facts and relationships, adding time and cost.
Foundry IQ helps developers connect unstructured business content to agents. Today we’re announcing grounding in SharePoint documents with permissions enforced, knowledge graphs that connect facts across those documents, and a specialized retrieval model that plans each search. Together, they are built to get an agent to the right answer in fewer searches.
Foundry IQ is part of Microsoft IQ Platform, the enterprise context management system that captures, governs and reuses the business context your apps and agents need. Microsoft IQ brings together four sources of context: Work IQ for how people work, Fabric IQ for governed business data, Web IQ for the web, and Foundry IQ for your unstructured data.
What’s new
Today we’re announcing the following capabilities in Foundry IQ:
- Knowledge graphs, public preview: Map entities during indexing using your ontology, or bring an existing graph, then query connections in natural language or GQL with source citations. Get started with the Foundry IQ plugin.
- SharePoint knowledge source with permissions, generally available: Index your SharePoint content into Foundry IQ, and each agent sees only what the person using it can see with user Entra group and SharePoint site group permissions.
- Foundry IQ Serverless, generally available in December 2026: Usage-based pricing that scales to zero when idle, with nothing to size up front and no commitments. General availability adds a production SLA, higher limits and more regions.
- Microsoft Retriever 1, public preview: A specialized retrieval model inside Foundry IQ, built to plan each search at lower cost, with no frontier model to deploy for the retrieval plan.
Knowledge graphs: connect the dots across your documents
Knowledge graphs in Foundry IQ give agents two ways to understand your business: entities connect the facts across your documents, and topics organize your content by theme.
As it indexes your documents, Foundry IQ uses your ontology to map how your documents, contacts and information relate. It records the source of every connection.
Entities connect facts that sit in different documents, and each connection keeps its source. Topics organize the same content by theme for broad questions. Topics answer the broad questions, such as which themes run across every contract. Flag a question as broad, and retrieval draws on the whole collection.
Knowledge graphs are available in public preview:
- Entities: Add a graph configuration with your ontology to a knowledge source you already use, or bring a graph you already have. Query through the knowledge base in natural language, or the graph directly in GQL. There’s no graph database to build and no second pipeline to run.
- Topics: Topics and subtopics identified automatically from your existing embeddings, with no ontology required, and organized into a sitemap of your corpus. Topics can also propose a candidate ontology for you to review.
Install the Foundry IQ plugin to get started with knowledge graphs.
Microsoft Discovery grounds R&D workflows with Foundry IQ
Knowledge graphs in Foundry IQ are a key component of Bookshelf, the knowledge experience in Microsoft Discovery. With Foundry IQ integration, Discovery customers can now leverage a wider variety of data sources, such as SharePoint and Work IQ, to ground their R&D workflows with high-quality scientific reasoning.
John Link, Partner Product Manager, Microsoft Discovery & Quantum
Find the answer in fewer steps
Every search an agent runs is a tool call you pay for in tokens. An agent without your knowledge in reach searches, then searches again. Foundry IQ plans each search for the agent, and it’s built to get agents to the right answer in fewer searches.
Grounding your agents is simpler, too. Foundry IQ is built into every new Foundry project, powered by serverless, so you can connect your enterprise knowledge and start building grounded agents without provisioning or managing a separate resource.
- Microsoft Retriever 1, public preview: Built to plan each search at lower cost. It runs on your existing index, with no frontier model to deploy for the retrieval plan.
- Foundry IQ built into every new Foundry project, public preview in December 2026: It’s now easier to ground agents with your documents, using a managed Foundry IQ resource provided in every Foundry project. Foundry IQ handles retrieval and indexing, while Foundry sets up and manages the underlying service for you.
- Foundry IQ Serverless, generally available in December 2026: Usage-based pricing that scales to zero when idle, with nothing to size up front and no commitments. General availability adds a production SLA, higher limits and more regions.
Sign up to get started with a managed Foundry IQ resource.
Connect to your enterprise content ecosystem
Your knowledge is distributed across various source in different formats. Foundry IQ brings it to your agents in Foundry, Microsoft IQ and third party agents, while honoring document and access permissions.
Your sources
- Work IQ SharePoint knowledge source, generally available: Index your SharePoint content through the Microsoft 365 indexer, with its permissions enforced at query time.
- More knowledge sources, generally available: Azure SQL Database and file upload knowledge sources are generally available, Fabric IQ semantic models and OneLake shortcuts and mirrored items are public preview.
- Semantic chunking and image descriptions, generally available: Azure Document Intelligence and Content Understanding in Foundry split documents by meaning and make the images in them searchable.
Your permissions
- Source permissions, generally available: Foundry IQ enforces SharePoint user, Entra group and site group permissions, ADLS Gen2 ACLs and RBAC, and Azure Blob container-level RBAC under the caller’s identity, for supported and configured sources. The same controls apply when the knowledge base is reached over MCP.
- OneLake item-level permissions, public preview: Results from OneLake respect each item’s permissions under the caller’s identity.
- Purview sensitivity-label enforcement, public preview: Label policy is evaluated when the agent reads, and the call fails closed when policy can’t be verified.
OneStream empowers finance teams with better work context and Foundry IQ
Frontier Finance teams are rethinking their decisioning and workflows around AI, and they need to work in Excel, Copilot, and other tools of choice to do that, without giving up governance or auditability.
Our integration with Foundry IQ means Finance teams have greater work context to their AI answers, while still being grounded in the financial intelligence from OneStream that make those answers auditable and contextualized. Together with Microsoft, we’re unlocking powerful, purpose-built AI that works where Finance teams work, and that creates answers a CFO can stand behind.
Tom Shea, Founder and CEO, OneStream
Innovating with Microsoft Foundry
Here’s how customers are using Foundry IQ to ground their agents in their own business knowledge.
McKinsey & Company grounds their AI operating system with Foundry IQ and Fabric IQ
AI needs more than access to data. AI needs to understand how a business works. McKinsey’s EcliptOS platform brings together Microsoft Fabric IQ and Microsoft Foundry IQ to connect business data with enterprise knowledge. We build on that foundation with McKinsey’s industry expertise, codified into impact recipes, and our transformation capabilities to turn business context into persistent business value.
Brian Elliott, McKinsey Senior Partner and Co-Lead of EcliptOS, McKinsey & Company
Rockwell Automation powers troubleshooting assistant with Foundry IQ
When a machine goes down on the night shift, our technicians can’t always reach an expert, so the answer they get has to be one they can trust. With Foundry IQ, our troubleshooting assistant draws on each plant’s own manuals, historical downtime records and expert insights, and links every step to its source. When the evidence isn’t there, it says so instead of guessing. That’s what gives our people the confidence to act.
Simon Lee, Senior Manufacturing Engineer, Rockwell Automation
Get started today
The easiest way to explore is through the Microsoft Foundry portal. Read the documentation and get started with the Foundry IQ plugin.
Watch Foundry IQ's Microsoft Ignite 2026 sessions in person or on-demand:
Updated Nov 17, 2026
Version 1.0pablocastro
Microsoft
Joined March 07, 2023
Microsoft Foundry Blog
Follow this blog board to get notified when there's new activity