n8n

VEXYL Voice Gateway with n8n basic configration

vexyl.ai

December 5, 2025

Get your AI voice assistant up and running in minutes with VEXYL Voice Gateway, n8n for workflow automation, Groq for lightning-fast speech-to-text, and Deepgram for natural text-to-speech.

This guide covers the minimal setup to have a working voice AI pipeline connected to your Asterisk PBX.

Prerequisites

Before starting, ensure you have:

Docker and Docker Compose installed
An Asterisk PBX (or FreePBX) with AudioSocket support
API keys from:
- Groq (free tier available)
- Deepgram (free credits on signup)
- Sarvam AI (required for base functionality)
n8n instance (self-hosted or cloud)

Architecture Overview

Here’s how the components work together:


┌─────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   Phone     │     │     VEXYL       │     │      n8n        │
│   Call      │────>│  Voice Gateway  │────>│   AI Workflow   │
│             │     │                 │     │                 │
└─────────────┘     └─────────────────┘     └─────────────────┘
                            │                       │
                     Groq STT (fast)          Your AI Logic
                     Deepgram TTS             (OpenAI, etc.)
                            │                       │
                            <───────────────────────┘

Step 1: Create Your Project Directory

Create a directory for your VEXYL deployment:

mkdir vexyl-voice-gateway
cd vexyl-voice-gateway

mkdir vexyl-voice-gateway
cd vexyl-voice-gateway

Step 2: Create the Environment File

Create a .env file with your configuration:

# .env - VEXYL Voice Gateway Configuration

# === REQUIRED API KEYS ===
SARVAM_API_KEY=your_sarvam_api_key_here
GROQ_API_KEY=your_groq_api_key_here
DEEPGRAM_API_KEY=your_deepgram_api_key_here

# === PROVIDER CONFIGURATION ===
# LLM Provider - using n8n for AI logic
LLM_PROVIDER=n8n

# STT Provider - Groq for fast, accurate transcription
STT_PROVIDER=groq

# TTS Provider - Deepgram for low-latency speech
TTS_PROVIDER=deepgram

# === n8n WEBHOOK CONFIGURATION ===
N8N_WEBHOOK_URL=https://your-n8n-instance.com/webhook/voice-assistant
N8N_TIMEOUT=30000

# === PORT CONFIGURATION ===
# HTTP API port
HTTP_PORT=8081
# AudioSocket port (Asterisk connects here)
AUDIOSOCKET_PORT=8080

# === AUDIO SETTINGS ===
# Buffer duration before processing (ms)
MAX_BUFFER_DURATION=3000
# Enable interruption detection
ENABLE_BARGE_IN=true
BARGE_IN_USE_VAD=true

# === OPTIONAL: TTS CACHING ===
TTS_CACHE_ENABLED=true

# === OPTIONAL: CALL TRANSFER ===
TRANSFER_ENABLED=true
TRANSFER_EXTENSION=operator

# .env - VEXYL Voice Gateway Configuration

# === REQUIRED API KEYS ===
SARVAM_API_KEY=your_sarvam_api_key_here
GROQ_API_KEY=your_groq_api_key_here
DEEPGRAM_API_KEY=your_deepgram_api_key_here

# === PROVIDER CONFIGURATION ===
# LLM Provider - using n8n for AI logic
LLM_PROVIDER=n8n

# STT Provider - Groq for fast, accurate transcription
STT_PROVIDER=groq

# TTS Provider - Deepgram for low-latency speech
TTS_PROVIDER=deepgram

# === n8n WEBHOOK CONFIGURATION ===
N8N_WEBHOOK_URL=https://your-n8n-instance.com/webhook/voice-assistant
N8N_TIMEOUT=30000

# === PORT CONFIGURATION ===
# HTTP API port
HTTP_PORT=8081
# AudioSocket port (Asterisk connects here)
AUDIOSOCKET_PORT=8080

# === AUDIO SETTINGS ===
# Buffer duration before processing (ms)
MAX_BUFFER_DURATION=3000
# Enable interruption detection
ENABLE_BARGE_IN=true
BARGE_IN_USE_VAD=true

