Choosing the Intelligence Model
Pick the right intelligence model for your AI Avatar. Speed-vs-depth tradeoff explained, when to swap, and what the model does not change.
Home › Help Center › AI Avatars › Choosing the Intelligence Model
The intelligence model is the "brain" behind every reply your avatar gives. It reads the System Prompt, the conversation history, and the Knowledge Base, then decides what to say. Picking the right one is mostly a speed-vs-depth tradeoff.
What the model actually does
Every time someone asks your avatar a question, the model:
- Reads the System Prompt (the persona's identity and instructions).
- Reads what's been said so far in the current conversation.
- Reads any relevant snippets pulled from the Knowledge Base.
- Decides the next thing the avatar should say.
A faster model does this in milliseconds; a more thorough model takes a beat longer but reasons more carefully. For most personas, the difference is invisible to the visitor.
The speed-vs-depth tradeoff
- Faster models — replies feel snappy and conversational. Perfect for support chat, FAQ-style interactions, anything where short responses matter. The recommended default fits here.
- More thorough models — replies take a moment longer but handle multi-step reasoning, complex policies, edge cases better. Pick these when your persona has to weigh several rules at once (medical-adjacent flows, complex pricing, technical troubleshooting).
Start with the recommended default. Switch only if you actually notice the avatar making the same mistake repeatedly — that's a sign you need a more thorough model. Don't pre-optimize.
What the model does not do
- It doesn't change the voice — that comes from the Replica.
- It doesn't change the face or lip-sync — also Replica.
- It doesn't change the language. Language is its own setting.
- It doesn't override Guardrails — those are checked separately.
What if I want to use my own model?
We support that on request — typically for enterprise customers with custom-trained models or strict data-routing requirements. Contact us through the support page if this is a need.
You can change the intelligence model anytime — it's not baked into the Replica. Test with one, switch to another, compare. The persona keeps the same face, voice, prompt, and attachments.
Related articles in AI Avatars
Help Center · Learn · Privacy Policy · Terms of Service · Contact