Sentri

A desktop AI assistant with persistent memory and a floating Clippy-style interface.

Features

• Floating Assistant UI - Draggable chat window with modern design
• Persistent Memory - Remembers user identity, preferences, and conversation context
• Semantic Memory Retrieval - Embedding-based search understands meaning, not just keywords
• Privacy Controls - User consent system for memory storage (full/limited/session-only)
• Memory Viewer - See, filter, and delete stored memories
• Intent Routing - Extensible tool system for time, system info, and more
• LLM Integration - Google Gemini API support
• Control Center - Qt-based GUI for configuration and management

Architecture

sentri/
├── brain/              # Memory and logic core
│   ├── controller.py   # Main orchestrator
│   ├── memory_store.py # Persistent storage
│   └── memory_retriever.py # Smart retrieval
├── clippy/             # Floating window UI
├── sentri_ui/          # Control center GUI
├── llm_adapters/       # LLM provider integrations
├── tools/              # Extensible tool system
└── sentri_core/        # Intent routing and registry

Setup

Prerequisites

• Python 3.10+
• Linux (tested on Ubuntu/Debian)

Installation

1. Clone the repository:

git clone <repo-url>
cd Sentri

2. Create virtual environment:

python3 -m venv venv
source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

Or manually:

pip install PySide6 google-genai python-dotenv sentence-transformers chromadb cryptography

4. Configure API key:

echo "GEMINI_API_KEY=your_api_key_here" > .env

Usage

Launch Control Center

python main.py

From the control center:

1. Go to LLM Settings tab
2. Enter your Gemini API key
3. Click Save Configuration
4. Go to Clippy tab
5. Click Launch Assistant

Memory System

Sentri automatically stores:

• Identity - Name, age, personal info
• Preferences - User likes/dislikes
• Relationships - People and connections
• Skills - What you know/can do
• Goals - Long-term objectives
• Habits - Patterns and routines
• Facts - General knowledge about the user
• Tasks - Reminders and to-dos

Semantic Search: Uses sentence-transformers (runs locally) to understand meaning:

• "What do I like?" finds "I enjoy hiking" and "My favorite is pizza"
• No exact keyword matching required
• Retrieval scored by: semantic similarity (60%), confidence (20%), recency (20%)

Privacy Controls: First-run consent dialog with three levels:

• Full - Remember everything
• Limited - Only preferences, tasks, goals
• Session Only - No persistence

Memory Viewer: Access from Control Center to view, filter, and delete stored memories.

Adding Tools

Create a new tool in tools/:

from tools.base_tool import BaseTool

class MyTool(BaseTool):
    name = "mytool"
    
    def execute(self, user_input: str) -> dict:
        return {"result": "tool output"}

Register in sentri_core/intent_router.py:

if "keyword" in text:
    return "mytool"

Configuration

Config stored in data/config.json:

{
    "provider": "gemini",
    "api_key": "your_key"
}

Development Status

Working:

• ✅ Memory storage and retrieval
• ✅ Semantic search with embeddings
• ✅ Privacy consent system
• ✅ Memory viewer UI
• ✅ Gemini LLM integration
• ✅ Floating window UI
• ✅ Intent routing
• ✅ Tool system foundation

In Progress:

• 🚧 Chat history UI
• 🚧 Settings panel
• 🚧 Animations
• 🚧 Local LLM support

Known Issues

• API key stored in plain text (use .env for now)
• No error handling for network failures
• Some UI components incomplete

Upgrading

If you're upgrading from an older version, see UPGRADE_GUIDE.md for migration instructions.

Roadmap

• Semantic search with embeddings
• Memory management UI
• Privacy consent system
• Secure credential storage
• Multi-LLM support (OpenAI, local models)
• Export/import conversations
• Plugin system
• Voice input/output
• Memory decay and consolidation

License

[Add your license here]

Contributing

[Add contribution guidelines]