[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f3h7qivqze38h":3},{"code":4,"message":5,"data":6,"timestamp":69},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":40,"summary_md":41,"topic":42,"github_topics":45,"license":65,"homepage":66,"updated_at":67,"history":68},"paperclipai\u002Fpaperclip",2,"The open-source app everyone uses to manage agents at work","TypeScript",106854,17747,25543,236,[16,22,28,32,36],{"id":17,"title":18,"points":19,"comments":20,"posted_at":21},47277096,"Paperclip: Open-source orchestration for zero-human companies",6,1,"2026-03-06",{"id":23,"title":24,"points":25,"comments":26,"posted_at":27},49831929,"Paperclip: Open-source orchestration for teams of AI agents",3,0,"2026-09-24",{"id":29,"title":30,"points":25,"comments":26,"posted_at":31},47903549,"Paperclip – a ticket-based multi AI agent orchestrator","2026-04-25",{"id":33,"title":34,"points":8,"comments":26,"posted_at":35},47339291,"Paperclip, Open-source orchestration for zero-human companies","2026-03-11",{"id":37,"title":38,"points":20,"comments":26,"posted_at":39},47324603,"Paperclip – Open-source orchestration for zero-human companies","2026-03-10","把多个异构编程 Agent 当成一家公司来管：目标、组织架构、预算和审计都在一个面板里。","## 解决什么问题\n\n当你同时跑着好几个 Agent——这里一个 Claude Code，那里一个 Codex，cron 上还挂着 OpenClaw 机器人——你很快就会失去掌控：谁在做哪件事、哪个任务卡住了发布、这个月 token 到底烧了多少钱、刚才动生产环境的 Agent 有没有权限。聊天窗口和终端标签回答不了这些问题。Paperclip 在 Agent 协作之上套了一个任务管理器式的界面：分配目标、定义角色和预算、审批产出、在一个地方查看审计记录。\n\n## 有什么不同\n\n现在多数 Agent 工具要么解决单 Agent 循环，要么绑定单一厂商运行时；LangGraph、CrewAI 这类框架给的是代码层的编排原语，IDE 里的托管 Agent 则维持单个会话的连贯性。Paperclip 换了个抽象：把每个 Agent 当成组织里的员工——组织架构、按范围隔离的密钥、审批关卡、成本跟踪——并且刻意做厂商无关，通过 adapter 接入 Claude Code、Codex、Cursor、Gemini CLI、OpenCode、Hermes 等。取舍在于它不去做模型 harness 本身，而是做你现有 harness 之上的管理层，所以真正的产品是治理和预算，而不是提示词。\n\n## 适合谁\n\n适合已经在每天跑多个编程 Agent、需要可见性、花费上限和评审检查点的小团队或个人开发者，尤其是人机混合协作的场景。不适合只在编辑器里跑一个 Agent 的人；也不适合今天就需要经过验证的企业级控制面的团队——README 里定位和截图很多，但架构、安全模型和失败行为写得很薄。\n\n## 上手成本\n\nNode.js 服务端加 React 前端，自托管，MIT 协议，没有厂商锁定，但部署和密钥都由你自己负责。每个支持的 harness 都要单独配凭据和 adapter，而且每多一个 Agent 都是实打实的月账单——预算功能之所以重要，是因为钱真的会花出去。项目很新（2026 年 3 月创建），README 里没有安装命令，得照着文档站来。厂商 CLI 变动时，adapter 层大概率要跟着改。\n\n## 点评\n\n在拥挤的赛道里，Paperclip 的\"Agent 公司\"框架最明确，组织架构加预算这个角度确实不同于代码层框架。热度很高：约八个月拿到约 107k stars、17.7k forks，最近一个月就涨了 25.5k（月榜第 4）——这是爆发而不是慢热，而且它没上过日榜、Hacker News 上讨论也不多，通常说明靠的是传播而非同行验证。结论是有潜力但未经验证：先在一个范围可控、能容忍 adapter 出问题的内部流程上试点，在治理能力被实践验证之前，自己保留一份审计日志。",{"slug":43,"name":44},"ai-engineering","AI 工程",[46,47,48,49,50,43,51,52,53,54,55,56,57,58,59,60,61,62,63,64],"acpx","agent-orchestration","agent-skills","agentic-ai","ai-agents","ai-governance","ai-tools","autonomous-agents","claude","claude-code","claude-skills","codex","hermes","llm","mcp","multi-agent","openclaw","project-management","workflow-automation","MIT","https:\u002F\u002Fpaperclip.ing","2026-10-24T14:10:24Z",[],1792870853]