Vico is an advanced memory management system that combines cutting-edge visual processing with sophisticated text analysis. Built with modern web technologies and optimized for Apple Silicon, it offers an intuitive interface for storing, managing, and searching through digital memories.
- Intelligent Image Analysis: Automated detailed image descriptions using MLX-powered vision models
- Visual Memory Storage: Efficient image processing and storage with base64 encoding
- Multi-Modal Search: Combined text and visual content search capabilities
- Real-time Operations: Instant memory creation, editing, and deletion
- Advanced Search: Natural language processing for intuitive memory retrieval
- Context-Aware Responses: AI-powered memory analysis and correlation
- Next.js 14 App Router with TypeScript
- React Server Components for optimal performance
- Tailwind CSS + shadcn/ui for component styling
- Client-side state management with React hooks
- Python FastAPI service for ML operations
- MLX for efficient model inference
- DuckDB for fast, reliable data storage
- CORS support for Chrome extension integration
- Qwen2.5-VL-72B for image analysis
- DeepSeek-R1-Distill-Qwen-14B for short-context memory reasoning
- Qwen2.5-14B-Instruct-1M for long-context memory processing
- Optimized inference using MLX on Apple Silicon
- Apple Silicon Mac (M1/M2/M3)
- Python 3.10+
- Node.js 18+
- Bun runtime
- Install Dependencies
bun install
pip install -r requirements.txt
- Run the Development Server
bun run dev
- Run the Inference Service
python python/inference_service.py
Built using shadcn/ui and ML implementations from mlx-vlm ❤️.
# Backup memories.duckdb twice daily (9am/9pm)
# Creates max 62 rotating backups (31 days × 2 backups/day)
# Format: memories-YEAR_DAY_HOUR.duckdb (e.g., memories-2024_15_09.duckdb)
0 9,21 * * * rsync -avz /path/to/vico/data/memories.duckdb backupserver:/path/to/backup/memories-$(date +\%Y_\%d_\%H).duckdb