Skip to main content

Creating an Agent with Private Knowledge

This guide will walk you through the process of creating a Connect agent that can access and leverage your organization's private knowledge to provide accurate, relevant responses.

What is a Private Knowledge Agent?​

A private knowledge agent is an AI assistant that has been granted access to your organization's specific information, allowing it to:

  • Answer questions based on your internal documents
  • Provide information from your knowledge base
  • Reference your proprietary data when generating responses
  • Combine general knowledge with your specific content

Prerequisites​

Before creating your private knowledge agent, you'll need:

  1. A Connect account with appropriate permissions
  2. Documents you want to make available to the agent (currently limited to 150KB per file)
  3. Clear understanding of what information is appropriate to include

Step 1: Create a New Agent​

  1. Log in to your Connect dashboard
  2. Navigate to the "Agents" section
  3. Click "Create New Agent"
  4. Provide a name for your agent
  5. Enter a system prompt that defines how your agent should behave
  6. Configure basic settings:
    • AI Provider: Select the LLM provider for your agent
    • Optional: Enable rephrasing and select a rephrasing provider
    • Optional: Enable guardrailing and configure guardrailing rules
    • Optional: Select an embedding provider for knowledge processing
    • Optional: Choose a chunking strategy for document processing
  7. Click "Create Agent"

Step 2: Upload Knowledge Documents​

  1. Navigate to your newly created agent's page
  2. Find the "Knowledge Base" section
  3. Click "Upload Document"
  4. Select the file you want to upload (supported formats include PDF, TXT, DOCX, etc.)
  5. The system will automatically process your document:
    • The document will be chunked according to your selected strategy
    • Text will be extracted and vectorized using the chosen embedding provider
    • The knowledge will be indexed for retrieval

Step 3: Monitor Document Processing​

  1. After uploading, your document will be listed with a "pending" or "processing" status
  2. Wait for the status to change to "completed"
  3. If processing fails, you'll see an error message with details
  4. You can retry processing if needed

Step 4: Add Integrations (Optional)​

  1. In your agent's configuration, find the "Integrations" section
  2. Browse available integrations from the marketplace
  3. Select integrations appropriate for your use case

Step 5: Test Your Agent​

  1. Use the Chat interface to test your agent
  2. Ask questions related to the content in your uploaded documents
  3. Verify that the agent provides accurate responses based on your knowledge base
  4. Test edge cases and different query formulations

Step 6: Publish Your Agent​

  1. Once you're satisfied with your agent's performance, toggle the "Published" setting
  2. This makes your agent accessible via public links and API
  3. You can unpublish at any time to restrict access

Best Practices​

Document Preparation​

  • Keep documents focused on specific topics
  • Ensure documents are well-structured for better chunking
  • Consider file size limitations (currently 150KB per file)
  • Use clear formatting that's easy to parse

System Prompt Design​

  • Be specific about how the agent should use the knowledge base
  • Include instructions for handling uncertain information
  • Define the tone and style of responses
  • Establish clear boundaries for what the agent should and shouldn't do

Knowledge Management​

  • Regularly update documents as information changes
  • Remove outdated content
  • Monitor which documents are frequently referenced
  • Add missing information based on user queries

Troubleshooting​

Agent Not Using Knowledge Base​

  • Verify documents were processed successfully
  • Check that questions are related to the document content
  • Review your system prompt for conflicting instructions
  • Try reformulating your questions

Processing Failures​

  • Check file size (must be under 150KB)
  • Ensure file formats are supported
  • Try simplifying complex documents
  • Use the "Retry" option for failed processing

Response Quality Issues​

  • Improve document quality and organization
  • Refine your system prompt
  • Try different chunking strategies
  • Consider using a different embedding provider

By following these steps and best practices, you'll create an AI agent that leverages your organization's specific knowledge to provide valuable, accurate responses.