AriLink

-powered telephony platform with real-time transcription, campaigns, and a full web dashboard

---

Overview

AriLink is a telephony platform built on Asterisk's ARI (Asterisk REST Interface). It combines a Nuxt 4 web dashboard, a Rust RTP audio pipeline, and local AI transcription into a single system for managing calls, running outbound campaigns, and monitoring everything in real time.

Key Features

• LLM Voice Agent - Full conversational AI via LlmChatBrain — streaming LLM responses with sentence-by-sentence TTS
• Barge-In - Interrupt bot speech mid-sentence. AEC3 echo cancellation + Silero VAD eliminate false triggers
• Pluggable Brains - Swap call logic via config: LLM Chat, IVR Transfer, Direct Dial, OpenClaw AI
• Local TTS - Text-to-speech via Kokoro (82M params) — zero API cost, ~100ms latency
• Local STT - Real-time transcription via Parakeet TDT (25 languages) or Whisper, with Google Cloud fallback
• Web Dashboard - Full management UI with real-time service status, active calls, logs, and configuration
• Campaign Engine - Automated outbound dialing with configurable concurrency and call scripts
• OpenClaw Integration - Bridge to OpenClaw AI agent ecosystem via Socket.IO
• Call History - SQLite-backed call records with search and transcription playback
• CLI Tool - arilink init/start/stop/status/logs for Docker deployments
• Built-in Softphone - SIP softphone directly in the dashboard for testing
• SSH Terminal & File Manager - Remote PBX management and audio file uploads via SFTP

Architecture

Built on Nuxt 4 + Nitro (single server for SPA, API, and Socket.IO), with a Rust RTP server for audio and Python services for transcription:

flowchart TD
subgraph Dashboard["Nuxt 4 Dashboard"]
UI[Vue 3 SPA]
API[Nitro API Routes]
SIO[Socket.IO]
end
subgraph Core["Node.js Core"]
CTRL[ARI Controller]
HARNESS[BrainHarness]
BRAINS["Brains<br/><small>LlmChat · IvrTransfer<br/>DirectDial · OpenClaw</small>"]
CAMP[Campaign Engine]
HIST[Call History]
end
subgraph Audio["Rust RTP Server"]
RTP[RTP Receiver :8000]
AEC["AEC3 Echo Cancellation<br/>+ Silero VAD"]
end
subgraph AI["AI Services"]
PKT[Parakeet STT]
WHSP[Whisper]
GCS[Google Cloud Speech]
KOKORO[Kokoro TTS]
end
PBX[Asterisk PBX] <-->|ARI WebSocket| CTRL
PBX -->|RTP Audio| RTP
RTP --> AEC
AEC -->|Clean audio| PKT
AEC -.->|Fallback| WHSP
AEC -.->|Fallback| GCS
AEC -->|user_speaking| CTRL
PKT -->|Transcription| CTRL
CTRL --> HARNESS --> BRAINS
BRAINS -->|speak| KOKORO
KOKORO -->|slin16 audio| PBX
CTRL --> HIST
CAMP --> CTRL
SIO <--> Core
SIO <-->|Real-time| UI
API --> HIST
BRAINS <-.->|Socket.IO| OC[OpenClaw AI]

Loading

Core Components

Engine

Orchestrates all backend services on startup:

• Starts the Rust RTP server and Parakeet transcription (in parallel)
• Connects to Asterisk ARI
• Wires up the Dashboard, Call History, and Campaign Engine
• Graceful shutdown of all child processes

ARI Controller

Interfaces with Asterisk PBX via ARI:

• Manages call flows, bridges, and DTMF input
• Delegates call logic to pluggable assistants
• Routes transcription results to active calls
• Handles contact lookups for voice-activated dialing

Rust RTP Server

High-performance audio pipeline with voice intelligence:

• Receives RTP audio from Asterisk ExternalMedia channels
• AEC3 echo cancellation — WebRTC-grade, 30-40dB attenuation with adaptive threshold
• Silero VAD — neural speech detection during TTS playback (barge-in trigger)
• RMS gate — filters noise below threshold before sending to STT
• Per-session audio routing with codec handling (slin16, Opus)
• Forwards clean audio to transcription services via WebSocket

Text-to-Speech (Kokoro)

Local TTS service for zero-latency, zero-cost speech synthesis:

• Kokoro 82M — lightweight neural TTS model
• Outputs 16kHz slin16 PCM directly for Asterisk playback
• Sentence-by-sentence streaming — start speaking before LLM finishes
• Configurable voice, speed, and language
• WebSocket protocol for low-latency communication

Transcription Providers

Multiple backend support with automatic failover:

• Parakeet TDT 0.6B-v3 (recommended) — 25 languages, 2000x+ real-time speed, runs on GPU
• Whisper — Slower but highly accurate
• Google Cloud Speech — Cloud-based fallback (requires API credentials)
• Streaming transcription with interim and final results
• Priority chain configured via TRANSCRIPTION_SERVICES env var

Pluggable Brains

Modular call logic — swap behavior via config.json without changing code:

• LlmChatBrain — Full conversational AI (streaming LLM + sentence-by-sentence TTS + barge-in)
• IvrTransferBrain — DTMF gate → voice → contact match → transfer
• DirectDialBrain — Voice → contact match → direct transfer
• OpenClawBrain — Bridge to OpenClaw AI agent ecosystem
• BrainHarness — shared voice intelligence (filler filtering, turn debouncing, barge-in handling) benefits all brains automatically
• Create custom brains by implementing the IBrain interface