# === OPTIONAL: TTS CACHING ===
TTS_CACHE_ENABLED=true

# === OPTIONAL: CALL TRANSFER ===
TRANSFER_ENABLED=true
TRANSFER_EXTENSION=operator

Step 3: Create Docker Compose File

Create a docker-compose.yml file:

version: '3.8'

services:
  vexyl-gateway:
    image: vexyl/vexyl-voice-gateway:latest
    container_name: vexyl-voice-gateway
    restart: unless-stopped
    ports:
      # AudioSocket port - Asterisk connects here
      - "8080:8080"
      # HTTP API port - health checks, session management
      - "8081:8081"
    env_file:
      - .env
    volumes:
      # Persist TTS cache between restarts
      - vexyl-cache:/app/cache
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8081/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s

volumes:
  vexyl-cache:

version: '3.8'

services:
  vexyl-gateway:
    image: vexyl/vexyl-voice-gateway:latest
    container_name: vexyl-voice-gateway
    restart: unless-stopped
    ports:
      # AudioSocket port - Asterisk connects here
      - "8080:8080"
      # HTTP API port - health checks, session management
      - "8081:8081"
    env_file:
      - .env
    volumes:
      # Persist TTS cache between restarts
      - vexyl-cache:/app/cache
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8081/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s

volumes:
  vexyl-cache:

Step 4: Set Up n8n Webhook Workflow

Create a new workflow in n8n with the following structure:

4.1 Add Webhook Node (Trigger)

Add a Webhook node
Set HTTP Method to POST
Set Path to voice-assistant
Copy the Production URL to your .env file

4.2 Add AI Agent Node

Add an OpenAI node (or your preferred LLM) with this system prompt:

You are a helpful voice assistant for [Your Company Name].

IMPORTANT: Keep responses concise (1-2 sentences) as this is a phone conversation.

You can help with:
- Answering questions about our services
- Scheduling appointments
- Providing order status

Guidelines:
- Be friendly and professional
- Ask clarifying questions when needed
- If you cannot help, offer to transfer to a human

When the caller wants to speak to a human, include in your response:
"shouldEscalate": true

When the conversation is complete, include:
"shouldHangup": true

You are a helpful voice assistant for [Your Company Name].

IMPORTANT: Keep responses concise (1-2 sentences) as this is a phone conversation.

You can help with:
- Answering questions about our services
- Scheduling appointments
- Providing order status

Guidelines:
- Be friendly and professional
- Ask clarifying questions when needed
- If you cannot help, offer to transfer to a human

When the caller wants to speak to a human, include in your response:
"shouldEscalate": true

When the conversation is complete, include:
"shouldHangup": true

4.3 Add Respond to Webhook Node

Configure the response to return JSON:

{
  "response": "{{ $json.message.content }}",
  "shouldEscalate": false,
  "shouldHangup": false
}

{
  "response": "{{ $json.message.content }}",
  "shouldEscalate": false,
  "shouldHangup": false
}

Complete n8n Workflow Structure

[Webhook] → [OpenAI Chat] → [Respond to Webhook]

Activate the workflow in n8n before proceeding.

Step 5: Start VEXYL Voice Gateway

Launch the gateway with Docker Compose:

# Pull the latest image
docker compose pull

# Start the gateway
docker compose up -d

# Check logs
docker compose logs -f

# Pull the latest image
docker compose pull

# Start the gateway
docker compose up -d

# Check logs
docker compose logs -f

You should see output like:

🚀 VEXYL Voice Gateway Starting...
📡 HTTP API listening on port 8081
🔌 AudioSocket server listening on port 8080
✅ STT Provider: groq
✅ TTS Provider: deepgram
✅ LLM Provider: n8n
✅ n8n webhook connected

🚀 VEXYL Voice Gateway Starting...
📡 HTTP API listening on port 8081
🔌 AudioSocket server listening on port 8080
✅ STT Provider: groq
✅ TTS Provider: deepgram
✅ LLM Provider: n8n
✅ n8n webhook connected

Step 6: Verify the Setup

Check that everything is working:

# Health check
curl http://localhost:8081/health

# Health check
curl http://localhost:8081/health

Expected response:

{
  "status": "ok",
  "providers": {
    "stt": "groq",
    "tts": "deepgram",
    "llm": "n8n"
  }
}

{
  "status": "ok",
  "providers": {
    "stt": "groq",
    "tts": "deepgram",
    "llm": "n8n"
  }
}

Step 7: Configure Asterisk

Add the AudioSocket application to your Asterisk dialplan.

