This project enables users to create personalized AI desktop companions with customizable characters, voice, and personality, offering near real-time (1-second) responses. It aims to replicate and expand upon the functionality of Neuro-sama, providing a platform for users to build AI agents with long-term memory, visual recognition, and voice cloning capabilities, targeting enthusiasts and developers interested in advanced AI companionship.

How It Works

The system orchestrates multiple microservices for core AI functionalities. It leverages a modular architecture that supports both open-source and closed-source Large Language Models (LLMs) and Text-to-Speech (TTS) engines, with a default integration of GPT-SoVITS for voice synthesis. Visual recognition is integrated for image analysis and context-aware activation of features. The project emphasizes low-latency responses through efficient service orchestration and optimized model inference.

Quick Start & Requirements

• Installation: A one-click deployment script (bash python neural_deploy.py) is available, followed by running Game-starts.bat. Alternatively, manual setup involves creating a Conda environment (conda create -n my-neuro python=3.11), installing dependencies (pip install -r requirements.txt, conda install ffmpeg, pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118), downloading models (python Batch_Download.py), and starting individual services (bert_api.py, asr_api.py, Mnemosyne-bert\api_go.py, tts-studio\tts_api.py).
• Prerequisites: Anaconda, Python 3.11, CUDA 11.8 (default, configurable), and an API key for LLM services (e.g., Deepseek, Qianwen, Zhipu AI) are required. For local LLM inference/fine-tuning, a minimum of 12GB VRAM is recommended.
• Resources: Initial setup requires downloading models. The project mentions less than 6GB VRAM for basic deployment.
• Links: Releases v3.9.2, GPT-SoVITS.

Highlighted Details

• Near 1-second response latency.
• Supports Live2D model customization.
• Features long-term memory and active dialogue initiation.
• Includes Bilibili live streaming integration and typing chat.

Maintenance & Community

• The project is actively developed with a roadmap including features like real-time interruption, emotional states, and game companionship.
• A QQ group is available for community support: 756741478.

Licensing & Compatibility

• The project utilizes the GPT-SoVITS project, which is under the Apache 2.0 license. The specific license for my-neuro itself is not explicitly stated in the README, but its components and dependencies should be reviewed for compatibility.

Limitations & Caveats

The one-click deployment is experimental and may fail. Some advanced features like real emotions, real-time interruption, and game companionship are still under development or planned for future releases. The project is described as approximately 30% complete, with significant development planned.