[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:stajic:zh":3,"public-menus:all":38,"post:openai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026:zh":205,"related:post:openai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026:zh:1":1585},{"statusCode":4,"data":5,"message":37},200,{"tenantId":6,"lang":7,"defaultLang":8,"siteUrl":9,"contactEmail":10,"brandName":11,"logoUrl":12,"siteName":11,"siteDescription":13,"ogImage":10,"robotsIndex":14,"socialLinks":10,"reservedSlugs":10,"seoPolicy":15},"stajic","zh","de","https:\u002F\u002Fstajic.de",null,"Stajic Platform","\u002FLogo_Planet.svg","Stajic Portal",true,{"branding":16,"relatedContent":17,"crossDomainLinks":18},{"logoUrl":12},{"enabled":14},[19,22,25,28,31,34],{"url":20,"label":21,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Ffigure.rocks","figure.rocks",{"url":23,"label":24,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Floving.rocks","loving.rocks",{"url":26,"label":27,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.com","bazify.com",{"url":29,"label":30,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.de","bazify.de",{"url":32,"label":33,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.at","bazify.at",{"url":35,"label":36,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.ba","bazify.ba","Portal settings resolved",[39,45],{"id":40,"name":41,"location":42,"isActive":14,"isDefault":43,"items":44},1,"main-navigation","header",false,[],{"id":46,"name":47,"location":48,"isActive":14,"isDefault":14,"items":49},4,"main-menu","sidebar",[50,66,79,93,103,118,133],{"id":51,"title":52,"url":60,"target":61,"icon":62,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":64,"portfolioId":10,"children":65},"item-18",{"de":53,"en":54,"es":55,"fr":56,"it":54,"ru":57,"sr":58,"zh":59},"Startseite","Home","Inicio","Accueil","Главная","Почетна","首页","\u002Ffull-stack-web-developer-munich-performance-seo-and-maintainable-builds","_self","i-lucide-home","page",111,[],{"id":67,"title":68,"url":75,"target":61,"icon":76,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":77,"portfolioId":10,"children":78},"item-22",{"de":69,"en":69,"es":70,"fr":69,"it":71,"ru":72,"sr":73,"zh":74},"Vision","Visión","Visione","Видение","Визија","想象","\u002Fueber-uns-webdesign-muenchen-webaplikation","i-lucide-eye",113,[],{"id":80,"title":81,"url":89,"target":61,"icon":90,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":91,"portfolioId":10,"children":92},"item-19",{"de":82,"en":83,"es":84,"fr":83,"it":85,"ru":86,"sr":87,"zh":88},"Leistungen","Services","Servicios","Servizi","Услуги","Услуге","服务","\u002Fservices-dienstleistungen-muenchen","i-lucide-wrench",116,[],{"id":94,"title":95,"url":99,"target":61,"icon":100,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":101,"portfolioId":10,"children":102},"item-23",{"de":96,"en":96,"es":96,"fr":96,"it":96,"ru":97,"sr":97,"zh":98},"Blog","Блог","博客","\u002Fblog","i-lucide-book-open",112,[],{"id":104,"title":105,"url":114,"target":61,"icon":115,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":116,"portfolioId":10,"children":117},"item-32",{"de":106,"en":107,"es":108,"fr":109,"it":110,"ru":111,"sr":112,"zh":113},"Neue Technologien","New Technologies","Nuevas tecnologías","Nouvelles technologies","Nuove tecnologie","Новые технологии","Нове технологије","新技术！","\u002Fneue-webtechnologien","i-lucide-sparkles",122,[],{"id":119,"title":120,"url":129,"target":61,"icon":130,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":131,"portfolioId":10,"children":132},"item-20",{"de":121,"en":122,"es":123,"fr":124,"it":125,"ru":126,"sr":127,"zh":128},"Kontakt","Contact us!","Contacto","Contact","Contatto","Контакт","Контактирајте нас","联系我们！","\u002Fcontact","i-lucide-mail",115,[],{"id":134,"title":135,"url":144,"target":61,"icon":145,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":147},"item-21",{"de":136,"en":137,"es":138,"fr":139,"it":140,"ru":141,"sr":142,"zh":143},"Unsere Arbeit","Our Work","Nuestro trabajo","Nos réalisations","I nostri lavori","Наши работы","Наши радови","文件夹","\u002Fportfolio","i-lucide-briefcase",114,[148,161,175,181,193],{"id":149,"title":150,"url":144,"target":61,"icon":159,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":160},"item-24",{"de":151,"en":152,"es":153,"fr":154,"it":155,"ru":156,"sr":157,"zh":158},"Alle Projekte","All Projects","Todos los proyectos","Tous les projets","Tutti i progetti","Все проекты","Сви