Draft Documentation

This guide is currently in development. Content may be incomplete or subject to change.

~20 minutes

Flow Steps Reference

Complete reference for all available flow step types. Learn how to configure each step to build powerful multichannel workflows.

Step palette in flow builder

Control Flow Steps

Control flow steps manage the execution path of your workflow, including branching, looping, and timing.

  1. 1

    Start

    Entry point of the flow. Every flow begins with a Start step.
    Settings: Trigger type (manual, scheduled, inbound)

  2. 2

    End

    Exit point of the flow. Marks successful completion.
    Settings: Final status (completed, failed)

  3. 3

    Condition

    Conditional branching based on expressions. Routes to different paths based on data.
    Settings: Condition expression, True/False branches, Multiple conditions (else-if)

  4. 4

    ForEachRow

    Iterate through dataset rows. Executes child steps for each row.
    Settings: Batch size, Delay between rows, Error handling

  5. 5

    ResponseBranch

    Branch based on user response content. Includes auto-reply matching.
    Settings: Branch conditions, Auto-reply patterns, Default branch, Timeout handling

  6. 6

    Delay

    Wait for a specified duration before continuing.
    Settings: Duration (seconds, minutes, hours), Business hours only option

  7. 7

    Parallel

    Execute multiple branches simultaneously.
    Settings: Parallel branches, Wait for all vs. wait for any

Condition Step Example

Condition step configuration
// Example condition expressions
row.balance > 100000           // Numeric comparison
row.status == "active"         // String equality
row.email != null              // Null check
row.score >= 80 && row.verified // Combined conditions

ResponseBranch Step

ResponseBranch with auto-replies

The ResponseBranch step is powerful for handling user replies. Configure:

  • Branch patterns: Keywords or phrases that trigger each branch
  • Auto-replies: Quick responses (1, 2, 3 or A, B, C)
  • AI matching: Use AI to understand intent, not just keywords
  • Default branch: Handle unmatched responses
  • Timeout: What happens if no response received

Messaging Steps

Send and receive messages via WhatsApp Business API.

  1. 1

    SendMessage

    Send a custom text message via WhatsApp.
    Settings: Message text, Variables {{field}}, Channel selection

  2. 2

    SendTemplate

    Send a pre-approved WhatsApp template message.
    Settings: Template selection, Template variables, Header media (optional)

  3. 3

    SendMedia

    Send images, videos, or documents via WhatsApp.
    Settings: Media type, File URL or upload, Caption (optional)

  4. 4

    WaitForResponse

    Wait for customer to reply via WhatsApp.
    Settings: Timeout duration, Timeout action, Store response in variable

SendMessage Configuration

SendMessage configuration

Use variables to personalize messages:

Hello {{name}}!

Your appointment is scheduled for {{appointment_date}}.

Reply with:
1 - Confirm
2 - Reschedule
3 - Cancel

SendTemplate Step

Template selection interface

Important: For first contact with new customers, you must use SendTemplate with a pre-approved WhatsApp template. Custom messages can only be sent within a 24-hour conversation window.

WaitForResponse Step

WaitForResponse timeout settings

Voice Steps

Make and receive AI-powered voice calls with natural conversation capabilities.

  1. 1

    VoiceCall

    Initiate an outbound AI-powered voice call.
    Settings: System prompt (AI instructions), Voice selection, Greeting message, Recording enabled, Max duration, Voice actions

  2. 2

    WaitInboundCall

    Wait for an inbound call on assigned DID.
    Settings: DID/Channel, Timeout, Call routing rules

  3. 3

    VoiceTransfer

    Transfer active call to agent, queue, or external number.
    Settings: Destination type, Destination (queue/number), Announcement, Context headers

  4. 4

    VoiceHangup

    End the active voice call.
    Settings: Final message (optional), Hangup reason

  5. 5

    WaitForVoiceOutcome

    Wait for voice call to complete and capture outcome.
    Settings: Outcome variable, Timeout handling

VoiceCall Configuration

VoiceCall with AI settings

System Prompt

The system prompt defines the AI agent's behavior, personality, and objectives:

You are a friendly customer service representative for Acme Corp.
Your goal is to confirm the customer's appointment.

Customer Information:
- Name: {{name}}
- Appointment: {{appointment_date}}

Guidelines:
- Be polite and professional
- If they want to reschedule, offer available times
- If they cancel, ask for the reason
- Keep the call under 2 minutes

Voice Selection

Available voices from OpenAI:

alloy · echo · fable · onyx · nova · shimmer

Voice Actions

Configure actions the AI can take during the call.Learn more about voice actions →

AI Steps

Leverage AI to classify, extract, and generate content within your flows.

  1. 1

    AIClassification

    Categorize text or responses using AI.
    Settings: Categories list, Input text/variable, Output variable, Confidence threshold

  2. 2

    AIExtraction

    Extract structured data from text using AI.
    Settings: Fields to extract, Field types, Input text, Output variables

  3. 3

    AIGenerate

    Generate AI-powered responses or content.
    Settings: Prompt template, Context variables, Output variable, Max tokens

  4. 4

    AgentExecution

    Run a configured AI agent with tools and memory.
    Settings: Agent selection, Input context, Available tools, Max iterations

AIClassification Example

AI Classification categories

Configure categories that the AI will use to classify responses:

Categories:
- interested: Customer wants to proceed
- not_interested: Customer declines
- callback: Customer wants to be contacted later
- question: Customer has questions
- complaint: Customer has a complaint

AIExtraction Example

AI Extraction fields

Define fields to extract from customer responses:

Fields:
- preferred_date (date): When they want to reschedule
- preferred_time (time): Morning, afternoon, or specific time
- reason (string): Why they're rescheduling
- contact_preference (enum): call, whatsapp, email

Data Steps

Manipulate and store data during flow execution.

  1. 1

    DatasetEnrich

    Write values back to the current dataset row.
    Settings: Field name, Value (static or variable), Append vs. replace

  2. 2

    SetVariable

    Set a context variable for use in later steps.
    Settings: Variable name, Value expression, Scope (row, flow, global)

  3. 3

    Transform

    Map and transform data between formats.
    Settings: Input fields, Transformation rules, Output fields

DatasetEnrich Step

DatasetEnrich configuration

Use DatasetEnrich to save results back to the dataset for later export:

// Example enrichments
classification_result: {{ai_classification}}
extracted_date: {{extracted.preferred_date}}
call_outcome: {{voice_outcome}}
final_status: "completed"

Integration Steps

Connect your flows to external systems and services.

  1. 1

    ApiCall

    Call an external REST API.
    Settings: API integration, Endpoint, Method, Parameters, Response mapping

  2. 2

    WebhookCall

    Trigger a webhook with event data.
    Settings: Webhook URL, Payload template, Headers, Retry policy

  3. 3

    Notification

    Send email or SMS notifications.
    Settings: Channel (email/SMS), Recipients, Subject/Message, Template

ApiCall Configuration

ApiCall configuration

Configure API calls to integrate with external systems:

  • Select integration: Choose from configured API credentials
  • Map parameters: Pass flow variables to API parameters
  • Map response: Extract API response data to flow variables
  • Error handling: Define behavior on API errors

Best Practices

Always handle timeouts
Configure timeout behaviors for WaitForResponse and WaitInboundCall steps to prevent rows from being stuck indefinitely.

Use DatasetEnrich liberally
Save intermediate results to the dataset row so you can export complete data after execution.

Test with small datasets first
Before running large campaigns, test your flow with 5-10 contacts to verify all paths work correctly.

Handle the default branch
In ResponseBranch and Condition steps, always configure a default/else branch to handle unexpected inputs gracefully.