Agentic RAG connection

Progress® Agentic RAG is a Retrieval-Augmented Generation (RAG) platform offered as SaaS, designed to index and query unstructured data (documents, audio, video, web content, etc.) and return grounded, generative answers via Large Language Models (LLMs). It emphasizes trust, quality metrics, and integration flexibility rather than being just a chat interface.

At a high level, Progress Agentic RAG provides:

• Indexing of unstructured and structured data in multiple formats and languages.
  
• RAG pipeline orchestration: retrieving relevant context and augmenting LLM input to reduce hallucination and increase relevance.
  
• AI agents to automate data ingestion, retrieval, enrichment and reasoning steps.
  
• APIs and SDKs for Python and JS/TS for programmatic use.

The Agentic RAG connector for Sitefinity CMS enables you to index selected content types and make them available to AI systems through Agentic RAG knowledge boxes.

For more information about knowledge boxes, see Agentic RAG Documentation.

PREREQUISITES: A Progress Agentic RAG account is required.

1. Navigate to Administration » Agentic RAG connection.
2. Click Create a connection.
3. Fill out a Connection name - A descriptive name to help you identify this connection later.
4. Select what content types can be used by the Agentic RAG system for AI answers in Index content types...
5. Select sites that can provide content to the Agentic RAG system in Index content from...
6. Add a Knowledge box by entering the following credentials:
   • NucliaDB API endpoint
     To get the NucliaDB API endpoint, open your Agentic RAG knowledge box. From the Home screen, copy the NucliaDB API endpoint and paste it here.
     

     NOTE:  Provide only the base domain part of the URL, before the /api/ component.

   • Knowledge box UID
     Open your Agentic RAG knowledge box and copy the Knowledge box UID from the screen, then paste it here.
   • API key
     Open your Agentic RAG knowledge box and navigate to Advanced » API Keys.
     Create a new API key with Writer permissions and paste it here.
7. Click Create.

Once the connection is established, Sitefinity automatically indexes the selected content types from the specified sites and sends them to the Progress Agentic RAG (PARAG) knowledge box as resources. In PARAG, each indexed Sitefinity item is stored as a resource, and the text extracted from its fields is split into searchable text blocks for retrieval and grounded answers.

The connector can index the following Sitefinity content types:

• Videos
• Static HTML in pages
• News items
• Lists
• List items
• Images
• Events
• Documents
• Content blocks
• Blogs
• Blog posts
• Dynamic content types

For each indexed item, the connector sends the following field types to PARAG:

• Short text fields
• Long text fields
• Number fields
• Tags and categories
• Custom classifications
  

  NOTE: Tags, categories, and custom classifications are supported as of version Sitefinity CMS 15.4.8627.

The indexed content becomes available for:

• AI-powered search queries
• Context-aware responses
• LLM-based content retrieval

By default only administrators can see the Agentic RAG connection page. To change that:

1. Navigate to Administration » Permissions » Feeds and notifications.
2. From the drop-down of providers, select ARAGProvider.
3. Change permissions.

NOTE: If you are on Sitefinity CMS 15.4.8627, only items with view permissions set to Everyone will be sent to the knowledge box. For newer versions, see Permissions in the PARAG connector.
When making third-party calls to the knowledge box, make sure the security field in the request body is populated with an access group. This ensures that the security lookup phase is performed, and content permissions are observed.

To learn about using Agentic RAG with the AI assistant widget, see AI Assistant widget.

You can use the Agentic RAG connection with AI Search widgets.
The AI Search widgets provide a conversational, context-aware search experience powered by Progress Agentic RAG (PARAG). For more information, see AI Search.