пројекти","所有项目","i-lucide-grid-3x3",[],{"id":162,"title":163,"url":171,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":174},"item-29",{"de":164,"en":165,"es":166,"fr":167,"it":168,"ru":169,"sr":170,"zh":143},"Local Roots, Global Reach","Local Roots - Global Reach","Empresa local ","Entreprise locale","Azienda locale","Местная компания","Локално предузеће глобално тржиште","\u002Fportfolio\u002Flocal-roots-global-reach-communication-media-systems-for-modern-business","i-lucide-folder","custom",[],{"id":176,"title":177,"url":179,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":180},"item-28",{"de":178,"en":178,"es":178,"fr":178,"it":178,"ru":178,"sr":178,"zh":178},"Solr Suggester","\u002Fportfolio\u002Fsolr-fuzzy-suggester-und-solr-infix-suggester-abfrage-ueber-ajax-und-filterung",[],{"id":182,"title":183,"url":191,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":192},"item-27",{"de":184,"en":185,"es":186,"fr":187,"it":188,"ru":189,"sr":190,"zh":185},"Firmenwebseite SEO","Company Website SEO","Sitio web corporativo SEO","Site web d’entreprise SEO","Sito web aziendale SEO","Корпоративный сайт SEO","Пословна веб-страница SEO","\u002Fportfolio\u002Fseo-sem-branding-mobile-webseite-muenchen",[],{"id":194,"title":195,"url":203,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":204},"item-31",{"de":196,"en":197,"es":198,"fr":199,"it":200,"ru":201,"sr":202,"zh":197},"Digitalisierungsportal","Digitalization Portal","Portal de digitalización","Portail de numérisation","Portale di digitalizzazione","Портал цифровизации","Портал за дигитализацију","\u002Fportfolio\u002Fdigitalisierungsportal-archiv-museum-bibliothek-ead-lido-mets-mods",[],{"statusCode":4,"data":206,"message":1584},{"id":207,"title":208,"slug":209,"content":210,"contentJson":211,"excerpt":795,"featuredImage":796,"featuredImageAlt":797,"featuredImageCaption":10,"featuredImageTitle":10,"featuredImageCopyright":10,"featuredImageAuthor":10,"featuredImageSourceUrl":10,"featuredImageLicense":10,"featuredImageIsAiGenerated":43,"status":798,"publishedAt":799,"createdAt":800,"updatedAt":801,"seoLocalePaths":802,"categories":811,"author":823,"translations":828},"473","OpenAI Agents API 与 Agents SDK 与 Responses API：2026 年你应该基于什么来构建？","openai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u003Cnav class=\"editorjs-toc\" data-editorjs-toc=\"true\" aria-label=\"目录\">\u003Cstrong class=\"editorjs-toc__title\">目录\u003C\u002Fstrong>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-0\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-6\" class=\"editorjs-toc__link\">架构已经改变：选择的是运行时边界，而不是库\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-9\" class=\"editorjs-toc__link\">简要对比\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-11\" class=\"editorjs-toc__link\">1. Agents API：托管执行框架，持久化云端智能体\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-15\" class=\"editorjs-toc__link\">使用 Agents API 你能获得什么\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-17\" class=\"editorjs-toc__link\">你仍然需要负责的部分\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-19\" class=\"editorjs-toc__link\">2. Responses API：掌控循环，使用平台原语\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-23\" class=\"editorjs-toc__link\">3. Agents SDK：仍受支持，但不再是默认的前进路径\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-28\" class=\"editorjs-toc__link\">Codex SDK 的定位\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-31\" class=\"editorjs-toc__link\">运行时所有权测试\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-34\" class=\"editorjs-toc__link\">新系统的决策树\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-36\" class=\"editorjs-toc__link\">什么不应驱动决策\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-38\" class=\"editorjs-toc__link\">迁移是架构变更，而不是重命名导入\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-41\" class=\"editorjs-toc__link\">迁移清单\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-43\" class=\"editorjs-toc__link\">公开测试版改变了风险模型\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-46\" class=\"editorjs-toc__link\">一种实用的默认架构\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-50\" class=\"editorjs-toc__link\">什么会改变这个答案？\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-53\" class=\"editorjs-toc__link\">局限性\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-56\" class=\"editorjs-toc__link\">结论\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-59\" class=\"editorjs-toc__link\">常见问题\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-61\" class=\"editorjs-toc__link\">术语表\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-63\" class=\"editorjs-toc__link\">主要来源与延伸阅读\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fnav>\n\u003Cp>OpenAI 的智能体技术栈在 2026 年 9 月发生了重大变化。