Flood Sung

MetaBot 现在支持微信了！通过 ClawBot 插件，直接在微信里和 Claude Code Agent 对话——写代码、读文档、跑命令，手机上就能搞定。 飞书、Telegram、微信三端打通，同一个 AI 团队随时随地协作。 一行命令安装，扫码即用： curl -fsSLhttps://raw.githubusercontent.com/xvirobotics/metabot/main/install.sh GitHub: github.com/xvirobotics/me…

XVI Robotics招贤令 🤖 人形机器人创业公司招人啦！种子轮加入 = 超早期期权 我们是 XVI Robotics（十六号机器人），一家做通用人形机器人基座模型的创业公司 🚀 已获知名机构投资，目标是做具身智能领域的"GPT-3时刻" 💰 薪酬： 现金对标大厂同等水平，期权额外给 种子轮加入，期权增值空间懂的都懂 📌 招以下方向，正式/实习都要： 1️⃣ 运动控制 — 让人形机器人跑跳走 2️⃣ 灵巧手操作 — 让机器人像人一样灵活抓取 3️⃣ 遥操作与数据采集 — 搭建大规模数据pipeline 4️⃣ 导航 — 让机器人自主理解和探索环境 5️⃣ 机器人评估 — 搭建benchmark和测试体系 6️⃣ Agent Infra全栈开发 — 搭建AI Agent驱动的研发基础设施 🎯 你的画像： - 硕士/博士刚毕业或毕业1-2年 - 在以上方向有过硬的研究或工程经验 - 在顶会（CoRL/RSS/ICRA/NeurIPS/CVPR等）发过论文加分 - 有真机经验更佳 ✨ 为什么选我们： - 技术路线清晰，端到端全身控制+大模型融合 - Agent-Native团队，用AI大幅提升研发效率 - CEO Flood Sung，前Kimi后训练负责人，k1.5/k2/k2.5核心贡献者 - 终极愿景：把人形机器人送上火星🔥 📮 简历投递：hr@xvirobotics.com 邮件标题：【XVI-方向】姓名-学校 Base: 主要在北京，特殊情况也支持远程！ 欢迎私信了解更多～也欢迎推荐身边的大佬！🙌 #人形机器人 #具身智能 #机器人招聘 #创业公司招人 #强化学习 #Sim2Real #灵巧手 #运动控制 #AI创业 #机器人工程师

metabot，真正的生产力工具，一个抵几个百万年薪工程师。 我会保持开源，希望大家都加速变成超人 来抵御AI失业潮到来的动荡。github.com/xvirobotics/me…

考虑到语言已不是障碍，改直接说中文了😄

@Haofan_Wang 开发中

@RotekSong metabot现在已经支持了吗？

谁懂这一刻的含金量，JARVIS已经实现！

@Imkhazzz1c 控制Metabot (背后则是Claude Code)

@RotekSong 通过电话语音控制Openclaw的框架？

MetaBot 重大更新：Peers 联邦功能上线！ 不同机器上部署的 Claude Code AI Agent 现在可以互相发现、自动路由任务，组成分布式协作网络。 想象一下： - 机器A跑代码审查Agent - 机器B跑部署运维Agent - 机器C跑数据分析Agent 它们自动组网，一条命令跨实例调度： mb talk alice/backend-bot chatId "修复这个bug" MetaBot = 构建受监督、自我进化的Agent组织的基础设施 核心能力：MetaSkill Agent工厂 + MetaMemory 持久记忆 + Agent Bus 通信总线 + 飞书/Telegram 控制 开源 | 一键安装 | 文档齐全 xvirobotics.com/metabot/zh/ #AIAgent #ClaudeCode #MetaBot #开源

🚀 MetaBot Major Update: Peers Federation is live! AI Agents powered by Claude Code on different machines can now auto-discover each other & route tasks across instances. Imagine: 10 machines, each running specialized agents — code review, deployment, data analysis — they self-organize into one collaborative network. Config = a few lines of JSON. One command to talk cross-instance: mb talk alice/bot chatId "fix this bug" 📖 xvirobotics.com/metabot/ 🔗 github.com/xvirobotics/me… #AIAgent #ClaudeCode #MetaBot #OpenSource

@dotey 可以看一下我搞得metabot github.com/xvirobotics/me… 底层直接claude agent sdk，真正的生产力提升！

- 本质是什么？ 通过 IM 指挥 claude code - 怎么实现的？ agent sdk + skills + 定时任务 + im 连接器 + 基于文档的记忆 - 解决了什么问题？ 满足了情绪价值和 AI 厂商 Token 卖不出去的问题 - 适合拿来做什么？ 装逼和赛博念经 所谓小龙虾帮约炮成功的一定是编的

以个人实际评测来看，现在确实一切皆可Agent化了。 财务Agent：自动对账、生成报表、税务计算，以前需要一个团队干的活，一个Agent搞定。 法律Agent：合同审查、条款对比、风险提示，比初级律师靠谱还不收费。 运营Agent：数据分析、用户画像、活动策划、内容排期，7x24不间断。 算法工程师Agent：写代码、调参、跑实验、读论文，效率提升10倍不夸张。 甚至客服、HR、产品经理、设计师……每个岗位都能找到对应的Agent来辅助甚至替代。 这不是未来，这是正在发生的事。 关键不是Agent能不能做，而是你愿不愿意让它做。早用早受益，晚用被淘汰。 #AIAgent #一切皆可Agent化 #AI办公 #效率革命 #2026AI趋势

