Profile

A Profile defines how your AI worker communicates—its name, voice, tone, and conversation style. Think of it as giving your worker a distinct character that resonates with your users.


Overview

Every worker needs a Profile. It controls:

  • Who the worker is - Name and purpose

  • How it sounds - Voice selection, speech speed, and audio settings

  • How it communicates - Tone, formality, and conversation style

  • What model powers it - LLM provider and configuration

You can create multiple Profiles for different use cases—a friendly support worker, a professional sales representative, or a concise technical assistant.


Creating a Profile

Navigate to Worker Setup > Profile and click Create Profile.

The configuration is organized into four tabs:

  • Identity & Brand Voice

  • Conversation Style

  • Intelligence

  • Voice & Runtime


Identity & Brand Voice

Defines who your worker is and how it presents itself.

Basic Information:

Field
Description

Worker Name

The name users see (e.g., "Sarah", "Alex", "Support Worker")

Description

Internal note describing this Profile's purpose

Training Keyword

Identifier used when generating training data

Primary Language

Default language for conversations (e.g., en-US)

Character & Communication:

Field
Options
Description

Actor Type

Conversational, Worker, Conversational + Worker, Orchestrator, Policy Manager

Determines worker capabilities

Modality

Voice + Chat, Chat Only, Voice Only

Communication channels supported

Tone

Friendly, Professional, Casual, Authoritative

Overall communication style

Formality

Low, Medium, High

Level of formal language

Jargon Level

None, Light, Technical

Industry terminology usage

Opening Greeting

Free text

First message when conversation starts

Actor Types Explained:

  • Conversational - Customer-facing dialogue worker

  • Worker - Background task execution without direct user interaction

  • Conversational + Worker - Handles dialogue and executes tasks

  • Orchestrator - Coordinates multiple workers (meta-agent)

  • Policy Manager - Monitors compliance across conversations (meta-agent)


Conversation Style

Fine-tune how the worker engages in dialogue.

Parameter
Range/Options
What It Controls

Cadence

1-10

Pace of responses (slower to faster)

Verbosity

Terse, Balanced, Detailed

Response length

Empathy

Low, Medium, High

Emotional responsiveness

Initiative

Reactive, Suggestive, Proactive

How actively worker guides conversation

Summarization

None, Periodic, Frequent

How often worker recaps information

Creativity

0.0-1.0

Variation in responses

Humor

None, Light, Moderate

Use of humor in responses

Emoji Use

None, Light, Moderate

Emoji frequency (chat only)


Intelligence

Select the AI model that powers your worker.

Execution Modes:

Mode
Description
Use Case

LiveKit Managed

Access to multiple providers through LiveKit

Standard deployments

Trainable

Models that support fine-tuning

Custom-trained workers

Bedrock

AWS Bedrock integration

Enterprise with existing AWS

Supported Providers:

Provider
Models

OpenAI

GPT-4o, GPT-4o Mini, GPT-4 Turbo, o1, o3 Mini

Google

Gemini 2.0 Flash, Gemini 1.5 Pro/Flash

DeepSeek

DeepSeek Chat, DeepSeek Reasoner

Anthropic (Bedrock)

Claude 3.5 Sonnet/Haiku, Claude 3 Opus

Amazon (Bedrock)

Nova Pro/Lite/Micro, Titan Text

Meta (Bedrock)

Llama 3.2 90B/11B, Llama 3.1 70B/8B

Temperature controls response creativity (0.0 = deterministic, 1.0 = creative).


Voice & Runtime

Configure voice synthesis and speech recognition. These settings only apply when Modality includes voice.

Text-to-Speech (TTS):

Engine
Characteristics

Cartesia

Ultra-low latency, recommended for real-time

ElevenLabs

Premium voice quality

OpenAI

Balanced quality and speed

Rime

Natural conversational voices

Google Cloud

Enterprise reliability

Azure Speech

Microsoft integration

Each engine provides multiple voice options with different characteristics (gender, accent, tone).

TTS Parameters:

Parameter
Range
Description

Speech Speed

0.5-2.0

Playback rate (1.0 = normal)

Volume

0.0-1.0

Audio output level

Speech-to-Text (STT):

Provider
Models

Deepgram

Nova-2 (recommended), Nova, Enhanced, Base

OpenAI

Whisper-1

AssemblyAI

Best, Nano

Google Cloud

Latest Short/Long, Command & Search

Azure

Multi-language support

Runtime Settings:

Setting
Options
Description

Interruptions

On/Off

Allow user to interrupt worker

Barge Mode

Off, On

Worker interruption behavior

Latency Profile

Low, Balanced, Quality

Speed vs. quality tradeoff

Turn Detection

Server VAD, Client VAD

Who detects speech turns


Modality Guide

Choose the right modality for your channel:

Modality
Best For
Voice Settings

Voice + Chat

Omnichannel support

Full TTS/STT config

Chat Only

Website widget, messaging

Voice settings hidden

Voice Only

Phone, voice assistants

Full TTS/STT config


Best Practices

Naming

  • Use human-sounding names for customer-facing workers ("Sarah", "Alex")

  • Use descriptive names for internal workers ("Order Status Worker")

  • Keep Worker Name under 20 characters

Voice Selection

  • Match voice gender and accent to your target audience

  • Test voice samples before deploying

  • Consider cultural expectations for your market

Tone & Style

  • Support workers: Friendly tone, medium formality, high empathy

  • Sales workers: Professional tone, suggestive initiative

  • Technical workers: Light jargon, balanced verbosity

Model Selection

  • Start with GPT-4o Mini for cost-effective testing

  • Use GPT-4o or Claude for complex reasoning

  • Consider fine-tuning for specialized domains


Impact Preview

Before deleting a Profile, the system checks if it's used by any active workers. If so, you'll see an impact preview showing:

  • Number of active workers using this Profile

  • Whether deletion is blocked or allowed

  • Affected worker names

This prevents accidental disruption to live deployments.


  • Jobs - Define the role and responsibilities for your worker

  • Topics & Goals - Structure conversation objectives

  • Actor Graph - Build multi-worker workflows

Last updated