新的 Agents API 引入了托管的 Codex 执行框架，用于持久化的云端智能体，而较旧的 Agents SDK 则进入了功能完备的维护阶段。Responses API 仍然是希望直接调用模型或自行掌控智能体循环的应用程序所使用的较低层级接口。这三者并非同一事物的三种可互换封装：它们将运行时边界置于不同的位置。\u003C\u002Fp>\n\u003Caside class=\"editorjs-callout editorjs-callout--info my-6 rounded-xl border p-5 border-blue-300 bg-blue-50 dark:border-blue-900 dark:bg-blue-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">直接回答\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\"><strong>对于 2026 年 9 月新建的 OpenAI 原生智能体应用，请从 Agents API 开始，除非你有具体理由需要自行掌控更多运行时。<\u002Fstrong>当你的应用程序应当掌控循环、状态策略和编排时，请使用 Responses API。对于现有应用程序，或 Agents API 尚不支持某项所需能力的短期缺口，请保留 Agents SDK。如果你想要 Codex 执行框架本身，但需要在你运营的基础设施中运行它，请评估 Codex SDK，而不是强行让 Agents SDK 承担该角色。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Caside class=\"editorjs-callout editorjs-callout--warning my-6 rounded-xl border p-5 border-amber-300 bg-amber-50 dark:border-amber-900 dark:bg-amber-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">时间敏感的架构决策\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">本文反映的是 OpenAI 平台在 <strong>2026 年 9 月 25 日<\u002Fstrong> 的状态。Agents API 于 2026 年 9 月 10 日启动公开测试。测试版接口、功能覆盖范围、定价和迁移指南都可能发生变化；在确定生产架构之前，请核实当前文档。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Caside class=\"editorjs-callout editorjs-callout--note my-6 rounded-xl border p-5 border-gray-300 bg-gray-50 dark:border-gray-700 dark:bg-gray-900\u002F40\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">此处使用的决策模型\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">下文的运行时所有权测试是一种实用的架构模型，而非 OpenAI 的产品分类法。它询问的是：智能体运行时的哪些部分必须由你的团队控制，哪些部分你希望由 OpenAI 来运营。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-6\">架构已经改变：选择的是运行时边界，而不是库\u003C\u002Fh2>\n\u003Cp>重要的决策不再只是“我应该安装哪个 SDK？”而是由谁拥有执行框架、智能体循环、持久化会话状态、上下文压缩、恢复、执行环境和应用程序生命周期。\u003C\u002Fp>\n\u003Cp>OpenAI 当前的 Agents 概览明确说明了这一边界。Agents API 运行托管的 Codex 执行框架，并管理编排以及持久化会话状态。Responses API 提供模型响应和托管能力，而由你的应用程序拥有周围的智能体循环。Agents SDK 在你的应用程序中运行循环，如今已功能完备，而不再是重大新智能体功能的前进路径。\u003C\u002Fp>\n\u003Ch2 id=\"section-9\">简要对比\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">选项\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">2026 年的最佳适用场景\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">谁拥有智能体循环？\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">会话 \u002F 上下文责任\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">战略状态\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Agents API\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">新的 OpenAI 原生持久化智能体\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">OpenAI 托管的 Codex 执行框架\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">OpenAI 管理会话、编排、压缩和恢复\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">新智能体应用的推荐起点；公开测试\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Responses API\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">直接模型集成和自定义智能体运行时\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">你的应用程序\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">由你选择响应链式调用、Conversations、存储和循环逻辑\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">核心 API 原语；对于新项目，推荐优先于 Chat Completions\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Agents SDK\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">现有 SDK 应用程序或临时能力缺口\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">你的应用程序通过 SDK 运行器\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">你的应用程序负责部署、存储和运行时行为\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">功能完备；维护和兼容性继续，不计划推出重大新功能\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Codex SDK\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">在你运营的基础设施中使用 Codex 执行框架\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">你环境中的 Codex 执行框架\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">你负责执行框架的托管和生命周期\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">当你想要执行框架但不想要托管的 Agents API 运行时时，这是一个独立选项\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-11\">1. Agents API：托管执行框架，持久化云端智能体\u003C\u002Fh2>\n\u003Cp>Agents API 通过 OpenAI 托管的服务公开 Codex 执行框架。OpenAI 管理会话、编排、上下文压缩和恢复。你的应用程序仍然提供工具并选择执行环境。\u003C\u002Fp>\n\u003Cp>最后这一点区别很重要。“托管智能体”并不一定意味着“所有计算都在 OpenAI 内部运行”。Agents API 架构支持无环境、OpenAI 托管环境，或连接到托管执行框架的自托管环境。使用自托管环境时，你的应用程序负责配置、重新连接、关闭和持久化文件，而执行框架仍由托管方管理。\u003C\u002Fp>\n\u003Cp>因此，Agents API 是一种运行时服务，而不仅仅是一种请求格式。会话可以持久化、流式传输进度、接收额外任务、使用工具、处理文件，并在长时间运行的工作中恢复。\u003C\u002Fp>\n\u003Ch3 id=\"section-15\">使用 Agents API 你能获得什么\u003C\u002Fh3>\n\u003Cul>\u003Cli>托管的 Codex 执行框架，无需自行构建和运营主要的智能体循环。