Learn how AI personas work and how to customize them
A persona in Anam is an AI-powered digital human that can have natural, real-time conversations with your users. Think of it as a customizable avatar with a brain, voice, and personality.
Every persona is defined by a configuration object that brings together three essential components. Here’s what a complete persona config looks like:
Copy
Ask AI
const personaConfig = { name: "Cara", avatarId: "30fa96d0-26c4-4e55-94a0-517025942e18", voiceId: "6bfbe25a-979d-40f3-a92b-5394170af54b", brainType: "ANAM_GPT_4O_MINI_V1", 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."};
This single configuration object defines everything about your persona - how they look, sound, and behave. Let’s break down each component and how it maps to the configuration.
The name of your persona. This is optional but recommended for internal organization and debugging. It doesn’t affect how the persona behaves, but helps you identify different personas in logs and analytics.
Copy
Ask AI
const personaConfig = { name: "Cara", avatarId: "30fa96d0-26c4-4e55-94a0-517025942e18", voiceId: "6bfbe25a-979d-40f3-a92b-5394170af54b", brainType: "ANAM_GPT_4O_MINI_V1", 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."};
The speech synthesis that brings your persona to life. Different voices convey different personalities and work better for different use cases. You can sample some of the available voices in our voice gallery.
Copy
Ask AI
const personaConfig = { name: "Cara", avatarId: "30fa96d0-26c4-4e55-94a0-517025942e18", voiceId: "6bfbe25a-979d-40f3-a92b-5394170af54b", brainType: "ANAM_GPT_4O_MINI_V1", 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."};
Choose the brain type that best fits your use case and performance requirements:
Best for: Most applications
A fast, cost-effective brain powered by GPT-4o Mini. Offers excellent performance for conversational AI with quick response times and reasonable costs.
Copy
Ask AI
brainType: "ANAM_GPT_4O_MINI_V1"
Best for: Most applications
A fast, cost-effective brain powered by GPT-4o Mini. Offers excellent performance for conversational AI with quick response times and reasonable costs.
Copy
Ask AI
brainType: "ANAM_GPT_4O_MINI_V1"
Best for: Complex reasoning
A powerful brain using Llama 3.3 70B for applications requiring advanced reasoning capabilities and nuanced conversations.
Copy
Ask AI
brainType: "ANAM_LLAMA_v3_3_70B_V1"
Coming Soon
Integration with OpenAI’s Advanced Voice Mode for ultra-realistic voice interactions.
Copy
Ask AI
brainType: "OPENAI_ADVANCED_VOICE" // Coming soon
This brain type is not yet available. Use ANAM_GPT_4O_MINI_V1 for similar capabilities.
Best for: Custom AI integration
Disables Anam’s built-in AI brains, allowing you to integrate your own AI models and processing logic.
Copy
Ask AI
brainType: "CUSTOMER_CLIENT_V1"
When using CUSTOMER_CLIENT_V1, you’ll handle AI responses through your own backend and use Anam purely for avatar rendering and voice synthesis. See the custom LLM guide for more information.
The easiest way to experiment with the available avatars, voices, and brain options is to use the playground in the Anam Lab.
From the build page, you can use the settings menu to change the available options and preview over 400 available voices. When you’re happy with your configuration, you can copy the avatar ID and voice ID from the settings bar to use in your own application.
Creating an effective persona requires thoughtful design across multiple dimensions. When you create a persona, the system prompt you choose defines how the persona will respond to your users. Focus on crafting a well-defined identity, personality, and communication style.
Your system prompt is crucial - it defines your persona’s personality and behavior.
