I love to talk on trending ⚡ technology ⚡, stay hungry
- 🤔 For engineers building LLM applications, what matters most is being pragmatic.
- 🌱 I want to build more Harnesses—and through them, fully unlock the capabilities of models.
- 📚 Enjoying reading and accumulating knowledge
- 📖 Novels: 《剑来》 and 《雪中悍刀行》
- 📺 Cartoon: One Piece and 《一人之下》 and 《星游记》
- Multi-agent collaboration is a long-term challenge. Like context management, it is fascinating and worth exploring.
- Agent evaluation is the foundation for continuously improving Agents—and the starting point for everything good.
- Anything Worth Learning and Building: Open source, English, LLM inference infrastructure, Coding Agents,
- 🌈 大模型应用开发 -上下文工程与运行空间实践指南 - 上下文工程是设计原则,Agent Harness 是构建目标
- 🪐 Actspace Agent - Actspace 是一个本地 Agent 桌面应用,是模型的Harness,从Coding Agent开始,最终愿景是彻底释放模型能力
- 🍍 agent-harness-dev - 一份构建Agent后端的Skill,从Context Engineering与Harness Engineering的设计理念整理出来的
- 🚄 code-develop-harness-init - 面向 Agent-first 开发的项目初始化模板(Harness)—— 人定方向,Agent 执行
- 🎶 音频制作Agent - AI 驱动的音频创作协同平台 —— 让 AI 智能体帮你完成配音、音效、BGM 的智能匹配与生成
- 🌵 Reason Code - Reason Code 是一款编码Agent,它集成于终端环境中
- 🦾 HeartClaw - 会自我进化的个人 AI Agent,长短期记忆与自治巡检,还能通过「天工」自动锻造新工具,让 Agent 不断长出新能力
公众号「锦康灵感铺」 · 微信号「a2385472291」 · 2385472291@qq.com · 稀土掘金 · 小红书
我目前对新的机会保持开放。如果你的团队在做大模型应用 /Agent 相关产品,或者只是想交流想法、都非常欢迎随时找我聊聊