Edit extensions.conf

[voice-assistant]
exten => 100,1,NoOp(Connecting to VEXYL Voice Gateway)
 same => n,Answer()
 same => n,Set(CHANNEL(audioreadformat)=slin)
 same => n,Set(CHANNEL(audiowriteformat)=slin)
 same => s,n,Set(RAW_UUID=${SHELL(uuidgen)})
 same => s,n,Set(SESSION_UUID=${TRIM(${RAW_UUID})})
 same => s,n,Set(CURL_RESULT=${CURL(http://your-vexyl-server:8081/session/${SESSION_UUID}/metadata,callerid=${CALLERID(num)}&name=${CALLERID(name)}&language_code=$>
 same => s,n,AudioSocket(${SESSION_UUID},your-vexyl-server:8080)
 same => n,Hangup()

[voice-assistant]
exten => 100,1,NoOp(Connecting to VEXYL Voice Gateway)
 same => n,Answer()
 same => n,Set(CHANNEL(audioreadformat)=slin)
 same => n,Set(CHANNEL(audiowriteformat)=slin)
 same => s,n,Set(RAW_UUID=${SHELL(uuidgen)})
 same => s,n,Set(SESSION_UUID=${TRIM(${RAW_UUID})})
 same => s,n,Set(CURL_RESULT=${CURL(http://your-vexyl-server:8081/session/${SESSION_UUID}/metadata,callerid=${CALLERID(num)}&name=${CALLERID(name)}&language_code=$>
 same => s,n,AudioSocket(${SESSION_UUID},your-vexyl-server:8080)
 same => n,Hangup()

Replace your-vexyl-server with the IP or hostname of your Docker host.

For FreePBX Users

Create a Custom Destination pointing to:

[voice-assistant]

[voice-assistant]

Then route an inbound route or IVR option to this destination.

Step 8: Test Your Voice Assistant

Call the extension you configured (e.g., dial 100)
Speak to the AI assistant
Check logs for any issues:

docker compose logs -f vexyl-gateway

docker compose logs -f vexyl-gateway

Environment Variables Reference

Here’s a quick reference of the key environment variables:

Variable	Description	Required
`SARVAM_API_KEY`	Sarvam AI API key	Yes
`GROQ_API_KEY`	Groq API key for STT	Yes*
`DEEPGRAM_API_KEY`	Deepgram API key for TTS	Yes*
`LLM_PROVIDER`	Set to `n8n`	Yes
`STT_PROVIDER`	Set to `groq`	Yes
`TTS_PROVIDER`	Set to `deepgram`	Yes
`N8N_WEBHOOK_URL`	Your n8n webhook URL	Yes
`HTTP_PORT`	API port (default: 8081)	No
`AUDIOSOCKET_PORT`	AudioSocket port (default: 8080)	No

*Required when using that provider

Troubleshooting

No audio response

Check that DEEPGRAM_API_KEY is valid
Verify TTS_PROVIDER is set to deepgram
Check Docker logs for TTS errors

Speech not recognized

Verify GROQ_API_KEY is valid
Ensure STT_PROVIDER is set to groq
Check if audio is reaching the gateway (look for “Audio received” in logs)

n8n not responding

Verify N8N_WEBHOOK_URL is correct and accessible
Ensure the n8n workflow is activated
Test the webhook manually with curl:

curl -X POST https://your-n8n-instance.com/webhook/voice-assistant \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello", "sessionId": "test-123"}'

curl -X POST https://your-n8n-instance.com/webhook/voice-assistant \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello", "sessionId": "test-123"}'

Asterisk connection issues

Ensure port 8080 is accessible from Asterisk
Check Asterisk has AudioSocket module loaded: module show like audiosocket
Verify the dialplan is loaded: dialplan show voice-assistant

Why This Stack?

Groq STT

Ultra-fast transcription (under 1 second)
High accuracy with Whisper models
Generous free tier

Deepgram TTS

Sub-200ms latency
Natural sounding voices
Streaming support

n8n

Visual workflow builder
400+ integrations (CRM, calendars, databases)
Self-hostable and open source
Easy to modify AI logic without code changes

Next Steps

Once your basic setup is working, consider:

Adding Redis for persistent session storage across restarts
Enabling TTS caching to reduce latency and costs
Setting up call transfer to route to human agents
Building advanced n8n workflows with database lookups and API integrations
Adding outbound calling for appointment reminders

Resources

This guide gets you started with a minimal configuration. For production deployments, refer to the complete documentation for security hardening, scaling, and advanced features.