MemU is an open-source memory framework designed for AI companions, offering high accuracy, fast retrieval, and cost efficiency. It functions as an intelligent memory manager that adapts to various AI companion scenarios, enabling companions to learn and grow with users through personalized interactions.

How It Works

MemU structures memories as an interconnected knowledge graph, organized into intelligent, self-managing folders. This approach avoids explicit memory modeling, allowing a memory agent to autonomously handle recording, modification, and archiving. Retrieval is optimized by categorizing information into documents, enabling focused searches rather than extensive embedding searches on fragmented data. An adaptive forgetting mechanism prioritizes information based on usage patterns, creating a dynamic, personalized information hierarchy.

Quick Start & Requirements

• Install via pip: pip install memu-py
• Requires Python 3.8+
• Cloud version requires an API key from app.memu.so/api-key/.
• Self-hosting backend is listed as "Coming next week."
• Example usage and integration details are available in the repository.

Highlighted Details

• Claims 92% accuracy on the Locomo dataset.
• Achieves up to 90% cost reduction through optimized online platform and batch processing of conversation turns.
• Features advanced retrieval strategies including semantic, hybrid, and contextual retrieval.
• Memories are organized as a file system with an interconnected knowledge graph and continuous self-improvement capabilities.

Maintenance & Community

• Active community engagement via Discord, X (Twitter), and Reddit.
• Enterprise support and inquiries are handled via email.
• Contributing guide and GitHub issues are available for community involvement.

Licensing & Compatibility

• Licensed under Apache License 2.0.
• Compatible with commercial use and closed-source linking.

Limitations & Caveats

The self-hosting backend is not yet available, limiting immediate local deployment options. While benchmarks are cited, a technical report is pending publication for detailed validation.