VoiceAI Installer

Fully local, modular voice AI infrastructure — built for real-time, self-hosted, production-ready systems.

---

🚀 Overview

VoiceAI Installer is a staged, end-to-end setup system for building a fully local voice AI stack.

It orchestrates multiple AI components into a single cohesive system:

• 🧠 Local LLM server
• 🎤 Speech-to-Text (STT)
• 🔊 Text-to-Speech (TTS)
• 🧩 Agent orchestration layer
• 🧠 Memory system (RAG with Qdrant)
• 🌐 Tooling (web search, extensions)
• 📞 Telephony integration
• 🎙 Real-time communication (LiveKit)

All components are designed to run locally, with modular architecture and clean separation of concerns.

---

🧱 Architecture

The system is built as a multi-service local AI stack:

User Voice
   ↓
STT → Agent → LLM → Tools / Memory
   ↓            ↓
 TTS ← Response ← RAG (Qdrant)
   ↓
Audio Output (LiveKit / Phone)

Each service runs independently and communicates over internal APIs.

---

⚙️ Features

• Fully local AI stack (no external dependency required)
• Modular service-based architecture
• Real-time voice pipeline
• Event-driven agent system
• Persistent memory (vector DB + embeddings)
• Hot-reload capable services
• Telemetry & monitoring layer
• Admin surfaces (operator-only)
• Production-oriented setup (not a toy project)

---

🧩 System Components

Core Services

• LLM Server (TabbyAPI)
• STT Service (Faster-Whisper)
• TTS Stack (Qwen3-TTS + Chatterbox)
• Agent Runtime
• Memory Layer (Qdrant)

Infrastructure

• LiveKit (real-time communication)
• Telemetry Service
• Admin Layer
• Systemd integration

AI Capabilities

• RAG (Retrieval-Augmented Generation)
• Persona system
• Voice activity detection (Silero VAD)
• Turn detection
• Tool integrations (web search, etc.)

---

🧠 Philosophy

This project is built with one goal:

Make local, production-grade voice AI systems reproducible.

Instead of fragmented setups and ad-hoc scripts, this provides:

• Structured installation
• Predictable environment
• Clean service boundaries
• Scalable architecture

---

🛠 Installation

git clone https://github.com/okaybeydanol/voiceai-installer.git
cd voiceai-installer

chmod +x bootstrap.sh
./bootstrap.sh

cd dashboard
chmod +x build-dashboard.sh
cd /home/{PROFILE_NAME}/ai-projects
# Enter the path where you downloaded it.
/home/{PROFILE_NAME}/Desktop/voiceai-installer/dashboard/build-dashboard.sh

cd /home/{PROFILE_NAME}/ai-projects/voiceai/bin/
chmod +x voiceai-ctl.sh

./voiceai-ctl.sh start
./voiceai-ctl.sh status
./voiceai-ctl.sh logs agent
./voiceai-ctl.sh health
./voiceai-ctl.sh validate

---

📂 Project Structure

bootstrap/
  ├── 00_shared_helpers.sh
  ├── 01_prepare_layout_and_environment.sh
  ├── 02_run_preflight_and_shared_tools.sh
  ├── ...
  └── 11_download_models_and_finalize.sh

dashboard/
  └── build-dashboard.sh

bootstrap.sh

Each stage is designed to be:

• Idempotent
• Isolated
• Maintainable

---

📊 Current Status

🚧 Active development

• Core infrastructure: ✅
• LLM / STT / TTS: ✅
• Agent system: ✅
• Memory (Qdrant): ✅
• Telephony layer: ✅
• Dashboard (Next.js): It's almost finished. Only prettier, types, eslint, and a couple of performance points remain. ⏳

---

🔮 Roadmap

• Complete Next.js dashboard
• Improve agent reasoning loop
• Add more tool integrations
• Enhance telephony support
• Optimize resource management

---

🤝 Contributing

This project is currently evolving rapidly.

Contributions, feedback, and ideas are welcome.

---

📄 License

MIT License

---

⚡ Author

Built by someone obsessed with making local AI systems actually usable.