​
🚀 Quick Start

Get your Gemini Live voice agent running in under 5 minutes with this comprehensive setup guide.

​
📋 Prerequisites

Before starting, ensure you have:

Google AI API Key

Access to Google’s Generative AI API with Gemini Live enabled

Twilio Account

For phone call integration (optional for web-only)

Node.js Environment

Node.js 18+ for running the voice service

TixAE Account

Your TixAE workspace with voice features enabled

​
🔑 Step 1: Google AI API Setup

​
Get Your API Key

  1. Visit the Google AI Studio
  2. Create a new API key or use existing one
  3. Important: Ensure your key has access to Gemini Live models
API Key Requirements: Your Google AI API key must have access to the latest Gemini models. Some keys may not have Live API access by default.

​
Verify API Access

Test your API key with this quick verification: 
curl -X POST "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash-exp:generateContent?key=YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{"parts": [{"text": "Hello, are you working?"}]}]
  }'

​
⚙️ Step 2: Configure Your Agent

​
1. Create Your Agent

In your TixAE dashboard:
  1. Navigate to Create Agent → Voice Agent
  2. Choose “Google Gemini Live” as your voice provider
  3. Set up your basic agent configuration

​
2. Model Selection

Choose the correct Gemini Live model: 
{
  "llmConfig": {
    "modelId": "gemini-2.0-flash-exp",
    "provider": "google-gemini-live",
    "temperature": 0.7,
    "maxTokens": 4000
  }
}
Model Recommendations:
  • gemini-2.0-flash-exp: Fastest, most reliable (Recommended)
  • gemini-2.0-flash-live-001: Stable alternative
  • Avoid: gemini-2.5-flash-exp-native-audio-thinking-dialog (tool calling issues)

​
🎙️ Step 3: Voice Configuration

​
System Instructions

Configure your agent’s personality and instructions:
System Prompt Example
You are a helpful AI assistant capable of having natural voice conversations.

Key Guidelines:
- Speak naturally and conversationally
- Use tools when needed to help users
- Keep responses concise for voice interactions
- Ask clarifying questions if unsure

Current date: {current_date}
Available tools: turn_on_lights, get_weather, search_knowledge

​
Voice Settings

Configure these essential voice parameters: 
{
  "speechConfig": {
    "voiceConfig": {
      "prebuiltVoiceConfig": {
        "voiceName": "Puck"
      }
    }
  },
  "responseModalities": ["AUDIO"],
  "inputAudioTranscription": {},
  "outputAudioTranscription": {}
}

​
🔧 Step 4: Environment Setup

​
Environment Variables

Add these to your .env file: 
# Google AI API Configuration
GOOGLE_GENAI_API_KEY=your_google_ai_api_key_here

# Twilio Configuration (if using phone calls)
TWILIO_ACCOUNT_SID=your_twilio_account_sid
TWILIO_AUTH_TOKEN=your_twilio_auth_token

# Optional: Advanced Settings
GEMINI_LIVE_DEBUG=false
ULTRA_FAST_AUDIO=true

​
Package Dependencies

Ensure these packages are installed:
package.json
{
  "dependencies": {
    "@google/genai": "^0.21.0",
    "twilio": "^4.19.0", 
    "ws": "^8.14.0",
    "@langchain/core": "^0.1.0"
  }
}

​
📞 Step 5: Phone Integration (Optional)

​
Twilio Setup

If you want phone call capabilities:
  1. Purchase a phone number in Twilio Console
  2. Configure webhook URL to point to your TixAE endpoint: 
    https://your-domain.com/api/voice/twilio
  3. Set HTTP method to POST

​
WebRTC Setup

For browser-based calling:
  1. Enable WebRTC in your agent settings
  2. Configure STUN/TURN servers if behind NAT
  3. Test browser permissions for microphone access

​
🧪 Step 6: Testing & Verification

​
Test Your Setup

1

API Key Test

Verify your Google AI API key works with Gemini Live models
2

Voice Test

Test basic voice input/output functionality
3

Tool Calling Test

Verify tools can be called during conversation
4

Phone Test

Make a test call to verify end-to-end functionality

​
Debug Commands

Use these commands to troubleshoot: 
# Test Gemini Live connection
node -e "console.log('Testing Gemini Live...'); /* your test code */"

# Check audio pipeline
curl -X POST "your-endpoint/test-audio" -H "Content-Type: audio/wav" --data-binary @test.wav

# Verify tool configuration  
curl -X GET "your-endpoint/tools" -H "Authorization: Bearer YOUR_TOKEN"

​
⚡ Performance Optimization

​
Ultra-Fast Audio Processing

Our implementation includes cutting-edge optimizations:
  • 20ms chunk processing for minimal latency
  • Loop-unrolled resampling (6x speed improvement)
  • Direct memory operations using bit shifts
  • Minimal validation for maximum throughput
Automatic Optimization: These performance enhancements are automatically applied when using TixAE’s Gemini Live integration. No additional configuration required!

​
🚨 Troubleshooting

​
Common Issues

​
🎯 Next Steps

Add Tool Integration

Integrate external APIs and function calling

Performance Optimization

Ultra-fast audio processing and latency tuning

Advanced Configuration

Reconnection logic, language detection, and more

Production Deployment

Deploy your voice agent to production

​
📞 Need Help?

Having trouble with setup?

Check out our troubleshooting guide or contact support for personalized assistance.
Your ultra-fast Gemini Live voice agent is ready to go! 🚀