Unified RAG (Retrieval-Augmented Generation) Integration

Overview

Unified RAG provides a built-in knowledge base system that enables agents to retrieve and respond using structured enterprise data. It enhances conversational intelligence by grounding responses in domain-specific content.

Key Benefits

• Knowledge-Driven Responses: Improves accuracy and relevance

• Centralized Knowledge Management: Upload and manage documents easily

• Multi-Document Retrieval: Query across multiple documents in a single knowledge base

Core Capabilities

• Create and manage knowledge bases

• Upload and index multiple documents

• Attach knowledge bases to agents

• Retrieve contextual responses based on user queries

Functional Behavior

Agent queries:

• Knowledge Base (primary source)

• LLM fallback (if enabled)

Supports:

• Cross-document search within a knowledge base

• Prompt-based control to restrict or allow LLM fallback

Setup Workflow

1. Create a knowledge base “Dentist Data” as example:

• Click on “New Knowledge Base”:

• Upload documents

Provide:

1. Knowledge Base name

2. Description

3. Option to share organization-wide

4. Advanced settings:

The Advanced Settings control how your knowledge base is indexed and searched, directly impacting the quality and efficiency of AI responses.

• Embedding: Embedding section defines how the platform converts your uploaded documents into a format that can be searched and retrieved efficiently during RAG (Retrieval-Augmented Generation). Embedding is a critical step in knowledge base creation, as it transforms text into vector representations that enable semantic search.

• Provider name: example: azure-openai indicates that embeddings are generated using Azure-hosted OpenAI models

• Model ID: Specifies the embedding model used to convert textual content

Upload knowledge sources by clicking on “Add Data Source” button and selecting the file from folders:

• View documents in KB “Dentist Data”:

• Select data sources to delete or re-index:

• Option to edit Knowledge Base details and Indexing Settings

• Indexing Settings:

• Chunk Size: This controls how big each “piece” of your document is when the system reads it.

• Chunk Overlap: This controls how much content overlaps between two chunks so nothing important is missed.

• Batch Size: This controls how many chunks are processed at the same time during setup.

• Enable LLM Moderation: This checks content for unsafe, inappropriate, or restricted information.

• Enable Redaction: This automatically removes or hides sensitive information (like names, phone numbers, etc.) from your data. User can define define what entities need to be hidden:

1. Attach knowledge base “Dentist Data” to an agent in flow

1. Configure agent instructions

Notes / Limitations

• Only one knowledge base per agent (current release)

• Multiple documents allowed within a knowledge base

• Multi-tool support (KB + APIs) is not yet available