EverMemOS – An Open-Source AI Long-Term Memory Operating System from Chen Tianqiao’s Team

What is EverMemOS?

EverMemOS is an open-source long-term memory operating system developed by Chen Tianqiao and the Shanda team, designed to provide AI assistants with persistent and retrievable memory capabilities. Inspired by human memory mechanisms, it adopts a four-layer architecture consisting of the Agent Layer, Memory Layer, Index Layer, and Interface Layer. By integrating data from tools such as Slack, Gmail, and Notion, EverMemOS transforms fragmented information into unified, searchable memory.Its core innovation is that it’s not just a memory “database”, but an “application processor” capable of influencing an AI model’s thinking and responses in real time. EverMemOS achieved a high score of 92.3% on the LoCoMo long-term memory benchmark, significantly outperforming existing systems. It supports both dynamic and static memory management and can adapt to diverse needs for teams and individuals.

Main Features of EverMemOS

Memory Extraction and Structuring

Extracts key information from conversations and converts it into structured memory units (MemCells) and user profiles, helping AI maintain context and recall past interactions.

Multi-scenario Memory Management

Supports dynamic memory and static memory, as well as personal and team-shared memory. It provides optimal memory organization and application strategies for different scenarios.

Proactive Intelligence and Reasoning

Capable of proactively identifying issues, driving decision-making, and following through on tasks. Through multi-hop reasoning, it traverses complex information chains to uncover deeper knowledge relationships.

Technical Architecture Design

Uses a four-layer architecture (Agent Layer, Memory Layer, Index Layer, Interface Layer) and integrates a variety of techniques, including RAG, vector retrieval, keyword search, and knowledge graphs. It supports integration with leading LLMs and intelligently selects the most suitable model.

Memory Portability

Users can export all data and knowledge at any time—including structured memories and insights—to avoid data lock-in.

Support for Multiple Applications

Through integration with Slack, Gmail, Notion, and other productivity tools, it supports both team collaboration and personal knowledge management to enhance efficiency.

Technical Principles of EverMemOS

Inspired by Human Memory Mechanisms

EverMemOS adopts a layered memory architecture modeled after how the human brain stores, retrieves, and applies memory, enabling efficient processing and management of memory data.

Four-layer Architecture

The system consists of the Agent Layer, Memory Layer, Index Layer, and Interface Layer.

• The Agent Layer handles task understanding and distribution.

• The Memory Layer manages long-term memory.

• The Index Layer optimizes memory retrieval.

• The Interface Layer enables external integrations.

Multi-model Integration

Combines RAG, vector search, keyword retrieval, and knowledge graph methods, and integrates major LLMs such as Gemini, Claude, and ChatGPT. The system intelligently selects the best model for each scenario.

Layered Memory Extraction

Extracts continuous semantic segments from conversations and converts them into episodic memory units, which are dynamically organized into structured memory. This approach overcomes the limitations of pure text-similarity searching and improves accuracy and efficiency.

Dynamic and Static Memory Management

Supports coordination between dynamic memory (e.g., real-time conversations) and static memory (e.g., documents and knowledge bases). The system provides optimal strategies depending on the scenario.

Memory Portability and Security

Users can export all data and structured insights at any time, ensuring portability and preventing data lock-in.

Proactive Reasoning Capabilities

Equipped with the ability to proactively identify problems, drive decisions, and track execution. Its multi-hop reasoning capability uncovers deep relationships within information chains.

Project Links

• Official Website: https://evermind-ai.com/

• GitHub Repository: https://github.com/EverMind-AI/EverMemOS/

Application Scenarios of EverMemOS

Team Collaboration

Integrates tools used by teams (Slack, Gmail, Notion, etc.) to unify knowledge management, improve collaboration efficiency, and ensure information sharing and knowledge retention.

Personal Knowledge Management

Helps individuals convert fragmented ideas and information into a structured knowledge system. Supports transforming initial thoughts into detailed plans to enhance personal productivity.

Enterprise Applications

Provides long-term memory management solutions for businesses, enabling persistent storage and intelligent retrieval of enterprise data. Supports knowledge management and decision-making.

Education Sector

Helps educators and students organize learning materials, document learning processes, and manage knowledge points to improve learning efficiency.

Intelligent Customer Support

Improves customer service quality and efficiency by enabling rapid retrieval of historical conversations and providing more accurate responses.

Content Creation

Supports creators in managing ideas and creative resources, assists in generating content, and improves the efficiency and quality of creative work.