\u003C\u002Fli>\u003Cli>持久化会话，适用于跨多轮和长时间运行的任务。\u003C\u002Fli>\u003Cli>托管的编排、上下文压缩和恢复。\u003C\u002Fli>\u003Cli>根据工作负载需求，可选择 OpenAI 托管或自托管的执行环境。\u003C\u002Fli>\u003Cli>用于进度和生命周期事件的流式传输和 Webhook。\u003C\u002Fli>\u003Cli>OpenAI 明确推荐用于新智能体应用的平台方向。\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch3 id=\"section-17\">你仍然需要负责的部分\u003C\u002Fh3>\n\u003Cul>\u003Cli>你的产品和应用服务器。\u003C\u002Fli>\u003Cli>函数工具实现和业务逻辑。\u003C\u002Fli>\u003Cli>围绕你自己系统的授权和策略决策。\u003C\u002Fli>\u003Cli>当你选择自托管计算时的执行环境生命周期。\u003C\u002Fli>\u003Cli>评估、验收标准、领域特定的防护措施，以及决定智能体被允许做什么。\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch2 id=\"section-19\">2. Responses API：掌控循环，使用平台原语\u003C\u002Fh2>\n\u003Cp>当你希望使用 OpenAI 模型和工具能力，而不想将整个智能体运行时委托出去时，Responses API 是更低层级的选择。OpenAI 将 Responses 描述为新项目推荐的 API 原语，以及 Chat Completions 的演进，具备内置工具、多轮状态选项、多模态输入和智能体工具使用能力。\u003C\u002Fp>\n\u003Cp>一个 Responses 请求本身可以调用工具，但当你围绕它构建智能体时，你的应用程序仍需负责更大的工作流。这意味着你的代码决定如何持久化应用状态、何时继续、如何恢复、如何协调专家、如何压缩长历史记录，以及如何表示可恢复的工作。\u003C\u002Fp>\n\u003Cp>这本身并不逊色。当智能体行为必须深度嵌入现有应用逻辑、当你需要自定义状态模型，或当托管式运行框架会隐藏你真正需要的控制权时，这就是正确的边界。\u003C\u002Fp>\n\u003Ch2 id=\"section-23\">3. Agents SDK：仍受支持，但不再是默认的前进路径\u003C\u002Fh2>\n\u003Cp>Agents SDK 仍然是一个开源框架，用于在你的应用程序中运行智能体工作流。它在 TypeScript 和 Python 中提供智能体定义、工具、交接、护栏、追踪、会话和运行器循环。\u003C\u002Fp>\n\u003Cp>但其战略地位已经改变。OpenAI 现在将 Agents SDK 标记为功能完备：维护、安全修复、关键错误修复和兼容性工作仍在继续，但不计划推出重大新功能。OpenAI 推荐新应用程序使用 Agents API。\u003C\u002Fp>\n\u003Cp>这并不意味着现有的 SDK 应用程序应立即重写。它意味着架构不应再假设 SDK 是下一个重大智能体运行时能力的落点。\u003C\u002Fp>\n\u003Caside class=\"editorjs-callout editorjs-callout--warning my-6 rounded-xl border p-5 border-amber-300 bg-amber-50 dark:border-amber-900 dark:bg-amber-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">不要将“功能完备”翻译为“已弃用”\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">OpenAI 并未将 Agents SDK 描述为不受支持。现有应用程序可以继续使用它，维护、安全修复、关键错误修复和兼容性工作仍在继续。重要的变化是，不计划推出重大新功能，并且新应用程序被引导至 Agents API。\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-28\">Codex SDK 的定位\u003C\u002Fh2>\n\u003Cp>当前的选择并不是简单的三岔路口。OpenAI 的运行时概览将 Codex SDK 列为在你运营的基础设施中运行 Codex 运行框架的选项。这在架构上既不同于托管的 Agents API，也不同于 Agents SDK。\u003C\u002Fp>\n\u003Cp>如果你的真实需求是“我想要 Codex 运行框架，但我需要自己运营它”，那么 Codex SDK 就是值得评估的界面。如果你的需求是“我想掌控围绕模型调用的循环”，请评估 Responses。如果你的需求是“我已经有一个可用的 Agents SDK 应用程序”，现有 SDK 可以继续有效，同时你可以根据其维护状态进行规划。\u003C\u002Fp>\n\u003Ch2 id=\"section-31\">运行时所有权测试\u003C\u002Fh2>\n\u003Cp>有用的架构决策始于确定你的团队必须拥有什么。将每项需求评分为必须控制、倾向于控制或倾向于托管。\u003C\u002Fp>\n\u003Csection class=\"editorjs-comparison my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">运行时所有权测试\u003C\u002Fh3>\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left dark:border-gray-700 dark:bg-gray-900\">\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">决策\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">如果你倾向于托管\u003C\u002Fth>\u003Cth class=\"border border-gray-300 bg-gray-50 px-4 py-3 text-left font-semibold dark:border-gray-700 dark:bg-gray-900\">如果你需要控制\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">智能体循环\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">持久会话\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">运行框架运行时\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">执行环境\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">编排语义\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">提供商 \u002F 传输灵活性\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-3 text-left font-semibold dark:border-gray-700\">运维负担\u003C\u002Fth>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-3 dark:border-gray-700\">\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-34\">新系统的决策树\u003C\u002Fh2>\n\u003Csection class=\"editorjs-process my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">根据控制边界选择运行时\u003C\u002Fh3>\u003Cdiv class=\"grid grid-cols-1 md:grid-cols-2 xl:grid-cols-3 gap-4\">\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">1\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">1. 这是一个新的智能体应用程序吗？\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">如果否，不要仅仅因为存在更新的界面就迁移。首先评估当前应用程序的实际约束。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">2\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">2. 你想要一个托管的长运行智能体运行框架吗？\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">如果是，从 Agents API 开始，这是 OpenAI 推荐的新智能体应用程序路径。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">3\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">3. 你需要 Codex 运行框架但必须自己运营它吗？\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">评估 Codex SDK，而不是在 Agents SDK 之上重建运行框架行为。