Skip to main content
Your AI’s behavior can be fine-tuned through a combination of engine settings and prompt adjustments.

​
Engine Settings Impact

Different AI engine modes affect behavior: Pick the engine in General → Engine type.

​
Dualplex

  • Lowest latency and the most natural turn-taking
  • Voice cloning available
  • Best for: most assistants, and anything where speed is noticed

​
Pipeline

  • Speech is transcribed, the model answers, the reply is spoken
  • The widest choice of voices and transcription tuning
  • There is a pause between turns, which filler audio can cover
  • Best for: assistants that need a specific voice or transcriber

​
Speech-to-speech

  • Native voice generation, no transcription step
  • No filler audio, because there is no gap to fill
How well the assistant reasons comes from the model, not the engine. Change it in Brain & prompt → Model.

​
Common Issues & Solutions

  1. AI Misunderstanding Context
    • Pick a stronger model in Brain & prompt → Model
    • Add more context to your prompt
    • Include example conversations
  2. Unnatural Conversation Flow
    • Enable Filler Audio for smoother transitions
    • Switch to the Dualplex engine for quicker back-and-forth
    • Add conversation examples in your prompt
  3. Repetitive Responses
    • Review and update your system prompt
    • Add variation examples
    • Consider a stronger model in Brain & prompt → Model

​
Optimization Tips

  • Sales Calls: Dualplex, where the fast back-and-forth keeps people engaged
  • Customer Support: a stronger model in Brain & prompt → Model for better problem solving
  • Lead Qualification: Dualplex, or Pipeline with filler audio if you need a particular voice

​
Testing Changes

After adjusting settings:
  1. Save the assistant, then use Test assistant — every test runs on the saved version
  2. Try different conversation scenarios
  3. Monitor call recordings for improvement
Tip: Start with Dualplex and the default model, then adjust based on what you hear on real calls.