What Is a PAL?

A PAL (Personified Application Layer) is the behavior configuration for your AI agent in Conversational Video Interface (CVI). It defines how the agent thinks, speaks, and acts during a conversation. The Face is separate: it is the on-screen likeness and default voice. A Conversation is the live session that puts a PAL and a Face together over WebRTC.

Piece	What it is	What you configure
PAL	Behavior, knowledge, tools, skills, and pipeline layers	System prompt, objectives, guardrails, LLM, perception, TTS overrides, tools, skills
Face	Visual appearance and default voice	Stock face or a trained face
Conversation	One live call	Which PAL and Face join, plus session options

Why the name matters

Docs and the API use PAL for the behavior object and Face for visual appearance and voice. Older docs and some endpoints still say persona and replica. Same ideas: PAL owns prompts, layers, tools, and skills; Face is the on-screen likeness and voice. Prefer PAL and Face in new work. Legacy names still work on existing endpoints and request fields (/v2/personas, /v2/replicas, persona_id, replica_id, and related aliases) for backward compatibility.

Where to go next

PAL Overview - fields, layers, and customization options
What Is CVI? - how PAL, Face, and Conversation fit in the full pipeline
CVI Quickstart - create a PAL and start a conversation

Getting Started

Build

Deploy

Debug

Guides

Resources

What Is a PAL?

Why the name matters

Where to go next

​Why the name matters

​Where to go next

Why the name matters

Where to go next