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">4\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">4. 你需要拥有智能体循环和状态模型吗？\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">使用 Responses API 作为更低层级的平台原语，并围绕它构建循环。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">5\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">5. 你已经在使用 Agents SDK 了吗？\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">如果它满足需求，请继续使用；不计划推出重大新功能，因此将未来平台采用视为明确的路线图决策。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">6\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">6. Agents API 是否缺少所需能力？\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">OpenAI 明确允许 Agents SDK 作为需要不受支持能力的新应用程序的短期选项。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv class=\"editorjs-process__step min-w-0  rounded-xl border border-gray-200 dark:border-gray-700 p-4\">\u003Cdiv class=\"text-xs font-semibold text-gray-500 dark:text-gray-400\">7\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 font-semibold text-gray-900 dark:text-gray-100\">7. 通过生产形态的验证性开发进行验证\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">在锁定架构之前，测试工具、环境、审批、延迟、可观测性、故障恢复和生命周期。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-36\">什么不应驱动决策\u003C\u002Fh2>\n\u003Cdiv class=\"overflow-x-auto\">\u003Ctable class=\"w-full border-collapse\">\u003Cthead>\u003Ctr>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">薄弱的决策规则\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">为何失败\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">更好的问题\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“最新的 API 一定最好。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">较新的方案可能在战略上更受青睐，但仍可能缺少你所需的能力。\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">哪些运行时职责应由平台管理，哪些应由应用自行拥有？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“我们已经熟悉这个 SDK。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">团队熟悉度可能维护着一个路线图已经改变的架构。\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">在下一个产品周期内，留下与迁移的成本分别是多少？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“托管就意味着没有基础设施。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Agents API 仍可使用自托管环境，你的应用仍拥有产品逻辑。\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">实际被委托出去的是哪一层基础设施？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“Responses 只适用于简单调用。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Responses 提供内置工具和有状态原语；它可以作为自定义智能体循环的基础。\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">我们需要平台拥有整个执行框架，还是只需要模型\u002F工具原语？\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">“功能完备就意味着我们必须立即迁移。”\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">SDK 仍会为现有应用继续维护。\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">继续留下会阻碍哪些具体的未来需求？\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-38\">迁移是架构变更，而不是重命名导入\u003C\u002Fh2>\n\u003Cp>从 Agents SDK 迁移到 Agents API 会改变所有权归属。在 SDK 中，循环运行在你的应用内。在 Agents API 中，OpenAI 运行执行框架和会话，而你的应用通过任务、事件、工具和环境边界进行集成。\u003C\u002Fp>\n\u003Cp>因此，真正的迁移计划需要梳理会话状态、自定义编排、交接、工具执行、审批、存储、追踪、重试、故障恢复、环境生命周期，以及任何特定于提供商的抽象。代码量可能下降，但运维假设会发生变化。\u003C\u002Fp>\n\u003Ch3 id=\"section-41\">迁移清单\u003C\u002Fh3>\n\u003Cul>\u003Cli>智能体定义和指令所有权。\u003C\u002Fli>\u003Cli>工具定义以及每个工具的执行位置。\u003C\u002Fli>\u003Cli>交接、管理者\u002F专家模式以及子智能体行为。\u003C\u002Fli>\u003Cli>会话标识符、对话状态、可恢复性和历史保留。\u003C\u002Fli>\u003Cli>人工审批和中断语义。\u003C\u002Fli>\u003Cli>自定义上下文裁剪或压缩逻辑。\u003C\u002Fli>\u003Cli>追踪、评估、可观测性和生产调试。\u003C\u002Fli>\u003Cli>自托管文件、容器、私有网络访问或其他执行依赖。\u003C\u002Fli>\u003Cli>提供商抽象或非 OpenAI 模型依赖。\u003C\u002Fli>\u003Cli>重试、超时、幂等性、恢复和生命周期假设。\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch2 id=\"section-43\">公开测试版改变了风险模型\u003C\u002Fh2>\n\u003Cp>Agents API 是新智能体应用的推荐方向，但它也处于公开测试阶段。这些事实并不矛盾。战略方向回答的是“平台将走向何方？”测试版状态回答的是“我应该为多少接口和运维变化预留预算？”\u003C\u002Fp>\n\u003Cp>对于生产系统，应将集成隔离在应用边界之后。尽可能将领域状态、权限、审计数据和业务规则保留在供应商特定的会话对象之外。这样更容易吸收 API 演进，而不会让智能体运行时成为整个产品的真相来源。\u003C\u002Fp>\n\u003Ch2 id=\"section-46\">一种实用的默认架构\u003C\u002Fh2>\n\u003Cp>对于许多新的 OpenAI 原生应用，2026 年一个合理的默认方案是：使用 Agents API 提供托管执行框架和持久会话，应用拥有领域服务和授权，业务操作使用显式函数工具，并根据数据和计算需求选择 OpenAI 托管或自托管执行。\u003C\u002Fp>\n\u003Cp>这样可以让智能体运行时保持强大，同时不让它成为业务真相的所有者。应用仍然决定用户可以做什么、哪些数据具有权威性、哪些操作需要审批，以及如何验证结果。\u003C\u002Fp>\n\u003Caside class=\"editorjs-referral my-6\">\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fai-agent-reliability-why-the-final-answer-is-not-enough\" class=\"flex flex-col sm:flex-row gap-4 rounded-xl border border-gray-200 dark:border-gray-700 p-4 transition hover:border-primary-500\">\u003Cdiv class=\"min-w-0 flex-1\">\u003Cstrong class=\"block text-lg text-gray-900 dark:text-gray-100\">AI 智能体可靠性：为什么最终答案还不够\u003C\u002Fstrong>\u003Cp class=\"mt-2 text-sm text-gray-600 dark:text-gray-300\">托管执行框架可以减轻运维负担，但可靠性仍然需要可追踪的执行、明确的验收标准，以及超越最终答案的评估。