开源AI工具：本地AI助手

Advanced STS Local AI Assistant

A fully local, privacy first AI assistant that runs entirely on your machine, no cloud, no subscriptions, no data leaving your PC. It combines speech recognition, a large language model, text to speech, a RAG database and a Powerfull MCP Server with a modular plugin architecture into a single easy to use desktop application.

Demo Video :

://www.youtube.com/watch?v=V0Aa8dKgbi0&t=141s

Key Features :

• 🎙️ STT - Faster Whisper with Silero VAD for accurate, low-latency voice detection/transcription
• 🤖 LLM Integration - Connects to LM Studio to run any local language model you wish
• 🔊 TTS - Coqui XTTS-V2 for natural, cloneable voice synthesis

   Par With Eleven Labs on Voice Quality

• 💬 Chat History - Persistent conversation log with full context management

   History is saved as a .txt file for backup and GUI continuity between sessions. It is also used to manually rebuild RAG Database if it gets corrupted/deleted

• 🧠 RAG Database - MiniLM-L6-v2 + ChromaDB for Semantic very-long-term memory and document retrieval:

  .pdf .txt

   is NO limit on how much it can remember😊

• 🔌 MCP Server - Modular plugin system for extending the assistant's capabilities

  : Google Services, Windows CLI, Home Assistant, Telegram, Signal and more to come

• 🐍 Architecture - Full Python

   two C++ backends: faster-whisper and ChromaDB

• 🖥️ GUI - PyQt5 for a clean and modern desktop interface

   Pannel Style

Minimum System Requirements :

Installation :

Step 1 - Install Python 3.12.6 X64

Download and install Python 3.12.6 X64 from the official website:

://www.python.org/downloads/release/python-3126/

 Python to PATH

Step 2 - Install Visual Studio Build Tools

Some packages require C++ compilation. Download and install the Build Tools from:

://visualstudio.microsoft.com/visual-cpp-build-tools/

 development with C++

Step 3 - Install NVIDIA CUDA 12.8 and cuDNN 9.12

CUDA 12.8:

://developer.nvidia.com/cuda-12-8-0-download-archive

://developer.nvidia.com/cudnn

Step 4 - Install LM Studio

Download and install LM Studio:

://lmstudio.ai

.gguf

 + shift + M

/Local Server

 settings

 CORS

 Model loading

 Keep Last JIT Model Loaded

Step 5 - Clone the Repository

git clone

://github.com/DIY-Engineering/Advanced-STS-Local-AI-Assistant.git

Step 6 - Run the

.py

The setup script handles everything automatically: - ✅ Verifies you are using Python 3.12.6 X64 - ✅ Creates the full project folder structure - ✅ Installs PyTorch with CUDA 12.8 - ✅ Installs all required Python packages - ✅ Downloads and installs all AI models

You can run the script with following options:

Setup.py --cpu                                   # Install PyTorch CPU version (no GPU)
Setup.py --skip-deps                             # Skip dependency installation (models only)
Setup.py --skip-models                           # Skip model downloads (deps only)
Setup.py --only-coqui                            # Download only the Coqui TTS model
Setup.py --only-whisper                          # Download only Faster-Whisper models
Setup.py --whisper-models small medium large-v3  # Specific Whisper models only

Step 7 - Manual Model Installation Alternative

If the automatic setup fails for the model downloads, you can install the models manually. See

 Models Download.txt

Step 8 - Manual python dependencyes install Alternative

If

.py

 install -r requirements.txt

Folder Structure

Advanced STS Local AI Assistant\
│
├── Advanced STS Local AI Assistant.py   ← Main application
├── Setup.py                             ← Automated setup script
│
├── Chat History\                        ← Conversation logs
├── Coqui TTS\
│   ├── Models\                          ← XTTS-v2 model files
│   └── Samples\                         ← Voice cloning reference audio
├── Debug Logs\                          ← Application logs
├── Dependencies\                        ← Additional local dependencies
├── Graphics\                            ← UI assets
├── MCP                                  ← Tool Chain Rules File
├── MCP Server\
│   ├── Graphics\
│   └── Plugins\                         ← MCP plugin scripts
├── Profiles\                            ← User settings and profiles
├── RAG Embedder\
│   └── MiniLM-L6-v2\                    ← Sentence embedding model
├── RAG Vector Database\                 ← ChromaDB knowledge base
├── Silero VAD\
│   └── Models\                          ← Voice activity detection models
├── System Prompt\                       ← System prompt configuration files
└── Whisper STT\
    └── Models\
        ├── tiny\
        ├── base\
        ├── small\
        ├── medium\
        └── large-v3\

Dependencies

All dependencies are installed automatically by

.py

Full pinned dependency list:

/Requirements.txt

🚀 First Launch 🚀

1. Make sure LM Studio is running, the local server is started and that you have at least one model downloaded, for starters i recommend

.gguf

 "Advanced STS Local AI Assistant.py"

 patient here, it has to load Heavy Dependencies😊

 default it uses Microsoft Sound Mapper for input/output

 Voice Sample

Troubleshooting :

.py

 Logs\Debug Log.txt

 out of memory

 choose model load parameters

 not working / Coqui refuses to load

.txt

 TTS\Models\

 Models Download.txt

 not detected

 Studio connection error

://localhost:1234

Acknowledgements: - Faster-Whisper -

://github.com/SYSTRAN/faster-whisper

://github.com/coqui-ai/TTS

://github.com/snakers4/silero-vad

://github.com/chroma-core/chroma

://lmstudio.ai

://github.com/UKPLab/sentence-transformers