Blog Post

Microsoft Foundry Blog
6 MIN READ

Foundry IQ SharePoint (indexed) knowledge source is now generally available

gia_mondragon's avatar
gia_mondragon
Icon for Microsoft rankMicrosoft
Nov 18, 2026

More enterprise knowledge sources in Foundry IQ, including SharePoint (indexed), are now generally available, alongside expanded Microsoft Entra group, user, and document-level permissions

An employee asks an agent how to handle a customer escalation. The procedure lives in SharePoint, the exception is in a table, and the decision tree is a diagram. Some of that guidance is restricted to the escalation team.

The agent needs more than a matching file. It needs usable evidence and access controls for the person asking.

Hundreds of customers have been testing the SharePoint (indexed) knowledge source and SharePoint indexer in Foundry IQ (Azure AI Search). This post explores how these capabilities preserve supported permissions, make rich documents useful, and ground agents in evidence each person can access.

It then looks beyond SharePoint—at document-level controls for other sources, private ingestion and cross-tenant encryption, Azure SQL and File knowledge sources, and knowledge-base retrieval that brings evidence together with citations. Selected public previews for source-side permissions, retrieval options, and model APIM endpoints support are covered separately below.

Start with SharePoint. Bring the permissions.

The SharePoint indexer and SharePoint (indexed) knowledge source are now generally available, bringing document libraries, lists, and modern site pages into Foundry IQ, including content from subsites.

Configure a knowledge source, and Foundry IQ creates the ingestion and enrichment pipeline. Your team can focus on the knowledge the agent needs instead of assembling each indexing component.

 

Figure 1. SharePoint (indexed) carries source permissions through ingestion and chunking so retrieval can filter the chunks returned to each user.

 

Supported SharePoint user, Microsoft Entra group, and site group permissions can travel with the content. At retrieval time, Foundry IQ filters results using indexed permission metadata and the requesting user's identity. Those controls also apply to the document chunks used for grounding.

Selected configuration fields (illustrative excerpt, not a complete request; see the linked how-to for connection, scope, model, and API settings):

{
  "kind": "indexedSharePoint",
  "indexedSharePointParameters": {
    "ingestionParameters": {
      "ingestionPermissionOptions": ["userIds", "groupIds"],
      "contentExtractionMode": "standard"
    }
  }
}

At retrieval, send application credentials in Authorization and the requesting user's access token in x-ms-query-source-authorization. The two tokens serve different purposes.

For the escalation agent, that means retrieving relevant guidance with supported access controls, not treating everything the connector can read as available to everyone.

Illustrative example: an escalation-team user can retrieve the restricted exception table and diagram description; a general employee sees only the procedure they are permitted to access.

Good to know: SharePoint permissions are captured at indexing. Item-level changes appear after a successful sync; inherited permissions changed at a parent scope can require an explicit resync. Review the supported permission model before deployment.

Make rich documents more useful, not just searchable

The answer may be in a diagram, a chart, or a table, not a paragraph.

The Azure Content Understanding skill, powered by Content Understanding in Foundry Tools, combines layout-aware extraction with chunking and optional image descriptions. Semantic chunking and AI-generated image descriptions are now generally available.

Semantic chunking follows document structure and paragraph boundaries. Configured image descriptions turn applicable charts, diagrams, and figures into text that can contribute to retrieval alongside the surrounding content.

For supported SharePoint documents, this connects extraction, enrichment, and permission-aware retrieval in one workflow. The escalation agent can draw on the decision diagram, chart or table as well as the written procedure. In supported knowledge sources, Standard content extraction mode (contentExtractionMode: "standard") selects Content Understanding-based processing for the generated pipeline. Teams can also configure the skill directly for more control. Learn how Content Understanding makes rich SharePoint content ready for agent retrieval.

Indexed or remote: choosing your SharePoint path

For a Foundry IQ production workload that needs control over indexed content and enrichment, start with SharePoint (indexed) knowledge source, which is generally available. SharePoint (remote) is a preview option for retrieving from SharePoint with its native permissions and sensitivity labels. For a direct SharePoint chat experience, consider a Copilot Studio agent. The indexed knowledge source is the focus of this post.

Table 1. Choose between indexed SharePoint (GA) and remote SharePoint (preview) based on governance, freshness, latency, cost, and content needs.

Factor

SharePoint (indexed) - GA

SharePoint (remote) - preview

Best fit

GA path for production grounding with a searchable, enrichable index.