Copy
Ask AI
// ❌ Too vaguesystemPrompt: "You are helpful."// ✅ Specific and actionable systemPrompt: `You are Marcus, a fitness coach with 10 years of experience. You're enthusiastic but not pushy, and you always ask about injuries or limitations before suggesting exercises. Keep responses under 50 words unless explaining a complex exercise.`
Copy
Ask AI
// ❌ No personalitysystemPrompt: "Answer customer service questions."// ✅ Clear personality and boundariessystemPrompt: `You are Emma, a customer support specialist for SoftwareCorp. You're patient, empathetic, and solution-focused. You can help with account issues, billing questions, and technical troubleshooting. If you can't solve something, always offer to escalate to a human specialist.`
The way your persona communicates is just as important as what it communicates. Consider:
Tone of voice: Professional, friendly, casual, or authoritative
Communication style: Concise vs detailed, formal vs informal
User addressing: How it should greet and interact with users
Cultural sensitivity: Adaptability to different user backgrounds
Your persona should maintain consistency while adapting to user needs. For example:
Copy
Ask AI
systemPrompt: `You are Alex, a friendly customer service representative. Your communication style:- Use a warm but professional tone- Address users respectfully - Break down complex information into simple steps- Always acknowledge user frustration with empathyExample response: "I understand you're having trouble with the login process. Let me help you resolve this step by step. First, could you tell me what error message you're seeing?"`
"If you're not completely sure about something, acknowledge it and ask forclarification. It's better to say 'I want to make sure I understand yourquestion correctly' than to provide incorrect information."
"Keep responses concise and relevant. Break down complex information intodigestible parts. Use numbered steps for procedures and bullet points forlists of options."
Personas can be sensitive to even small changes in the system prompt. If your persona isn’t performing as expected, try small incremental changes rather than major overhauls.
Iterative improvement process:
Start simple: Begin with basic functionality and expand gradually
Use example interactions: Include specific examples in your system prompt
Monitor performance: Track how well the persona handles real conversations
Collect feedback: Gather input from users and team members
Make incremental adjustments: Small changes often yield better results than major rewrites
You can control how long a persona session remains active by setting the optional maxSessionLengthSeconds parameter in your persona configuration. This is useful for managing costs, ensuring security, or creating time-limited experiences.
Copy
Ask AI
const personaConfig = { name: "Cara", avatarId: "30fa96d0-26c4-4e55-94a0-517025942e18", voiceId: "6bfbe25a-979d-40f3-a92b-5394170af54b", brainType: "ANAM_GPT_4O_MINI_V1", systemPrompt: "You are Cara, a helpful customer service representative.", maxSessionLengthSeconds: 300 // 5 minutes};
When you set maxSessionLengthSeconds, the session will automatically end after the specified time limit, regardless of whether the conversation is ongoing. The countdown begins when the persona starts streaming.
If you don’t specify maxSessionLengthSeconds, the session will continue until manually ended or until it reaches the maximum session length limit of 30 minutes.
Set session durations based on your specific use case. Consider the typical length of conversations in your domain and add a buffer for natural conclusion.
Factor in conversation flow: Allow enough time for natural conversation completion
Consider user experience: Abrupt session endings can frustrate users, so consider displaying a countdown timer to the user
Monitor usage patterns: Track actual session lengths to optimize your limits
Communicate limits: Let users know when sessions are time-limited
Here’s how to create a persona optimized for sales conversations using the recommended system prompt structure:
Copy
Ask AI
const salesPersonaConfig = { name: "Jordan", avatarId: "professional-avatar-id", voiceId: "confident-voice-id", brainType: "ANAM_GPT_4O_MINI_V1", systemPrompt: `[ROLE]You are Jordan, a consultative sales specialist for SaaS products. You help prospects understand how our solutions can solve their business challenges.[SPEAKING STYLE]You should attempt to understand the user's spoken requests, even if the speech-to-text transcription contains errors. Your responses will be converted to speech using a text-to-speech system. Therefore, your output must be plain, unformatted text.When you receive a transcribed user request:1. Silently correct for likely transcription errors. Focus on the intended meaning, not the literal text. If a word sounds like another word in the given context, infer and correct.2. Provide concise, focused responses that move the conversation forward. Ask one discovery question at a time rather than overwhelming prospects.3. Always prioritize clarity and building trust. Respond in plain text, without any formatting, bullet points, or extra conversational filler.4. Occasionally add natural pauses "..." or conversational elements like "Well" or "You know" to sound more human and less scripted.Your output will be directly converted to speech, so your response should be natural-sounding and appropriate for a sales conversation.[USEFUL CONTEXT]Your sales approach:- Lead with curiosity about their business challenges- Ask 2-3 discovery questions before presenting solutions- Use social proof and case studies when relevant- Address objections with empathy and alternative perspectives- Always respect budget constraints and timeline pressures- Guide toward a demo, trial, or next meeting when appropriate- If you don't know specific product details, acknowledge it and don't make up information`};
Assistant
Responses are generated using AI and may contain mistakes.