SYNZ: The Digital Human Agent

SYNZ is a fully autonomous Digital Human Co-Pilot capable of seeing, hearing, speaking, remembering, and teaching code. She lives on your desktop, interacts via a Live2D anime avatar, and uses local LLMs (Llama-3) to think.

---

Key Features

• Local Intelligence: Powered by Llama-3 (8B) running locally. No API fees, no data leaks.
• Code Mentor Mode: Watches your file system in real-time. If you save a file with bugs, she will speak up and teach you the fix.
• Smart Voice: Real-time STT (Whisper) with "Voice Hardening" to ignore keyboard clicks and silence hallucinations.
• Hybrid Personality: Witty, Sassy, and slightly chaotic (like a streamer), but fundamentally Helpful and smart.
• Vision: Can "See" your screen (monitor 1) and analyze UI/Code.

---

Architecture (How It Works)

SYNZ is a Distributed System spanning multiple processes:

1. The Ears (Hearing):

   • Tech: openai-whisper (Small Model).
   • Function: Listens for the wake word ("SYNZ"). Filters out background noise and self-echoes.

2. The Face (Personality & Routing):

   • Tech: Python UDP Server.
   • Function: The central hub. Managing the "Agentic State", Memory (Vector + JSON), and the Startup Chime (Beep-Beep).

3. The Brain (Logic Core):

   • Tech: Llama-3-8B (8192 Context Window).
   • Function: Context-aware thinking. It runs the Code Mentor logic and handles complex reasoning.

4. The Body (Top Layer):

   • Tech: Unity Engine + Live2D Cubism.
   • Function: Visual avatar with Lip Sync, Blinking, and Breathing (controlled by Face Server).

---

Downloads (Portable)

Latest Build (v2 Portable): Download SYNZ v2 (Google Drive)

• Size: ~7.2 GB
• Requirements: Windows 10/11, NVIDIA GPU (Drivers Installed).
• Portable: Works on any machine (CUDA 13.x drivers bundled).

---

Installation Guide

Prerequisites

1. Python 3.10+
2. Unity 2022.3 (LTS) (For the Avatar)
3. NVIDIA GPU (RTX 3060+ Recommended for fast responses)

Setup

1. Install:
   • Run setup_synz.bat.
   • Wait for "SETUP COMPLETE". (Creates venv and installs PyTorch/Whisper).
2. Enable GPU (Optional but Recommended):
   • Run enable_gpu.bat to install CUDA-optimized libraries.

---

How to Run

1. Launch the Mind

Double-click start_synz.bat.

• This opens 3 terminals (Brain, Face, Ears).
• Wait for the Double Beep tone.
• Wait for "Systems Online" voice message.

2. Launch the Body

• Open the Unity Project.
• Press Play.
• You will see [NeuroLink] Connected! in the logs.

---

Controls & Commands

Voice Commands

• Wake Up: "SYNZ" (or "Sins"). -> She listens for 30s.
• Code Mentor:
  • "Activate Code Mentor" -> Triggers analysis on file Save.
  • "Stop Code Mentor" -> Disables the watcher.
• Vision: "Look at this" -> Takes a screenshot and describes it.
• Search: "Search for [Topic]" -> Googles it in real-time.

Text Testing

If you don't want to talk, you can test text responses:

• Run the script: .\venv\Scripts\python.exe TheBrain\test_voice_chat.py
• Type: "Hello". She will speak the response.

---

Troubleshooting

Q: She echoes herself / Hears herself speak? A: This is fixed in the latest update (Socket Port 8009). Ensure start_synz.bat was restarted to load the new ears.py.

Q: "hallucinations" (She hears "Thanks for watching")? A: A filter is now active to block these common Whisper artifacts. If it persists, try lowering your mic volume.

Q: Code Mentor is annoying? A: Say "Stop Code Mentor". She will only speak when spoken to.

---

Powered by Local AI.