Bring your SharePoint data to your Agents with Foundry IQ

Enterprises sit on mountains of knowledge locked inside SharePoint — HR policies, product specs, compliance docs. The challenge is giving AI agents access to that knowledge without building a custom RAG pipeline from scratch.

Foundry IQ solves this. It is a managed knowledge layer inside Microsoft Foundry that turns enterprise data into reusable, permission-aware knowledge bases for AI agents. You create a knowledge base, point it at your SharePoint content, and any Foundry agent can query it — with citations, access control, and incremental refresh built in.

Think of it as RAG-as-a-Service — point it at your data, get a knowledge base, skip the plumbing.

In this article, I walk you through a complete, Infrastructure-as-Code deployment that connects a SharePoint site to a Foundry Agent through Foundry IQ, using an indexed SharePoint knowledge source.

The source code is available on GitHub.

What is Foundry IQ?

Microsoft provides three “IQ” layers that give agents access to different aspects of your organization:

• Work IQ (Microsoft 365) — the intelligence layer that grounds M365 Copilot and agents in real-time work context: files, emails, meetings, chats, and people signals across your organization. Built on three layers — Data, Memory, and Inference — it gives agents persistent, contextual understanding of how work happens.

• Fabric IQ (Microsoft Fabric) — a workload for unifying data across OneLake and organizing it according to your business vocabulary through ontologies. It exposes unified data to analytics, AI agents, and applications with consistent semantic meaning. Includes graph databases, data agents, and Power BI semantic models.

• Foundry IQ (Microsoft Foundry) — a managed knowledge layer that connects structured and unstructured data across Azure, SharePoint, OneLake, and the web into reusable knowledge bases for AI agents.

Each IQ workload is standalone, but you can use them together to provide comprehensive organizational context for agents.

What Foundry IQ Provides

At its core, Foundry IQ provides three things:

• Knowledge bases — reusable, topic-centric collections of knowledge sources. Multiple agents can share the same knowledge base.
• Knowledge sources — connections to indexed or remote content (Azure Blob Storage, SharePoint, OneLake, the web).
• Agentic retrieval — a multi-query pipeline that decomposes complex questions into subqueries, executes them in parallel, semantically reranks results, and returns unified responses with citations.

Sharepoint Knowledge source: Indexed vs. Remote

When you connect SharePoint to Foundry IQ, you choose between two approaches. This choice drives the entire architecture.

Indexed SharePoint

The indexed approach crawls your SharePoint site, extracts content, chunks it, generates vector embeddings, and stores everything in an Azure AI Search index. When an agent queries the knowledge base, it searches this pre-built index.

What you get:

• Full-text search, vector search, and hybrid queries
• Multimodal support — images in documents are verbalized (described in natural language) and become searchable
• Custom chunking with Content Understanding skills
• Incremental refresh via scheduled indexers
• No per-query dependency on SharePoint availability

What you give up:

• Data is replicated into Azure AI Search (storage and compute costs)
• Initial indexing takes time depending on volume

Remote SharePoint

The remote approach queries SharePoint directly at runtime through the Copilot Retrieval API. No index is created — content stays in SharePoint.

What you get:

• Zero data replication — content stays where it is
• Real-time freshness — always querying the latest version
• Simpler setup — no indexer pipeline to configure
• SharePoint permissions and Purview labels are enforced natively

What you give up:

• Text only — no multimodal retrieval (tables, images, charts not supported)
• Other limitations, such as Rate limit, Query length … — Microsoft Learn

👉 My demo repository uses the indexed approach to demonstrate the full power of AI Search and multimodal retrieval.

Authentication Deep Dive

Authentication is where most SharePoint + AI Search projects get stuck. The Azure AI Search SharePoint indexer supports three authentication methods:

• Delegated Permissions (OBO):

SharePointOnlineEndpoint=[site url];ApplicationId=[app id];
TenantId=[tenant id]

• Application Permissions with Client Secret:

SharePointOnlineEndpoint=[site url];ApplicationId=[app id];
ApplicationSecret=[secret];TenantId=[tenant id]

• Federated Credential (Passwordless) ← What My Repo Uses:

SharePointOnlineEndpoint=[site url];ApplicationId=[app id];
FederatedCredentialObjectId=[managed identity object ID];
TenantId=[tenant id]

Pros:

• Zero secrets to manage — passwordless
• No expiration, no rotation
• Same capabilities as client secret (scheduled runs, ACL preservation)
• Aligns with Zero Trust principles

The Skillset — Content Understanding & Image Verbalization

The skillset is where the magic happens. It uses Azure AI Search’s Content Understanding skill to process SharePoint documents:

1. Chunking — PDF, DOCX, and other formats are broken into manageable text sections with location metadata (page numbers, section boundaries).
2. Image extraction — embedded images are automatically extracted from documents.
3. Image verbalization — each extracted image is passed through a Chat Completion skill (GPT-4.1) that generates a natural-language description. Diagrams, charts, and figures become searchable text.
4. Vectorization — each text chunk is embedded using text-embedding-3-large (3072 dimensions) for semantic search.
5. Index projection — chunks, vectors, and verbalized images are projected into the search index.

Try It Yourself

The entire deployment is packaged as an Azure Developer CLI (azd) template. Three commands — azd auth login, azd env set, azd up — and you have a working Foundry agent grounded on your SharePoint content.

The repo handles everything: Bicep provisioning of Foundry + AI Search + Entra app registration, postdeploy hooks that wire up the knowledge source and agent, and predown cleanup. Detailed prerequisites, customization options, and step-by-step instructions are in the README:

👉 **github.com/arnaud-tincelin/foundryiq-sharepoint**

What’s Next?

This demo shows the indexed approach with passwordless authentication. Here are several directions you could take it:

• Add ACL synchronization — the indexed SharePoint source supports basic ACL sync in preview, so the agent only returns content the user is authorized to see
• Combine with other knowledge sources — add Azure Blob, OneLake, or web sources to the same knowledge base for cross-source retrieval
• Custom skillsets — extend the skillset with OCR, entity recognition, or custom skills
• Tune retrieval reasoning effort — Foundry IQ lets you set retrieval reasoning effort to minimal, low, or medium to balance cost vs. quality
• Keep in mind that FoundryIQ and its connectors are in preview and subject to changes.