RAG Integration

RAG (Retrieval-Augmented Generation) connects your knowledge base to your agent for accurate, contextual responses.

How RAG Works

1. User asks question: "What's your return policy?"
2. Retrieval: System searches knowledge base for relevant info
3. Augmentation: Retrieved context is added to LLM prompt
4. Generation: LLM generates response using the context

Configuration

Basic Setup

{
  "knowledgeBaseId": "kb_abc123",
  "ragConfig": {
    "topK": 5,
    "minScore": 0.7
  }
}

Advanced Configuration

{
  "ragConfig": {
    "topK": 5,
    "minScore": 0.7,
    "reranking": true,
    "hybridSearch": true
  }
}

System Prompt Integration

Tell your agent how to use retrieved information:

You have access to a knowledge base with company information.

When answering questions:
1. Use the provided context to answer accurately
2. If the context doesn't contain the answer, say so
3. Never make up information not in the context
4. Cite the source when relevant

If no relevant information is found, respond:
"I don't have specific information about that. Would you like me to connect you with a specialist?"

Tuning RAG Performance

Low Recall (Missing Relevant Info)

• Decrease minScore (e.g., 0.5)
• Increase topK (e.g., 10)
• Check document quality

Low Precision (Irrelevant Info)

• Increase minScore (e.g., 0.8)
• Decrease topK (e.g., 3)
• Improve document chunking

Best Practice Scores

Debugging

Enable RAG debugging to see what's being retrieved:

1. Go to Call History
2. Open a call transcript
3. View RAG Context for each turn