\u003C\u002Fp>\u003Cspan class=\"mt-3 inline-flex text-sm font-medium text-primary-600 dark:text-primary-400\">阅读可靠性文章 →\u003C\u002Fspan>\u003C\u002Fdiv>\u003C\u002Fa>\u003C\u002Faside>\n\u003Ch2 id=\"section-50\">什么会改变这个答案？\u003C\u002Fh2>\n\u003Cp>如果 Agents API 增加或移除能力、以不同的契约退出测试版、改变环境或定价边界，或引入减少所有权差异的迁移工具，那么建议就会改变。如果你的应用依赖提供商可移植性、自定义编排语义、仅本地执行，或托管执行框架无法支持的能力，建议也会改变。\u003C\u002Fp>\n\u003Cp>对于现有的 Agents SDK 应用，答案还会随迁移成本而变化。如果系统稳定、经过充分评估，并且没有被 SDK 的功能完备状态所阻碍，那么立即迁移可能带来的风险大于价值。如果产品路线图依赖于仅在 Agents API 中提供的能力，那么延迟迁移可能会造成另一种债务。\u003C\u002Fp>\n\u003Ch2 id=\"section-53\">局限性\u003C\u002Fh2>\n\u003Cp>本比较聚焦于运行时所有权和 OpenAI 所声明的平台方向。它不会针对特定工作负载对延迟、质量或总成本进行基准测试。这些属性取决于模型选择、工具使用、环境、任务长度、缓存、沙箱使用和应用架构。\u003C\u002Fp>\n\u003Cp>Agents API 也足够新，生产经验仍在积累中。因此，设计应通过具有代表性的工作负载进行验证，而不是仅根据产品定位来选择。\u003C\u002Fp>\n\u003Ch2 id=\"section-56\">结论\u003C\u002Fh2>\n\u003Cp>2026 年 OpenAI 智能体决策的核心在于运行时所有权。Agents API 意味着 OpenAI 拥有更多 harness 和持久会话机制。Responses 意味着你的应用程序拥有围绕平台原语的循环。Agents SDK 对现有系统仍然有效，但已不再是重大新智能体运行时功能的默认目标。\u003C\u002Fp>\n\u003Cp>对于新应用程序，除非有实际需求推动你向技术栈更低层移动，否则应遵循平台方向。从 Agents API 开始，当你需要拥有循环时转向 Responses，当你需要将 harness 部署在自己的基础设施中时评估 Codex SDK，并在现有投资或临时能力缺口使其合理时保留 Agents SDK。\u003C\u002Fp>\n\u003Ch2 id=\"section-59\">常见问题\u003C\u002Fh2>\n\u003Csection class=\"editorjs-faq my-6 rounded-xl border border-gray-200 p-5 dark:border-gray-700\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">2026 年 OpenAI 智能体运行时选择\u003C\u002Fh3>\u003Cdiv id=\"faq1\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">对于新项目，我应该使用 OpenAI Agents API 还是 Agents SDK？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">OpenAI 目前推荐新智能体应用程序使用 Agents API。Agents SDK 功能已完备，并继续为现有应用程序提供支持，维护、安全、关键错误修复和兼容性工作仍在继续。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq2\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">Agents SDK 是否已弃用？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">OpenAI 将其描述为功能已完备，而非不受支持。不计划推出重大新功能，但维护、安全修复、关键错误修复和兼容性工作仍在继续。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq3\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">何时应使用 Responses API 而非 Agents API？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">当你的应用程序应拥有智能体循环、状态策略、编排和延续逻辑，同时仍使用 OpenAI 模型和平台工具时，请使用 Responses。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq4\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">Agents API 是否需要 OpenAI 托管的计算资源？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">不需要。Agents API 可以使用无执行环境、OpenAI 托管环境，或连接到托管 harness 的自托管环境。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq5\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">Codex SDK 适用于什么场景？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">OpenAI 将 Codex SDK 定位为在你运营的基础设施中运行 Codex harness。当你想要 harness 但不想要托管的 Agents API 运行时，它是相关选项。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003Cdiv id=\"faq6\" class=\"border-t border-gray-200 py-4 first:border-t-0 dark:border-gray-700\">\u003Ch4 class=\"font-semibold text-gray-900 dark:text-gray-100\">现有的 Agents SDK 应用是否应立即迁移？\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">不应自动迁移。评估当前系统是否因 SDK 的功能已完备状态而受阻，所需能力是否存在于 Agents API 中，以及迁移收益是否超过运营和架构变更成本。\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-61\">术语表\u003C\u002Fh2>\n\u003Csection class=\"editorjs-glossary my-6 rounded-xl border border-gray-200 dark:border-gray-700 p-5\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">关键运行时术语\u003C\u002Fh3>\u003Cdl>\u003Cdiv id=\"harness\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">Harness\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">协调模型调用、工具、上下文、会话和智能体执行的运行时循环及支持机制。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"agents-api\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">Agents API\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">OpenAI 的托管 API，用于使用托管 Codex harness 的持久云智能体。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"responses-api\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">Responses API\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">OpenAI 的较低层 API 原语，用于模型响应、托管工具和有状态交互，应用程序可围绕其构建自己的智能体循环。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"agents-sdk\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">Agents SDK\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">OpenAI 的开源框架，用于在应用程序代码中运行智能体工作流；截至 2026 年 9 月功能已完备。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"codex-sdk\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">Codex SDK\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">OpenAI 列出的运行时选项，用于在你运营的基础设施中运行 Codex harness。