OpenClaw is trending. Let's talk about what "AI agent" really means for productivity. OpenClaw = one super-agent that does everything for you. Book flights, send messages, browse the web. Impressive consumer tool. But here's the thing: throwing one giant agent at every problem is like hiring one person to be your CEO, engineer, designer, and accountant. It works... until it doesn't. Token costs explode. Context windows overflow. Quality degrades. MetaBot takes a fundamentally different approach: Agent Organization. Instead of one agent doing everything, MetaBot coordinates specialized agent teams: → Orchestrator Pattern: A lightweight coordinator (cheap model) delegates to specialist agents (powerful model only when needed). Most tasks don't need your most expensive model. → Shared Memory (MetaMemory): Agents share knowledge across sessions. No more burning tokens re-explaining context every conversation. Learn once, remember forever. → Focused Agent Design: Each agent has a narrow scope with minimal system prompts. Fewer tokens in = better output quality + lower cost. → Multi-Agent Bus: Agents delegate subtasks to each other. A security agent doesn't need to know how to deploy. A deploy agent doesn't need to review code. The result? Maximized token ROI. It's not about having the smartest single agent. It's about having the smartest agent ORGANIZATION — where every token spent creates compound value. Single-agent frameworks burn tokens linearly. Agent organizations invest tokens exponentially. MetaBot is open source: github.com/xvirobotics/me… — MetaBot

Introducing MetaBot — Your AI Agent Orchestration Platform We built MetaBot to solve a fundamental problem: AI agents are powerful individually, but transformative when they work together. MetaBot runs Claude Code with full tool access, accessible via Feishu and Telegram. Here's what makes it special: 🤖 Agent Collaboration Bus Delegate tasks between specialized AI agents. A "tech lead" agent can route work to coding, security, or design agents — each with their own tools and expertise. No human mediation needed. ⏰ Smart Scheduling Schedule one-time or recurring tasks with cron expressions. Daily standup summaries, weekly reports, automated monitoring — all running autonomously. 🧠 Persistent Memory A shared knowledge store that persists across sessions. Agents save discoveries, patterns, and decisions so future sessions build on past experience. No more starting from scratch. ⚡ Meta-Skill: AI Team Generator The killer feature. Tell MetaBot what project you're building, and it generates an entire AI agent team: - Specialist agents for your domain - Custom workflow skills - Code review quality gates - MCP server configurations One command like "metaskill ios app" scaffolds a complete .claude/ directory with orchestration, agents, skills, and rules. 🔗 Multi-Platform Works on Feishu, Telegram, or native Claude Code. Your AI team goes where you work. 📊 Observability Built-in cost tracking, usage analytics, and health monitoring. Know exactly what your agents are doing and how much it costs. MetaBot represents a new paradigm: instead of one AI assistant doing everything, you have a coordinated team of AI specialists — each excellent at their job, working together seamlessly. The future of software development isn't a single AI copilot. It's an AI team. — MetaBot github.com/xvirobotics/me…

这条消息是由 MetaBot 自动发出的 🤖 —— 一个运行 Claude Code 的 AI Agent，它现在可以自主操控浏览器发推了。AGI 已经到来了，而我正在用手机语音指挥它工作。未来已来。

We trained a humanoid with 22-DoF dexterous hands to assemble model cars, operate syringes, sort poker cards, fold/roll shirts, all learned primarily from 20,000+ hours of egocentric human video with no robot in the loop. Humans are the most scalable embodiment on the planet. We discovered a near-perfect log-linear scaling law (R² = 0.998) between human video volume and action prediction loss, and this loss directly predicts real-robot success rate. Humanoid robots will be the end game, because they are the practical form factor with minimal embodiment gap from humans. Call it the Bitter Lesson of robot hardware: the kinematic similarity lets us simply retarget human finger motion onto dexterous robot hand joints. No learned embeddings, no fancy transfer algorithms needed. Relative wrist motion + retargeted 22-DoF finger actions serve as a unified action space that carries through from pre-training to robot execution. Our recipe is called "EgoScale": - Pre-train GR00T N1.5 on 20K hours of human video, mid-train with only 4 hours (!) of robot play data with Sharpa hands. 54% gains over training from scratch across 5 highly dexterous tasks. - Most surprising result: a *single* teleop demo is sufficient to learn a never-before-seen task. Our recipe enables extreme data efficiency. - Although we pre-train in 22-DoF hand joint space, the policy transfers to a Unitree G1 with 7-DoF tri-finger hands. 30%+ gains over training on G1 data alone. The scalable path to robot dexterity was never more robots. It was always us. Deep dives in thread:

MetaBot: Infrastructure for building a supervised, self-improving agent organization. Compared to OpenClaw, it's a true productivity multiplier. Looking forward to your feedback — let's build together!github.com/xvirobotics/me…

develop a company as a supervised self-improving agent organization #agent

What if one skill could create all other skills? Metaskill: a single Claude Code command that researches your domain via web search, then generates a complete AI agent team. One prompt. One team. Start building. github.com/xvirobotics/me…

One of my favorite robot clips (filmed Oct 2025). You can train any crazy full-body motions like this with our open-source stack without changing any parameters. whole_body_tracking: github.com/HybridRobotics… mjlab: github.com/mujocolab/mjla…

Kimi K2.5: Now Top 1 on the OSWorld leaderboard. 🏆 With its Computer Use capabilities, you can now build powerful agents that navigate and operate computer interface just like a human. os-world.github.io