MemoryOS provides a hierarchical memory operating system for personalized AI agents, aiming to enhance coherence and context awareness in interactions. It targets developers building AI agents who need robust long-term memory capabilities, offering significant performance improvements on memory benchmarks.

How It Works

MemoryOS employs a four-module architecture: Storage, Updating, Retrieval, and Generation. This design mirrors OS memory management principles, enabling efficient handling of short-term, mid-term, and long-term memory. It supports pluggable memory modules and various LLMs, facilitating flexible integration.

Quick Start & Requirements

• Installation: pip install memoryos-pro (from PyPI) or clone from GitHub and install requirements.txt. Docker images are also available.
• Prerequisites: Python >= 3.10. OpenAI API key is required for default LLM integration, though other models like Deepseek and Qwen are supported.
• Setup: Basic setup involves initializing the Memoryos class with user and API details. Docker setup is also straightforward.
• Documentation: https://bai-lab.github.io/MemoryOS/docs

Highlighted Details

• Achieves SOTA results on the LoCoMo benchmark, boosting F1 by 49.11% and BLEU-1 by 46.18%.
• Offers a plug-and-play architecture for memory modules (storage, update, retrieval).
• Provides a MemoryOS-MCP server for easy integration into agent clients.
• Supports a wide range of LLMs including OpenAI, Deepseek, and Qwen.

Maintenance & Community

The project is actively updated with recent additions including ChromaDB support, Docker integration, and performance optimizations. Community engagement is encouraged via GitHub issues/pull requests and a Discord server. Contact for cooperation: baiting@bupt.edu.cn.

Licensing & Compatibility

The repository does not explicitly state a license. The project name "MemoryOS" is noted as being used for academic discussion purposes, with a disclaimer regarding affiliation with memoryos.com. Commercial use implications are unclear without a stated license.

Limitations & Caveats

A detailed documentation is still under development. The project name disclaimer suggests potential naming conflicts or legal considerations. Access to the "MemoryOS Playground" requires an invitation code.