现在越来越觉得，Claude Code 或 Codex 最大的问题，已经不是写不出代码。

而是它记不住，我们对话中聊过的方案、踩过的坑，以及项目开发进度等等信息。

偶然发现 EverOS 开源了一个 Claude Code 插件，给 AI 装上了能自我进化的长期记忆。

GitHub：http://github.com/EverMind-AI/EverOS…

一行命令安装插件后，与 Claude Code 的每次对话都会自动保存，新对话还会自动检索记忆。

还提供一个 GitHub 风格的记忆面板，可随时查看每个项目以往跟 Claude 的对话内容。

此外，EverOS 仓库里有超过 20 个真实落地的应用案例，Claude Code 插件只是其中一个。

还有学习陪伴助手、阿尔茨海默症患者记忆辅助、AI 可穿戴设备等等，均用到 EverOS 这个记忆框架。

而且所有案例均开源可用，我们可以直接克隆代码下来学习，或者进行二次开发变成自己的东西。

有需要的同学可以看下，也可以 Star 关注一下 EverOS 后续更多的落地案例开源。

EverMind-AI/EverOS

Source: https://github.com/EverMind-AI/EverOS

Project Overview

EverOS is a unified home for applying, building, and evaluating long-term memory in self-evolving agents. The repository is organized around three essential parts:

At the center of EverOS is EverCore, a long-term memory operating system for agents. If you are new to the project, scan the use cases first to see what memory enables, then follow the Quick Start to run EverCore locally. The architecture and benchmark sections below give you the deeper reference material when you are ready to compare systems or reproduce results.

Use Cases

Use cases show what persistent memory makes possible in real products and workflows. Some examples are packaged in this repository; others point to external demos or integrations you can study and adapt.

Quick Start

Choose the path that matches your goal:

git clone https://github.com/EverMind-AI/EverOS.git
cd EverOS

Each component has its own installation guide, dependency configuration, and usage examples.

EverCore

The fastest way to run a memory system locally is to start with EverCore:

cd methods/EverCore

# Start Docker services
docker compose up -d

# Install dependencies
curl -LsSf https://astral.sh/uv/install.sh | sh
uv sync

# Configure API keys
cp env.template .env
# Edit .env and set:
#   - LLM_API_KEY (for memory extraction)
#   - VECTORIZE_API_KEY (for embedding/rerank)

# Start server
uv run python src/run.py

# Verify installation
curl http://localhost:1995/health
# Expected response: {"status": "healthy", ...}

Server runs at http://localhost:1995 · Full Setup Guide

Basic Usage

Store and retrieve memories with simple Python code:

import requests

API_BASE = "http://localhost:1995/api/v1"

# 1. Store a conversation memory
requests.post(f"{API_BASE}/memories", json={
    "message_id": "msg_001",
    "create_time": "2025-02-01T10:00:00+00:00",
    "sender": "user_001",
    "content": "I love playing soccer on weekends"
})

# 2. Search for relevant memories
response = requests.get(f"{API_BASE}/memories/search", json={
    "query": "What sports does the user like?",
    "user_id": "user_001",
    "memory_types": ["episodic_memory"],
    "retrieve_method": "hybrid"
})

result = response.json().get("result", {})
for memory_group in result.get("memories", []):
    print(f"Memory: {memory_group}")

More Examples · API Reference · Interactive Demos

Architecture Methods

These are the memory architectures currently included in EverOS. Use them as runnable systems, research references, or starting points for your own agent memory layer.

Benchmarks

These benchmarks provide shared standards for measuring memory quality and agent self-evolution across systems.

Evaluation

Use the evaluation runner to reproduce EverCore results or compare another memory system against the same benchmark tasks.

Benchmark Results

Supported Benchmarks

• LoCoMo — Long-context memory benchmark with single/multi-hop reasoning
• LongMemEval — Multi-session conversation evaluation
• PersonaMem — Persona-based memory evaluation

Run Evaluations

cd methods/EverCore

# Install evaluation dependencies
uv sync --group evaluation

# Run smoke test (quick verification)
uv run python -m evaluation.cli --dataset locomo --system everos --smoke

# Run full evaluation
uv run python -m evaluation.cli --dataset locomo --system everos

# View results
cat evaluation/results/locomo-everos/report.txt

Full Evaluation Guide · Complete Results

Citations

If EverOS helps your research, please cite the relevant paper:

@article{hu2026evermemos,
  title   = {EverMemOS: A Self-Organizing Memory Operating System for Structured Long-Horizon Reasoning},
  author  = {Chuanrui Hu and Xingze Gao and Zuyi Zhou and Dannong Xu and Yi Bai and Xintong Li and Hui Zhang and Tong Li and Chong Zhang and Lidong Bing and Yafeng Deng},
  journal = {arXiv preprint arXiv:2601.02163},
  year    = {2026}
}

@article{yue2026hypermem,
  title   = {HyperMem: Hypergraph Memory for Long-Term Conversations},
  author  = {Juwei Yue and Chuanrui Hu and Jiawei Sheng and Zuyi Zhou and Wenyuan Zhang and Tingwen Liu and Li Guo and Yafeng Deng},
  journal = {arXiv preprint arXiv:2604.08256},
  year    = {2026}
}

@article{hu2026evaluating,
  title   = {Evaluating Long-Horizon Memory for Multi-Party Collaborative Dialogues},
  author  = {Chuanrui Hu and Tong Li and Xingze Gao and Hongda Chen and Yi Bai and Dannong Xu and Tianwei Lin and Xiaohong Li and Yunyun Han and Jian Pei and Yafeng Deng},
  journal = {arXiv preprint arXiv:2602.01313},
  year    = {2026}
}

Stay Tuned

Star the repo or join the community links above to follow new architecture methods, benchmark releases, and memory-enabled use cases.

Contributing

Contributions are welcome across the whole repository: architecture methods, benchmark coverage, use-case examples, documentation, and bug fixes. Browse Issues to find a good entry point, then open a PR when you are ready.

Welcome all kinds of contributions 🎉

Help make EverOS better. Code, documentation, benchmark reports, use-case write-ups, and integration examples are all valuable. Share your projects on social media to inspire others.

Connect with one of the EverOS maintainers @elliotchen200 on 𝕏 or @cyfyifanchen on GitHub for project updates, discussions, and collaboration opportunities.

Code Contributors

Contribution Guidelines

Read the Contribution Guidelines for setup, pull request expectations, and use-case submission notes. For responsible disclosure, see the Security Policy.

License, Conduct, and Acknowledgments

Apache 2.0 • Code of Conduct • Acknowledgments