\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"runtime-ownership\" class=\"border-t border-gray-200 dark:border-gray-700 py-3 first:border-t-0\">\u003Cdt class=\"font-semibold text-gray-900 dark:text-gray-100\">运行时所有权\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">描述智能体循环、会话状态、执行环境和生命周期中哪些部分由平台运营、哪些由应用程序运营的架构边界。\u003C\u002Fdd>\u003C\u002Fdiv>\u003C\u002Fdl>\u003C\u002Fsection>\n\u003Ch2 id=\"section-63\">主要来源与延伸阅读\u003C\u002Fh2>\n\u003Ca href=\"https:\u002F\u002Fopenai.com\u002Findex\u002Fintroducing-the-agents-api\u002F\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — 推出 Agents API\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">2026 年 9 月 10 日 Agents API 的发布公告，介绍了托管的 Codex harness 和公开测试版。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — Agents 运行时概述\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">当前对 Agents API、Codex SDK 和 Responses API 的比较，包括 Agents SDK 的支持状态。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Foverview\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — Agents API 概述\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">关于持久云智能体、会话、编排、上下文压缩、恢复和环境选择的文档。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Farchitecture\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — Agents API 架构\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">托管 harness、应用服务器以及无\u002FOpenAI 托管\u002F自托管执行环境之间的架构边界。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\u002Fsdk\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — Agents SDK\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">当前 Agents SDK 支持通知以及对应用程序拥有的智能体循环的说明。\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fmigrate-to-responses\" target=\"_blank\" rel=\"noopener noreferrer\" class=\"editorjs-link-tool block border border-gray-200 dark:border-gray-700 rounded-lg p-4 transition text-gray-900 dark:text-gray-100 hover:border-primary-500 hover:bg-primary-50 dark:hover:bg-gray-900 hover:text-gray-900 dark:hover:text-gray-100\">\u003Cstrong class=\"block font-semibold\">OpenAI — 迁移到 Responses API\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">当前 Responses API 的定位、内置工具、有状态上下文和智能体原语。\u003C\u002Fp>\u003C\u002Fa>\n\u003Cnav class=\"editorjs-toc\" data-editorjs-toc=\"true\" aria-label=\"目录\">\u003Cstrong class=\"editorjs-toc__title\">目录\u003C\u002Fstrong>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-0\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-6\" class=\"editorjs-toc__link\">架构已经改变：选择的是运行时边界，而不是库\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-9\" class=\"editorjs-toc__link\">简要对比\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-11\" class=\"editorjs-toc__link\">1. Agents API：托管执行框架，持久化云端智能体\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-15\" class=\"editorjs-toc__link\">使用 Agents API 你能获得什么\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-17\" class=\"editorjs-toc__link\">你仍然需要负责的部分\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-19\" class=\"editorjs-toc__link\">2. Responses API：掌控循环，使用平台原语\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-23\" class=\"editorjs-toc__link\">3. Agents SDK：仍受支持，但不再是默认的前进路径\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-28\" class=\"editorjs-toc__link\">Codex SDK 的定位\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-31\" class=\"editorjs-toc__link\">运行时所有权测试\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-34\" class=\"editorjs-toc__link\">新系统的决策树\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-36\" class=\"editorjs-toc__link\">什么不应驱动决策\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-38\" class=\"editorjs-toc__link\">迁移是架构变更，而不是重命名导入\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-41\" class=\"editorjs-toc__link\">迁移清单\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-43\" class=\"editorjs-toc__link\">公开测试版改变了风险模型\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-46\" class=\"editorjs-toc__link\">一种实用的默认架构\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-50\" class=\"editorjs-toc__link\">什么会改变这个答案？