Skip to main content

​
Knowledge Sources in Agentic Retrieval

A knowledge source specifies content used for agentic retrieval. It either encapsulates a search index or provides a direct connection to remote data.

​
Knowledge Source Types

​
Indexed Sources

Content stored in Azure AI Search indexes:
  • Search Index: Existing search index with text and vector fields
  • Azure Blob: Auto-generated pipeline from blob containers
  • OneLake: Auto-generated pipeline from lakehouse
  • Indexed SharePoint: Crawled and indexed SharePoint sites

​
Remote Sources

Live data queried at retrieval time:
  • Remote SharePoint: Direct SharePoint Graph API queries
  • Web (Bing): Real-time web search results

​
Creating Knowledge Sources

{
  "name": "product-docs",
  "type": "searchIndex",
  "indexName": "products-index",
  "description": "Product documentation and specifications",
  "alwaysQuery": false
}

​
Usage Control

  • alwaysQuery: Include in every query (true/false)
  • retrievalInstructions: Guidance for when to use
  • description: Helps LLM select appropriate sources

​
Next Steps

See Create Knowledge Base to combine sources into a knowledge base.