Getting started with Conversation Flows

Creating conversation flows might seem daunting at first, but in Virbe platform it's designed to be as simple or sophisticated as you need it to be. Think of it like building with blocks – you can start with just two pieces to create a functional conversation, then add more blocks to create increasingly complex and engaging interactions.

Whether you're building a simple customer service bot or a sophisticated virtual assistant, the same basic principles apply.

The beauty of the system is that while a basic flow can handle natural conversations right out of the box, you have the power to add logic, branching paths, and integrations as your needs grow.

Minimal Viable Flow

The simplest functional flow consists of just two elements:

1. Start node

2. LLM Response node

This basic setup is powerful enough to handle open-ended conversations because:

• Each user input triggers the flow

• Conversation context is maintained automatically

• The LLM generates contextually appropriate responses

Add specific system instructions for the LLM node to provide general guidance for the virtual being on how they should approach the conversation with the end user. The node detailed form also accepts additional context – note that System instruction is the most prioritized part of the LLM prompt, additional context bears less weight on the behavior and should be used for less important elements of the prompt.

Building More Complex Flows

Flow Control Mechanisms

As you build more sophisticated flows, two key mechanisms help control conversation progression:

1. Wait for User Input This toggle appears in many nodes and determines whether the flow should:

• Wait for user response before proceeding (toggle on)

• Continue immediately to the next node (toggle off)

1. Checkpoints Checkpoints allow you to:

• Save the conversation state at specific points

• Return to these points upon the next incoming event or input

• Handle complex conversation branches

Example: Guided Conversation Flow

You can use Checkpoint nodes to make a new "starting point" for the conversation, so that each incoming trigger doesn't take the user through the same path from the very beginning. For example, you can use it to make a logical branching based on user's needs:

Start 
  → Intent Matcher (understand user need)
    → Checkpoint (save the status)
      → LLM Response (provide personalized response for user's needs)