AIMING Lab 开源 AutoClaw & MetaClaw：自我迭代的 AI 代理框架

兄弟们！来自 AIMING Lab的自我迭代的OpenClaw 有点东西啊！ 2两半狂揽3.3 k Star 🌟并且还在持续增长！

这项目基于OpenClaw 迭代出一个进化成“会自己变聪明”的版本 ——AutoClaw& MetaClaw 正式开源！🦞🧬

Just talk to your agent，它就自动学习、提取 Skills、持续进化！

核心功能：

• 透明代理层在OpenClaw / CoPaw / IronClaw / NanoClaw 等前面

• 每一次正常聊天都变成学习信号 → 自动注入最相关 Skills + 会话后自动总结新 Skill

• 三种模式：skills_only（轻量） / rl（立即训练） / MadMax（默认！睡觉、开会、闲置时偷偷云端 RL 训练，永不打扰你）

• 完全不用本地 GPU！用 Tinker / MinT 云端 LoRA 就行

• v0.3.2 新增多爪支持 + OpenRouter

bash metaclaw setup metaclaw start

GitHub github.com/aiming-lab/Met…YbU

!Image 2: Tweet image

!Image 3: Huaxiu Yao

#### Huaxiu Yao

@HuaxiuYaoML · 1d ago

🦞 AutoResearchClaw hit 3.3K⭐ in 2.5 days. What if it could get better after every single run?

Enter MetaClaw — the self-evolution engine.

Zero retraining. Zero code changes. Just metaclaw start:

📉 40% fewer refine cycles

📉 24.8% fewer stage retries

📈 18.3% higher robustness

No GPU needed. MetaClaw sits as a proxy, injects skills learned from past failures, and meta-learns from every conversation.

Your agent doesn't just run. It evolves. �github.com/aiming-lab/Met…ZJwwXr

Built with @openclaw and @thinkymachines

Kudos to the team @richardxp888, @JimChenjw, @Xinyu2ML, @lillianwei423, @StephenQS0710, @HaoqinT, @JiaqiLiu835914, @yuyinzhou_cs, @zhengop, @cihangxie

!Image 4: Tweet image

13

31

158

9,963

1 Replies

0 Retweets

17 Likes

2,280 Views 

One Sentence Summary

AIMING Lab has officially open-sourced AutoClaw and MetaClaw, a self-iterating AI agent framework based on OpenClaw, capable of automatically learning, extracting skills, and continuously evolving through dialogue, all without requiring a local GPU.

Summary

This tweet introduces AIMING Lab's open-source AutoClaw and MetaClaw projects, a self-iterating AI agent framework built upon OpenClaw. Its core highlight is the agent's ability to automatically learn, extract skills, and continuously evolve through every conversation, achieving a "self-improving" functionality. The tweet details its core features, including a transparent proxy layer, transforming chats into learning signals, three operational modes (skills_only, rl, MadMax), and the advantage of cloud LoRA training without needing a local GPU. Furthermore, it mentions that v0.3.2 adds multi-claw support and OpenRouter integration, providing a GitHub link and simple startup commands, making it highly valuable for AI agent developers.

AI Score

88

Influence Score 7

Published At Yesterday

Language

Chinese

Tags

AutoClaw

MetaClaw

OpenClaw

AI Agent

Self-Iteration