Agent is a common word used in the doc, and here’s a wiki on what it is, and how you can customize it.

What’s AI Voice Agents

An AI voice agent is a conversational entity equipped with customizable settings, allowing it to assume a specific identity and interact with users to achieve predefined goals.

Customization Options

Check Create Agent for detailed API spec.

Customize Agent Response (LLM)

The text responses of the agent is generated usually from a LLM model. We recommend using Retell LLM to get started. You can also integrate your own LLM or other response systems. Following the previous guide, you will have a Websocket URL to input to agent, where the agent connects to send conversation updates like transcript and get responses from.

You can set up the LLM easily in the dashboard UI for a quicker start.


Check out Language Guide for more details.

Other Customization

  • Voice: Determines the agent’s speech characteristics. Read more about voice selection, and language support at Voice Guide

    • Voice Temperature: Controls how stable / emotional the voice is.
    • Voice Speed: Controls how fast the agent speaks.
  • Speech Controllability: Control and direct how the speech is generated. Read more at Speech Controllability Guide.

  • Speech-to-text

    • Language: Decides what language the speech recognition operates in.
    • Boosted Keywords: Bias speech-to-text model to certain keywords.
  • Interaction Configuration: Controls how agent interacts with user, read more at Interaction Configuration Guide.

    • Who Speaks First: Decides who speaks first in a call.
    • Backchannel: Configure whether to, how often, and what to backchannel.
    • Reminder: Configure whether to, how often to remind user when user is unresponsive.
  • Ambient Sound: Add ambient environment sound to the call to make experience more realistic.

    • Volume: Controls the volume of the ambient sound.
  • LLM Websocket URL: URL to specify what response system (LLM, prompt) it would use, as mentioned in the previous section.

  • Webhook: Specific webhook for the agent so that you can have your server receive real-time updates whenever a call starts or ends, and get retrieve information like transcript, audio, from it. Read more at Webhook Guide.

Feel free to use Update Agent for changing options of an agent. For this API, you don’t need to supply the full agent parameters, you just need to specify the parameters you want to change.

Should I Create One Agent Per Customer / Per Call?

This depends on whether you want the agent to sound differently, respond differently for them. Generally speaking, you would create agent for different scenarios / different customer segments, like one for customer support, one for sales, etc.

You have full control over what the agent says, and various customization around Agent persona, voice, etc.

Usually you don’t need to create an agent per call, as you can customize the prompt for each call, read more at Dynamic Variables.