Skip to main content
Transitions define how and when your voice agent moves from one stage to another. They’re the connections (edges) between nodes on the canvas.

​
Overview

A transition consists of:
  • Source: The node where the transition originates
  • Target: The node where the conversation continues
  • Condition: What triggers the transition (optional)
┌─────────────┐                    ┌─────────────┐
│   Source    │ ──── Condition ───►│   Target    │
│    Node     │                    │    Node     │
└─────────────┘                    └─────────────┘

​
Transition Types

​
Any Message

Triggers when the caller says anything. 
Type: any_message
Use cases:
  • Start node → First conversation stage
  • After static messages
  • When any response should continue the flow

​
Immediate

Automatically triggers after the node executes. 
Type: immediate
Use cases:
  • After greeting plays
  • After function execution
  • Chaining nodes without user input

​
Keyword

Triggers when the caller says specific words or phrases. 
Type: keyword
Config:
  keywords: ["cancel", "stop", "quit"]
  caseSensitive: false
  matchType: "contains"  # or "exact"
Use cases:
  • IVR-style menus (“say ‘sales’ for sales”)
  • Escape phrases (“say ‘agent’ for a human”)
  • Command detection

​
All Extracted

Triggers when all required variables have been collected. 
Type: all_extracted
Use cases:
  • After Data Extraction completes
  • Moving to processing after data collection
  • Form completion flows

​
Function Success

Triggers when a function call completes successfully. 
Type: function_success
Use cases:
  • After API lookups
  • After database operations
  • After external integrations

​
Timeout

Triggers after a period of silence. 
Type: timeout
Config:
  seconds: 10
  message: "Are you still there?"
  resetOnInput: true
Use cases:
  • Handling unresponsive callers
  • Re-prompting after silence
  • Ending abandoned calls

​
Variable Comparison

Triggers based on a variable value. 
Type: variable_comparison
Config:
  variable: "customer_tier"
  operator: "=="
  value: "premium"
Use cases:
  • Routing based on collected data
  • Conditional flow paths
  • A/B testing flows

​
Condition Prompt

Uses AI to evaluate a natural language condition. 
Type: condition_prompt
Config:
  condition: "The customer wants to speak to a human agent"
Use cases:
  • Intent detection
  • Sentiment-based routing
  • Complex conditional logic

​
Transition Priority

When multiple transitions could trigger, they’re evaluated in order:
  1. Keyword - Checked first for exact matches
  2. Condition Prompt - AI evaluates intent
  3. Variable Comparison - Checks data values
  4. Timeout - Fires after silence period
  5. Default/Fallback - Catches everything else
If no transition matches, the conversation stays in the current node. Always ensure at least one transition can fire.

​
Creating Transitions

​
From Conversation Nodes

  1. Open a Conversation node
  2. Scroll to Transitions section
  3. Click Add Transition
  4. Enter a condition (natural language)
  5. Select the target node
Condition: "The customer asks about returns or refunds"
Target: Returns Handler

​
From the Canvas

  1. Click the source handle (bottom of node)
  2. Drag to the target handle (top of node)
  3. Release to create the connection
  4. Click the edge to configure

​
Edge Labels

Edges display their condition as a label. Click the label to edit.

​
Transition Conditions

​
Writing Effective Conditions

Good: “The customer explicitly asks to speak with a human agent”Bad: “The customer needs help”
Good: “The customer says goodbye, thanks you, or indicates they’re done”Bad: “The customer says bye”
Good: “The customer provides an order number in format XX-123456”Bad: “The customer gives a number”

​
Condition Examples

ScenarioCondition
Transfer request”The customer asks to speak with a human, manager, or real person”
End conversation”The customer says goodbye, indicates they’re done, or thanks the agent”
Billing inquiry”The customer asks about charges, billing, invoices, or payments”
Escalation”The customer expresses frustration, anger, or threatens to cancel”
Confirmation”The customer confirms, agrees, or says yes”
Denial”The customer declines, disagrees, or says no”

​
Multiple Transitions

Nodes can have multiple outgoing transitions: 
┌─────────────────┐
│  Conversation   │
└────────┬────────┘
         │
    ┌────┴────┬────────────┬────────────┐
    │         │            │            │
    ▼         ▼            ▼            ▼
[Returns] [Billing] [Transfer] [End Call]

Conditions:
- "asks about returns" → Returns
- "asks about billing" → Billing
- "wants human" → Transfer
- "says goodbye" → End Call

​
Default Transitions

Some nodes require or benefit from a default path:
  • If/Else: Must have a default for unmatched conditions
  • Conversation: Consider a default for unexpected intents
  • Data Extraction: Handle max attempts exceeded

​
Testing Transitions

  1. Open the Test panel
  2. Send messages that should trigger each transition
  3. Watch the canvas - active transitions highlight in green
  4. Verify the flow moves to the expected node
Test edge cases: What happens with unexpected input? Does the flow handle it gracefully?

​
Best Practices

One Clear Path

For each user intent, there should be exactly one matching transition.

Handle Everything

Ensure every possible input has a path forward.

Test Thoroughly

Test each transition with multiple phrasings.

Keep Conditions Simple

Complex conditions are harder to maintain and debug.

​
Debugging Transitions

If transitions aren’t firing as expected:
  1. Check condition wording - Is it too specific or ambiguous?
  2. Test in isolation - Does the transition work on its own?
  3. Check order - Is another transition firing first?
  4. Review logs - What did the AI interpret?

Conversation Stage

Primary node for transitions

If/Else Stage

Variable-based routing

Flow Canvas

Visual transition management