Skip to main content
The Visual IDE is a no-code graph-based editor that provides deep control over your AI agents without sacrificing technical depth. Build sophisticated conversation flows through a unified visual interface that eliminates context switching between configuration panels. Visual IDE

​
Overview

While accessible to non-engineers, the Visual IDE exposes granular control over:
  • System prompts and AI behavior
  • Variable states and data flow
  • Tool definitions and integrations
  • Conditional logic and workflows
  • Multi-language conversation paths

​
Core components

​
Scripts

Scripts are the fundamental building blocks of your agent’s behavior. Each script contains:
  • Nodes - Individual conversation steps or actions
  • Edges - Connections defining the flow between nodes
  • Variables - Data storage accessible throughout the script
  • General settings - Metadata like name and description
Scripts can be composed together using the Add Script to Context system tool, allowing you to create modular, reusable conversation components.

​
Nodes

Nodes represent discrete steps in your conversation flow. Each node has:
  • A unique ID for reference
  • A position on the canvas (X, Y coordinates)
  • Type-specific configuration
  • Input and output ports for connections
See Script nodes for detailed information about available node types.

​
Edges

Edges connect nodes together to define conversation flow. Each edge specifies:
  • Source node and port
  • Target node and port
The edge system allows for complex branching logic, including:
  • Conditional routing based on user responses
  • Multiple outcomes from DTMF input
  • Dynamic navigation with GoToNode

​
Variables

Variables store data throughout the conversation lifecycle. Each variable includes:
​
Key
string
required
Alphanumeric identifier with underscores (e.g., customer_pin)Used in Scriban templates as {{ variables.customer_pin }}
​
Type
enum
required
Data type: String, Number, or Boolean
​
DefaultValue
string
Initial value, converted at runtime based on Type
​
IsVisibleToAgent
boolean
default:"true"
If true, the variable is injected into the LLM system promptIf false, hidden from the AI (useful for sensitive data like PINs)
​
IsEditableByAI
boolean
default:"false"
If true, the AI can modify this variable via extraction or tool outputIf false, the variable is read-only
​
Description
object
Multi-language descriptions explaining the variable’s purpose
Use IsVisibleToAgent: false combined with IsEditableByAI: false for secure data collection scenarios where the AI should never see raw values.

​
Building your first script

1

Create a new script

Navigate to your agent and create a new script. Every script automatically includes a Start Node.
2

Add conversation nodes

Drag nodes from the palette onto the canvas:
  • AI Response - What the agent should say
  • User Query - What you expect the user to ask
  • System Tools - Actions like End Call, DTMF Input, Send SMS
3

Connect the flow

Draw edges between nodes to define the conversation path. The flow always starts from the Start Node.
4

Configure node details

Click each node to configure its behavior, including multi-language content and examples.
5

Test your script

Save the script and assign it to an agent, then test via phone call or WebRTC.

​
Multi-language support

The Visual IDE natively supports parallel language contexts. Most content fields (like Response, Query, Examples) are multi-language dictionaries: 
{
  "en": "Welcome to our service. How can I help you today?",
  "ar": "مرحباً بك في خدمتنا. كيف يمكنني مساعدتك اليوم؟",
  "es": "Bienvenido a nuestro servicio. ¿Cómo puedo ayudarte hoy?"
}
The agent automatically switches between languages mid-conversation, loading completely different neural configurations, TTS voices, and STT models.

​
Design patterns

​
Modular scripts

Break complex agents into smaller scripts:
  • Main script - Greeting and routing logic
  • FAQ script - Common questions handler
  • Payment script - Secure transaction flow
Use the Add Script to Context tool to inject scripts dynamically based on user needs.

​
Error handling

Always provide fallback paths:
  • Add timeout handling on DTMF Input nodes
  • Create “I didn’t understand” responses
  • Include escape routes to human transfer

​
Variable management

Organize variables by visibility:
  • Public variables - Visible to AI for context
  • Private variables - Hidden from AI, set by deterministic tools
  • Constants - Read-only data like business hours
Never store sensitive data (passwords, full credit card numbers) in variables marked as IsVisibleToAgent: true. Use secure sessions instead.

​
Best practices

  1. Use descriptive node IDs - Makes debugging easier
  2. Provide examples - Help the AI understand expected patterns
  3. Test edge cases - Verify all paths work as expected
  4. Keep scripts focused - One script per major conversation goal
  5. Document variables - Add descriptions for team collaboration

​
Next steps

Agent configuration

Configure personality, integrations, and settings

Script nodes

Learn about all available node types

Action flows

Build deterministic workflows

Secure sessions

Collect sensitive data securely

Build docs developers (and LLMs) love

Get started for freeTalk to us