@HuaxiuYaoML
Meet MetaClaw π¦β Just talk to your agent, it learns and evolves. π¬ Conversations become training trajectories β‘ Models update live with hot-swapped weights π§ Failures generate new reusable skills π» No GPU cluster required Under the hood: π Online SkillRL training Training runs asynchronously while the model continues serving. π Skill evolution When the agent fails, an LLM analyzes the trajectory and generates new reusable skills. π Skill injection Relevant skills are retrieved and injected into the system prompt at each step to guide behavior in real time. Built on Kimi-2.5 via Tinker cloud LoRA β fine-tuning costs about $10 and requires no GPU cluster π» Fully open-sourced https://t.co/RR24ZJvZ7T Built with @openclaw and @thinkymachines Kudos to the team @richardxp888, Jianwen Chen, @Xinyu2ML, @lillianwei423, @StephenQS0710, Zeyu Zheng, @cihangxie!