An intelligent CLI assistant featuring a sophisticated dual-LLM system: one model analyzes intent and context, while another generates personality-driven responses. Includes persistent memory management, structured note-taking, and adaptive conversation flow.
- Intent Analyzer: Processes user input to determine mode, technical level, correction needs, and response parameters
- Personality Engine: Generates contextually appropriate responses based on intent analysis and conversation history
- Multi-Scope: User preferences, system config, temporary context
- Multi-Level: Short, medium, and long-term retention
- Importance Weighting: Prioritizes critical information
- Tag-Based Organization: Semantic categorization
- Automatically extracts and saves structured information from conversations
- Category, priority, and status management
- Tag-based retrieval system
- 🧠 Context-Aware Responses: Maintains conversation history and adapts tone
- 💾 Persistent Memory: Remembers user preferences and important information across sessions
- 📝 Automatic Note Creation: Extracts and structures information from natural conversation
- 🎭 Adaptive Personality: Adjusts technical depth, tone, and emoji usage based on context
- 📊 Complete Logging: Full LLM request/response audit trail
- 🔧 Modular Architecture: Clean separation of concerns for maintainability
- Python 3.14
- Google Gemini API (dual models)
- SQLite (memory, notes, history, logs)
- prompt_toolkit (rich CLI interface)
- Python 3.8+
- Google Gemini API key (get one here)
- Clone the repository
git clone https://github.com/lionelponce/sayuri-v2.git
cd sayuri-v2- Install dependencies
pip install -r requirements.txt- Configure API key
cp config.json.example config.json
# Edit config.json and add your Gemini API key- Run Sayuri
python main.pyTu: ¿Cómo funciona el sistema de memoria?
Sayuri: El sistema tiene tres niveles de retención basados en importancia...
Tu: Guarda una nota: reunión el viernes con el equipo de DevOps
Sayuri: [Nota guardada] 📝
Tu: factory_reset
[Resets all databases and starts fresh]
sayuri_v2/
├── main.py # CLI interface
├── config.json # Configuration (API keys, models)
├── requirements.txt # Dependencies
├── sayuri/
│ ├── core.py # Main orchestration logic
│ ├── database.py # Database context manager
│ ├── googleapis.py # Gemini API integration
│ ├── memory.py # Memory management
│ ├── notes.py # Note-taking system
│ ├── history.py # Conversation history
│ └── llmlogs.py # LLM request/response logging
├── prompts/
│ ├── personality.txt # Personality engine prompt
│ └── intention.txt # Intent analyzer prompt
├── databases/ # SQLite databases (auto-created)
└── logs/ # LLM interaction logs
When you send a message, the Intent Analyzer determines:
- Interaction mode (casual, technical, correction, support)
- Required technical depth (low, medium, high)
- Response length and tone
- Whether to store memory or create notes
The system automatically stores important information:
{
"scope": "user",
"level": "long",
"content": "Prefers Python over JavaScript",
"importance": 3,
"tags": ["preference", "programming"]
}The Personality Engine generates responses tailored to:
- Detected intent
- Relevant memories
- Conversation history
- User preferences
- Personal Assistant: Remembers preferences, manages notes
- Technical Companion: Adapts explanations to your skill level
- Learning Tool: Stores knowledge and retrieves context
- Conversation Partner: Natural, context-aware dialogue
- Local-First: All data stored locally in SQLite
- No Cloud Storage: Conversations stay on your machine
- API Privacy: Only sends necessary context to Gemini API
- Web interface
- Voice interaction (TTS/STT)
- Multi-user support
- Plugin system for custom skills
- Export/import memory and notes
GNU GPL v3.0 License
Lionel Ponce
- GitHub: @lionelponce
Built to explore advanced AI architectures and human-AI interaction patterns.