Diva - Local Voice Assistant with MCP

A privacy-focused voice assistant powered by local LLMs with MCP (Model Context Protocol) tool integration.

Features

Voice Recognition

• STT: Faster-Whisper for accurate speech-to-text
• VAD: Silero VAD for voice activity detection
• Wake Word: Listen mode with interrupt support

LLM Brain

• Local Models: Runs Ollama models (gemma3, llama3.2)
• Streaming: Token-by-token streaming responses
• Chain-of-Thought: Shows reasoning process before answering
• Conversation History: Maintains context across interactions

MCP Tool Integration

• LLM-Native Tool Calling: Models decide when to use tools
• Real-time Data: Weather, web search from live sources

Tools Available

Weather (get_weather)

• Get current weather for any city
• Temperature, conditions, humidity, wind speed
• Uses OpenWeatherMap API

Web Search (web_search)

• Search the web for real-time information
• Uses Tavily API (1000 free searches/month)

Thinking Process

The assistant shows step-by-step reasoning:

Let me think about this...
The user is asking about...
I should use [tool_name] to get accurate data...
[answer]

Setup

1. Install Dependencies

pip install -r requirements.txt

2. Install Ollama

# Download from https://ollama.ai 
ollama pull llama3.2 # or any models

3. Configure API Keys

Create .env file:

OPENWEATHERMAP_API_KEY=your_key
TAVILY_API_KEY=your_key

API Keys Required

Get free API keys:

• OpenWeatherMap: get-key-from-here
• Tavily: get-key-from-here

4. Run

python main.py

Usage

Voice Commands

• Say your question naturally
• Use "bye" or "exit" to stop

Examples

"What's the weather in London?"
"Search for latest tech news"
"Who is the president of USA?"
"Hello, how are you?"

Architecture

┌─────────────────────────────────────────────┐
│                  main.py                    │
│  (VAD → STT → Brain → LLM → Response)       │  
└─────────────────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────────┐
│              diva_brain.py                  │
│  • Detects tool needs                       │
│  • Shows thinking process                   │
│  • Streams tokens                           │
└─────────────────────────────────────────────┘
                    │
          ┌─────────┴─────────┐
          ▼                   ▼
┌─────────────────┐  ┌─────────────────┐
│    diva_mcp.py  │  │    Ollama LLM   │
│  • Tool schemas │  │  (llama3.2)     │
│  • Executes     │  │                 │
└─────────────────┘  └─────────────────┘
          │
          ▼
┌─────────────────────────────────────────────┐
│              MCP Servers                    │
│  • weather_mcp_server.py                    │
│  • web_search_mcp_server.py                 │
└─────────────────────────────────────────────┘

working on......

• Add TTS streaming integration (XTTS v2)
• Add email MCP server
• Add news MCP server
• Voice cloning
• Conversation memory
• Tool for calendar/tasks