Flows

Flows & Dialogs

Build conversation flows for virtual assistants.

---

Overview

Dialog flows define how your virtual assistant handles conversations:

• Intents — What users want to accomplish
• Entities — Information extracted from messages
• Dialogs — Conversation logic and responses
• Transitions — How conversations move between states

---

Dialog Types

Standard Dialog

Structured conversation flow for specific tasks:

[User: Check order status]
         │
         ▼
[Collect Order Number]
         │
         ▼
[Validate Order Number]
         │
    ┌────┴────┐
    ▼         ▼
[Valid]   [Invalid]
    │         │
    ▼         ▼
[Fetch    [Request
 Status]   Again]
    │
    ▼
[Deliver Response]

Agent Flow

Complex flows with AI agent capabilities:

Agent Flow: Research Assistant
Capabilities:
  - search_web
  - analyze_documents
  - summarize_findings
Behavior: |
  Research the user's question using available tools.
  Synthesize findings into a comprehensive answer.
  Cite sources when providing information.

Sub-Dialog

Reusable dialog components:

Sub-Dialog: Collect Address
Entities required:
  - street_address
  - city
  - state
  - zip_code
Validation: address_verification_api
Reusable: true

---

Building Dialogs

Dialog Builder

Visual drag-and-drop builder:

Node Type	Purpose
Message	Display text, cards, or media
Prompt	Collect user input
Condition	Branch based on conditions
API Call	Execute integrations
Script	Run custom code
Agent	Invoke AI agent
Transfer	Hand off to human

Node Configuration

Message node example:

Node: Welcome Message
Type: message
Content:
  text: "Hello! I'm your order assistant. How can I help you today?"
  quick_replies:
    - "Check order status"
    - "Start a return"
    - "Track my package"
Transitions:
  - intent: check_order → order_status_flow
  - intent: start_return → return_flow
  - default → fallback

Transitions

Define conversation flow:

Transition Type	Description
Intent	Move based on detected intent
Entity	Move based on extracted entity
Condition	Move based on variable value
Default	Fallback when no match
Error	Handle errors gracefully

---

Entity Management

Entity Types

Type	Use Case	Example
Built-in	Common data types	Date, Number, Email
List	Defined options	Product names, sizes
Regex	Pattern matching	Order ID: ORD-\d
Composite	Grouped entities	Full address

Entity Extraction

Configure extraction:

Entity: OrderNumber
Type: regex
Pattern: "(?:order|ORD)?[-#]?(\d{5,8})"
Validation: |
  length >= 5 AND length <= 8
  starts_with_digit
Prompts:
  initial: "What's your order number?"
  retry: "I didn't catch that. Order numbers are usually 5-8 digits."
Max retries: 3

Entity Disambiguation

Handle ambiguous inputs:

Disambiguation: Size
Values: ["small", "medium", "large"]
On ambiguity:
  type: confirm
  message: "Did you mean {{matched_value}}?"
On multiple:
  type: select
  message: "Which size? Small, medium, or large?"

---

Context Management

Context Variables

Store and use conversation context:

// Set context
context.session.orderNumber = "12345"
context.session.customerTier = "premium"

// Use in messages
"Your order {{context.session.orderNumber}} is being processed."

// Use in conditions
if (context.session.customerTier === "premium") {
  // Premium customer handling
}

Variable Scopes

Scope	Lifetime	Use Case
Turn	Single turn	Temporary processing
Session	Conversation	Order details, selections
User	Across sessions	Preferences, history
Global	All users	Configuration, flags

---

Interruption Handling

Configuration

Handle topic switches:

Interruption Handling:
  allow_interruptions: true
  hold_and_resume:
    enabled: true
    message: "No problem, let me help with that first."
  implicit_confirmation: true
  max_depth: 3

Scenarios

Scenario	Handling
New intent	Hold current, start new
Clarification	Handle inline, resume
Cancel	Confirm and exit
Go back	Return to previous step

---

Testing

Conversation Testing

Test dialog flows:

1. Open Test panel
2. Enter test utterances
3. Verify intent detection
4. Check entity extraction
5. Validate flow navigation
6. Review responses

Batch Testing

Run test suites:

Test Suite: Order Status Flow
Tests:
  - name: "Valid order number"
    utterance: "Where is my order 123456?"
    expected:
      intent: check_order_status
      entities:
        order_number: "123456"
      response_contains: "order status"

  - name: "Missing order number"
    utterance: "Where is my order?"
    expected:
      intent: check_order_status
      next_prompt: "order number"

Debug Mode

Enable detailed debugging:

• Intent confidence scores
• Entity extraction details
• Flow transitions
• Variable values
• API call results

---

Best Practices

Dialog Design

• Keep flows focused on single tasks
• Use sub-dialogs for reusable components
• Handle errors gracefully
• Provide clear escape paths
• Test edge cases thoroughly

Intent Training

• Add 15-25 utterances per intent
• Include variations and typos
• Use real customer language
• Regularly review misclassifications
• Avoid overlapping intents

Conversation UX

• Confirm understanding before actions
• Provide clear options when ambiguous
• Keep messages concise
• Use quick replies for common paths
• Allow natural conversation flow

---

Related

• Automation AI
• Channels
• Integrations