Overview
Conversation nodes handle the back-and-forth dialogue between your agent and the caller. They can:- Generate dynamic AI responses based on a prompt
- Deliver static pre-written messages
- Use knowledge base content for accurate answers
- Route to other stages based on conversation context
Conversation nodes are colored blue on the canvas.
Configuration
Mode
Choose how the node generates responses:- Prompt Mode
- Static Mode
The AI generates responses dynamically based on your instructions.Best for:
- Open-ended conversations
- Q&A interactions
- Complex decision-making
- Personalized responses
Prompt
When using Prompt Mode, the prompt tells the AI how to behave in this stage. Writing effective prompts:Be Specific
Be Specific
Tell the AI exactly what it should and shouldn’t do.
Set Boundaries
Set Boundaries
Define what’s out of scope to prevent unwanted responses.
Include Context
Include Context
Give the AI background information it needs.
Static Message
When using Static Mode, enter the exact message the agent should speak.Skip Response
Toggle to skip AI response generation. Useful when:- You only need to evaluate transitions
- The response comes from a previous node
- You’re using the node purely for routing
Transitions
Transitions define when and where to route the conversation. Each transition has:- Condition: When this transition should trigger
- Target: Which node to go to
Adding Transitions
- Scroll to the Transitions section in the node config
- Click Add Transition
- Enter a condition in natural language
- Select the target node
Condition Examples
| Condition | Use Case |
|---|---|
| ”The customer wants to speak to a human” | Route to Human Transfer |
| ”The customer is asking about pricing” | Route to Pricing node |
| ”The customer confirms they’re done” | Route to End Call |
| ”The customer mentions an order number” | Route to Order Lookup |
Conditions are evaluated by the AI using natural language understanding. Write them as clear descriptions of when the transition should trigger.
Knowledge Base
Attach a knowledge base folder to enable RAG (Retrieval-Augmented Generation):| Setting | Description |
|---|---|
| Folder | Select which KB folder to use |
| Top-K | Number of results to retrieve (default: 5) |
| Reranking | Enable for better accuracy (default: on) |
Training Examples
Improve AI performance with examples:Free-form Examples
Add industry-specific patterns or edge cases:Conversation Examples
Structured agent/caller exchanges to guide behavior:Example Configurations
General Support Handler
General Support Handler
Product Recommendation
Product Recommendation
Legal Disclaimer
Legal Disclaimer
Best Practices
Single Responsibility
Each conversation node should handle one topic or task. Split complex flows into multiple nodes.
Clear Transitions
Write unambiguous conditions. “Customer wants help” is vague; “Customer asks about returns” is specific.
Test Thoroughly
Test each path through the node. Try edge cases and unexpected inputs.
Use Knowledge Base
For factual information, use the knowledge base rather than hardcoding in prompts.