This project provides an MCP server for Claude Desktop, enabling persistent semantic memory and long-term context storage. It targets users of Claude Desktop who need to manage conversational history and recall information across sessions, offering features like natural language recall and tag-based retrieval.

How It Works

The service leverages ChromaDB for persistent vector storage and sentence transformers for generating semantic embeddings. It processes natural language commands to store, retrieve, and manage memories, supporting semantic search, time-based recall, and tag-based filtering. The architecture is designed for cross-platform compatibility and includes hardware-aware optimizations.

Quick Start & Requirements

• Installation: Clone the repository, create a virtual environment, and run python install.py. A Windows-specific installer (scripts/install_windows.py) is available. Docker installation is also supported via docker-compose up.
• Prerequisites: Python 3.10+ is recommended. Hardware accelerators like MPS (Apple Silicon), CUDA (NVIDIA), or DirectML (Windows) are supported for performance.
• Links: Installation Guide, Invocation Guide

Highlighted Details

• Semantic search with customizable embedding models.
• Natural language time-based recall (e.g., "last week").
• Persistent storage with ChromaDB and automatic backups.
• Cross-platform support including Apple Silicon (MPS), CUDA, and DirectML.
• Database optimization and duplicate cleanup tools.

Maintenance & Community

The project is maintained by doobidoo. Community support is available via Telegram.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive license suitable for commercial use and integration with closed-source applications.

Limitations & Caveats

Windows users may face PyTorch installation issues, necessitating the use of the provided Windows-specific installation script or wrapper. Performance on CPU-only configurations may be limited without hardware acceleration.