> ## Documentation Index
> Fetch the complete documentation index at: https://docs.voicerix.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Conditions

> How the agent decides which path to take after each node.

Conditions are the arrows connecting nodes on the canvas. After each node completes its turn, the agent evaluates all conditions attached to that node and follows the one that best matches what happened in the conversation.

Every condition has three fields:

## Condition Label

A short name for this path — shown on the canvas edge and in call logs. Keep it clear and specific so you can read the call flow at a glance.

Examples: `Customer agreed`, `Wrong number`, `Move to Main Agenda`

## Condition

A natural language description of when the agent should take this path. The AI reads the conversation and picks the condition whose description best matches the outcome of the current node.

Write it as a clear statement, not a vague one. The more specific the condition, the more accurately the agent picks the right path.

Example:

```
Choose this pathway when you have introduced yourself, confirmed you are
speaking with the right person, and the customer is willing to hear more.
```

<Tip>
  If two conditions are similar, add more detail to tell them apart. Vague conditions like "Customer responded" lead to the agent picking the wrong path.
</Tip>

## Transition Speech

Optional text or audio the agent speaks right before moving to the next node. This plays while the next node is loading — it fills the brief pause so the caller does not experience silence.

* **Text** — The agent speaks this line using its voice
* **Audio** — A pre-recorded clip is played instead

Transition Speech is **not** added to the conversation context, so the next node does not "remember" it. Use it only as a natural filler line.

Example: `Sure, let me check that for you.`

<Warning>
  Transition Speech is spoken exactly as written. For agents that handle multiple languages, phrase it carefully so it works across all expected languages.
</Warning>