Understanding Personas

Learn how AI personas work and how to customize them

A persona is a complete AI agent: a face, a voice, an LLM, and a system prompt packaged together. You configure a persona, create a session token, and stream it to your users.

The persona configuration

Every persona is defined by a single configuration object:

const personaConfig = {
  name: "Cara",
  avatarId: "30fa96d0-26c4-4e55-94a0-517025942e18",
  avatarModel: "cara-4",
  voiceId: "6bfbe25a-979d-40f3-a92b-5394170af54b",
  llmId: "a7cf662c-2ace-4de1-a21e-ef0fbf144bb7",
  systemPrompt:
    "You are Cara, a helpful customer service representative. You're friendly, knowledgeable, and always try to solve problems efficiently. Keep responses conversational and under 50 words unless explaining something complex.",
};

Each field maps to a component of the persona:

Field	What it controls	Learn more
name	Internal label for logs and analytics	Optional
avatarId	The face and expressions users see	Avatar Gallery, Custom Avatars
avatarModel	Which avatar model renders the persona	Models, Video Options
voiceId	How the persona sounds	Voice Gallery, Custom Voices
llmId	Which language model powers the conversation	Available LLMs, Custom LLMs
systemPrompt	Personality, behavior, and instructions	Prompting Guide

You can also extend a persona with a Knowledge Base for document-grounded answers, and Tools for taking actions during a conversation. Personas aren’t limited to your own app either: you can invite them to join your video meetings.

Persona lifecycle

1. Configure

   Define the persona’s appearance, voice, and personality in a configuration object.

2. Create a session token

   Your server exchanges the persona configuration and your API key for a short-lived session token.

3. Stream

   The client uses the session token to start the avatar stream and begin the conversation.

4. Conversation

   Real-time back-and-forth between the persona and the user.

5. Session end

   The conversation ends and the session is cleaned up.

Experimenting in the Lab

The easiest way to try different avatars, voices, and LLMs is the playground in Anam Lab. From the build page you can swap components, preview over 400 voices, and copy IDs into your own code when you’re happy with the result.

Next steps

Prompting Guide

Write system prompts that produce natural, conversational personas.

Session Options

Configure session duration, voice detection, and greeting behavior.