Campaign Engine

Automated outbound dialing:

• Upload phone lists and run campaigns from the dashboard
• Configurable concurrency, retry logic, and call scripts
• Real-time campaign progress via Socket.IO
• Per-call results stored in campaign-results/

Dashboard Pages

Page	Description
Dashboard	Service status (Asterisk, Rust RTP, Transcription), active calls
Calls	Call history with transcription, search, and playback
Assets	Upload/manage Asterisk audio files via SFTP
Terminal	SSH terminal for remote PBX management
Campaigns	Create and run outbound call campaigns
Contacts	Manage contacts for voice-activated dialing
Assistants	Configure call handling logic per assistant
Logs	Real-time and historical application logs
Config	Edit environment variables and server settings
Docs	Built-in documentation viewer

Getting Started

Docker (Recommended)

The fastest way to try AriLink — no Node.js, Python, or Asterisk installation needed:

docker compose up -d

This starts 4 services: Asterisk (PBX), Parakeet (STT), Kokoro (TTS), and AriLink (dashboard + engine). Open localhost:3011.

Test with a SIP phone (Zoiper, Linphone): server localhost:5060, extension 1001, password demo1001.

# With GPU acceleration (NVIDIA) — ~1000x faster STT + TTS
docker compose -f docker-compose.yml -f docker-compose.gpu.yml up -d
# With Rust RTP server (AEC3 echo cancellation + Silero VAD for barge-in)
docker compose --profile rust-rtp up -d
# With live code editing (mount local source)
docker compose -f docker-compose.yml -f docker-compose.dev.yml up -d

See docs/docker.md for full Docker guide, configuration, and troubleshooting.

Manual Setup

For development or connecting to an existing FreePBX server:

Prerequisites

1. Set up FreePBX server:
   • New installation? FreePBX Installation Guide
   • Already installed? FreePBX ARI Configuration
2. Node.js 23+ and npm
3. UV (Python package manager) for transcription services:

   # Windows PowerShell
   irm https://astral.sh/uv/install.ps1 | iex

Installation

1. Clone and install dependencies (one command installs everything):

   npm install

   This automatically installs both the core server and dashboard dependencies.

2. Set up transcription (local speech recognition):

   cd transcription-services/parakeet-service
   python -m venv .venv
   uv pip install -r requirements.txt

3. Configure environment:

   cp .env.example .env
   # Edit .env with your FreePBX IP, ARI credentials, etc.

Running

# Development (hot reload, auto-starts Parakeet if configured)
npm run dev
# Production (builds if needed, then starts)
npm start
# Build only
npm run build

Set AUTO_START_TRANSCRIPTION=true in .env to have the launcher automatically start Parakeet in parallel with the server. No separate terminal needed.

The dashboard will be available at http://localhost:3011. All settings can be managed from the Config page once the server is running.

Assistant Modes

npm run start:ivr # IVR with transfer (default)
npm run start:dial # Direct dial mode
npm run start:autodialer # Auto-dialer campaign mode

Configuration

The system uses a central .env file. Key settings:

Category	Variables
PBX	PBX_IP, ASTERISK_LOGIN, ASTERISK_PASSWORD
Transcription	TRANSCRIPTION_SERVICES, AUTO_START_TRANSCRIPTION, SPEECH_LANG
TTS	TTS_SERVICE, TTS_VOICE, TTS_SPEED, TTS_LANG
LLM	LLM_PROVIDER, LLM_MODEL, LLM_API_KEY, LLM_ENDPOINT
Rust RTP	USE_RUST_RTP, RUST_SERVER_URL
Server	PORT (default 3011), LISTENER_SERVER, EXTERNAL_HOST
Assistants	DEFAULT_ASSISTANT, per-assistant config.json files

See .env.example for all options with documentation.

Tech Stack

• Frontend: Nuxt 4, Vue 3.5, @nuxt/ui, Socket.IO client, SIP.js (softphone)
• Backend: Nitro (Node.js), TypeScript, Socket.IO, ari-client, better-sqlite3
• Audio: Rust RTP server with AEC3 echo cancellation + Silero VAD (slin16, Opus)
• STT: Parakeet TDT 0.6B-v3 (NeMo), Whisper, Google Cloud Speech
• TTS: Kokoro 82M (local, 16kHz slin16 output)
• Infrastructure: Asterisk PBX (FreePBX), Docker Compose, arilink CLI

Roadmap

• Web dashboard with live calls, campaigns, call historyDone
• Local STT (Parakeet) + TTS (Kokoro) — zero API costDone
• Pluggable brain architecture (BrainHarness + IBrain)Done
• LLM conversational voice agent (LlmChatBrain)Done
• Barge-in with AEC3 echo cancellation + Silero VADDone
• Docker Compose + CLI toolDone
• OpenClaw AI integrationDone
• MCP tool use in LlmChatBrain (calendar, CRM, SMS)
• Call analytics and latency metrics
• WebRTC softphone improvements
• Additional STT providers: Deepgram, Azure Speech, ElevenLabs Scribe

See docs/ROADMAP.md for the full product roadmap.

License

This project is licensed under the AriLink Community License (ACL) v1.0 - see the LICENSE file for details.

• Free for personal & internal use - View, study, and modify for personal/internal use.
• Commercial use restricted - Internal commercial operations allowed; no redistribution or SaaS.
• No Redistribution - You may not redistribute, white-label, or sell the software.
• Commercial Licensing: For SaaS hosting, resale, or third-party deployment, contact Alexi Pawelec (Discord: alexispace).

---

Built with care for modern telephony