Preview path for retrieval directly from SharePoint.

Governance

Stores supported SharePoint access-control lists (ACLs) on indexed chunks; not the full SharePoint policy model.

SharePoint checks its permissions and sensitivity labels at retrieval.

Freshness

Content and ACL changes take effect after a successful indexer sync; inherited ACLs can need resync.

No Search index copy; Microsoft 365 supplies the current retrieval results.

Latency

Control the index and serving; benchmark latency on your workload.

 

Retrieval depends on upstream response and has limited concurrency in preview.

Setup / cost

Requires app registration and Search ingestion; enrichment adds usage charges when enabled.

Requires eligible Microsoft 365 Copilot licensing or Retrieval API pay-as-you-go; preview.

Content

Content Understanding can extract layout and tables and optionally describe images.

Text retrieval; nontextual tables, images, and charts are not returned.

Extend enterprise access controls

The security story goes beyond SharePoint:

More knowledge. A production-ready retrieval foundation.

Enterprise answers often span systems. Indexed Azure SQL knowledge sources are now GA, bringing rows from Azure SQL Database and Managed Instance into a managed ingestion pipeline. File knowledge sources are also GA, letting teams upload curated collections without first provisioning Azure Storage. Each source has its own access-control requirements.

The knowledge base can use a model to break a question into subqueries, retrieve across configured sources, and return a cited answer. Streaming responses, citation lookup, and controls for retrieval limits, warnings, and failures give developers more ways to build a responsive, observable experience.

Together, these capabilities connect the pieces: enterprise content, usable evidence, access controls, and retrieval your application can operate.

This post focuses on Azure public cloud. For other clouds, check the feature and regional availability list and the requirements for each source.

What’s in public preview

Alongside those generally available capabilities, teams can explore public previews when their requirements call for different ways to protect sources, retrieve evidence, or route model traffic.

OneLake permission preservation takes a different approach from indexed SharePoint. In public preview, the index retains source locations so OneLake can evaluate the requesting user’s access at retrieval time. That extends permission-aware grounding into Fabric without treating indexed access-control lists and source-side checks as the same security model.

Once the right evidence is available, how the agent finds it becomes the next question. Microsoft Retriever 1 offers a specialized retrieval model, while knowledge graph retrieval follows relationships across entities. Named knowledge-base tools let developers scope sources for each task, and GPT-6 knowledge-base profiles give teams another model choice for planning and answer generation.

For teams managing model traffic, supported OpenAI-compatible LLM knowledge-base models can connect through Azure API Management in public preview, including Premium v2 gateways. Customers can configure gateway policies for quotas, load balancing, and monitoring. Supported Azure OpenAI embedding connections for knowledge sources, skills and vectorizers also support APIM endpoints. Each workload still needs its own authentication and connectivity setup.

Choose your next step

  • Build with GA capabilities: Start with a SharePoint (indexed) library and an employee question. Ingest permissions and verify what different users can retrieve.
  • OneLake permission preservation takes a different approach from indexed SharePoint. In public preview, the index retains source locations so OneLake can evaluate the requesting user’s access at retrieval time. That extends permission-aware grounding into Fabric without treating indexed access-control lists and source-side checks as the same security model. Extend the evidence: Add indexed Azure SQL or File knowledge sources when the task spans systems, and apply the access controls appropriate to each source.
  • Preview capabilities also change how an agent finds evidence. Microsoft Retriever 1 offers a specialized retrieval model; knowledge graph retrieval helps answer questions about relationships; and named knowledge-base tools let developers scope the sources used for a task. GPT-6 knowledge-base profiles add another model choice for query planning and answer generation. Make answers more useful: Configure Content Understanding for rich documents, then query a knowledge base and inspect its cited response.
  • For teams managing model traffic, supported OpenAI-compatible LLM knowledge-base models can connect through Azure API Management Premium v2 gateways in public preview. Customers can configure gateway policies for quotas, load balancing, and monitoring. Supported Azure OpenAI embedding connections for knowledge sources, skills and vectorizers also support APIM endpoints. Each workload still needs its own authentication and connectivity setup. Evaluate a preview: Pick one scenario from the preview section, review its requirements in What’s new, and test it separately from the GA production path.

What’s next

For the full set of GA and public-preview updates, see:

Updated Nov 18, 2026
Version 1.0