Replace Words in Response

The Replace Words in Response node allows you to define multiple find-and-replace rules to modify your LLM responses. Each replacement can use either plain text or regex patterns, with support for case-sensitive matching.

​

Functionality

The Replace Words in Response node acts as a text transformer, finding and replacing specified patterns in your LLM responses. This can be useful for:

• Response Formatting: Standardize output formats and styling
• Content Filtering: Remove or modify unwanted content patterns
• Brand Consistency: Ensure consistent terminology in responses
• Data Transformation: Convert units or standardize numerical formats
• Response Sanitization: Clean up or modify AI-generated content

​

Node Properties

For each replacement rule you add, you can configure:

• Find:
  • Text to Find: The pattern or text to search for in the response
  • Use Regex: Toggle to enable regular expression pattern matching
  • Case Sensitive: Toggle to make the pattern matching case sensitive
• Replace:
  • Replacement Text: The text to insert in place of matched content

​

Usage Examples

​

Scenario: Response Standardization

Let’s say you want to ensure consistent terminology in AI responses:

1. Add a Replace Words in Response node after your LLM node
2. Add multiple replacements:
   • Find: “AI assistant”, Replace: “Virtual Assistant”
   • Find: “artificial intelligence”, Replace: “AI”
   • Find: “I think”, Replace: “Based on the available information”
3. Enable “Case sensitive” where appropriate
4. This helps maintain consistent branding and professional tone in responses

​

Scenario: Response Sanitization

Let’s say you want to remove or mask sensitive information that the LLM might inadvertently include:

1. Add a Replace Words in Response node after your LLM node
2. Enable “Use regex”
3. Configure a replacement:
   • Find: /[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}/
   • Replace: “[REDACTED EMAIL]”
4. This will replace any email addresses in the response with “[REDACTED EMAIL]“

​

Tips and Best Practices

• Place this node immediately after LLM nodes to process their outputs
• Order your replacements carefully as they are applied sequentially
• Use regex for complex pattern matching needs
• Test replacements with various response types to ensure desired outcomes
• Consider case sensitivity requirements for your use case
• Monitor replaced content to ensure accuracy
• Keep replacement rules simple and focused for better maintenance
• Regularly review and update rules as LLM response patterns evolve