\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-53\" class=\"editorjs-toc__link\">局限性\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-56\" class=\"editorjs-toc__link\">结论\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-59\" class=\"editorjs-toc__link\">常见问题\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-61\" class=\"editorjs-toc__link\">术语表\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-63\" class=\"editorjs-toc__link\">主要来源与延伸阅读\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fnav>",{"time":212,"blocks":213,"version":794},1790365885872,[214,222,228,236,243,250,255,260,265,270,306,311,316,321,326,331,345,350,361,366,371,376,381,386,391,396,401,407,412,417,422,427,432,477,482,511,516,545,550,555,560,565,581,586,591,596,601,606,611,620,625,630,635,640,645,650,655,660,665,670,700,705,730,735,745,754,763,772,781,790],{"id":215,"data":216,"type":220,"tunes":221},"75qcDhG72q",{"title":217,"maxLevel":218,"minLevel":219},"目录",3,2,"tableOfContents",{},{"id":223,"data":224,"type":226,"tunes":227},"intro",{"text":225},"OpenAI 的智能体技术栈在 2026 年 9 月发生了重大变化。新的 Agents API 引入了托管的 Codex 执行框架，用于持久化的云端智能体，而较旧的 Agents SDK 则进入了功能完备的维护阶段。Responses API 仍然是希望直接调用模型或自行掌控智能体循环的应用程序所使用的较低层级接口。这三者并非同一事物的三种可互换封装：它们将运行时边界置于不同的位置。","paragraph",{},{"id":229,"data":230,"type":234,"tunes":235},"direct",{"body":231,"title":232,"variant":233},"\u003Cstrong>对于 2026 年 9 月新建的 OpenAI 原生智能体应用，请从 Agents API 开始，除非你有具体理由需要自行掌控更多运行时。\u003C\u002Fstrong>当你的应用程序应当掌控循环、状态策略和编排时，请使用 Responses API。对于现有应用程序，或 Agents API 尚不支持某项所需能力的短期缺口，请保留 Agents SDK。如果你想要 Codex 执行框架本身，但需要在你运营的基础设施中运行它，请评估 Codex SDK，而不是强行让 Agents SDK 承担该角色。","直接回答","info","callout",{},{"id":237,"data":238,"type":234,"tunes":242},"date-warning",{"body":239,"title":240,"variant":241},"本文反映的是 OpenAI 平台在 \u003Cstrong>2026 年 9 月 25 日\u003C\u002Fstrong> 的状态。Agents API 于 2026 年 9 月 10 日启动公开测试。测试版接口、功能覆盖范围、定价和迁移指南都可能发生变化；在确定生产架构之前，请核实当前文档。","时间敏感的架构决策","warning",{},{"id":244,"data":245,"type":234,"tunes":249},"model-note",{"body":246,"title":247,"variant":248},"下文的运行时所有权测试是一种实用的架构模型，而非 OpenAI 的产品分类法。它询问的是：智能体运行时的哪些部分必须由你的团队控制，哪些部分你希望由 OpenAI 来运营。","此处使用的决策模型","note",{},{"id":251,"data":252,"type":42,"tunes":254},"h-boundary",{"text":253,"level":219},"架构已经改变：选择的是运行时边界，而不是库",{},{"id":256,"data":257,"type":226,"tunes":259},"p-boundary-1",{"text":258},"重要的决策不再只是“我应该安装哪个 SDK？”而是由谁拥有执行框架、智能体循环、持久化会话状态、上下文压缩、恢复、执行环境和应用程序生命周期。",{},{"id":261,"data":262,"type":226,"tunes":264},"p-boundary-2",{"text":263},"OpenAI 当前的 Agents 概览明确说明了这一边界。Agents API 运行托管的 Codex 执行框架，并管理编排以及持久化会话状态。Responses API 提供模型响应和托管能力，而由你的应用程序拥有周围的智能体循环。Agents SDK 在你的应用程序中运行循环，如今已功能完备，而不再是重大新智能体功能的前进路径。",{},{"id":266,"data":267,"type":42,"tunes":269},"h-short",{"text":268,"level":219},"简要对比",{},{"id":271,"data":272,"type":304,"tunes":305},"short-table",{"content":273,"stretched":43,"withHeadings":14},[274,280,286,292,298],[275,276,277,278,279],"选项","2026 年的最佳适用场景","谁拥有智能体循环？","会话 \u002F 上下文责任","战略状态",[281,282,283,284,285],"Agents API","新的 OpenAI 原生持久化智能体","OpenAI 托管的 Codex 执行框架","OpenAI 管理会话、编排、压缩和恢复","新智能体应用的推荐起点；公开测试",[287,288,289,290,291],"Responses API","直接模型集成和自定义智能体运行时","你的应用程序","由你选择响应链式调用、Conversations、存储和循环逻辑","核心 API 原语；对于新项目，推荐优先于 Chat Completions",[293,294,295,296,297],"Agents SDK","现有 SDK 应用程序或临时能力缺口","你的应用程序通过 SDK 运行器","你的应用程序负责部署、存储和运行时行为","功能完备；维护和兼容性继续，不计划推出重大新功能",[299,300,301,302,303],"Codex SDK","在你运营的基础设施中使用 Codex 执行框架","你环境中的 Codex 执行框架","你负责执行框架的托管和生命周期","当你想要执行框架但不想要托管的 Agents API 运行时时，这是一个独立选项","table",{},{"id":307,"data":308,"type":42,"tunes":310},"h-agents-api",{"text":309,"level":219},"1. Agents API：托管执行框架，持久化云端智能体",{},{"id":312,"data":313,"type":226,"tunes":315},"p-api-1",{"text":314},"Agents API 通过 OpenAI 托管的服务公开 Codex 执行框架。OpenAI 管理会话、编排、上下文压缩和恢复。你的应用程序仍然提供工具并选择执行环境。",{},{"id":317,"data":318,"type":226,"tunes":320},"p-api-2",{"text":319},"最后这一点区别很重要。“托管智能体”并不一定意味着“所有计算都在 OpenAI 内部运行”。Agents API 架构支持无环境、OpenAI 托管环境，或连接到托管执行框架的自托管环境。使用自托管环境时，你的应用程序负责配置、重新连接、关闭和持久化文件，而执行框架仍由托管方管理。",{},{"id":322,"data":323,"type":226,"tunes":325},"p-api-3",{"text":324},"因此，Agents API 是一种运行时服务，而不仅仅是一种请求格式。会话可以持久化、流式传输进度、接收额外任务、使用工具、处理文件，并在长时间运行的工作中恢复。",{},{"id":327,"data":328,"type":42,"tunes":330},"h-api-gain",{"text":329,"level":218},"使用 Agents API 你能获得什么",{},{"id":332,"data":333,"type":343,"tunes":344},"api-gains",{"meta":334,"items":335,"style":342},{},[336,337,338,339,340,341],"托管的 Codex 执行框架，无需自行构建和运营主要的智能体循环。","持久化会话，适用于跨多轮和长时间运行的任务。","托管的编排、上下文压缩和恢复。","根据工作负载需求，可选择 OpenAI 托管或自托管的执行环境。","用于进度和生命周期事件的流式传输和 Webhook。","OpenAI 明确推荐用于新智能体应用的平台方向。","unordered","list",{},{"id":346,"data":347,"type":42,"tunes":349},"h-api-own",{"text":348,"level":218},"你仍然需要负责的部分",{},{"id":351,"data":352,"type":343,"tunes":360},"api-own",{"meta":353,"items":354,"style":342},{},[355,356,357,358,359],"你的产品和应用服务器。","函数工具实现和业务逻辑。","围绕你自己系统的授权和策略决策。","当你选择自托管计算时的执行环境生命周期。","评估、验收标准、领域特定的防护措施，以及决定智能体被允许做什么。",{},{"id":362,"data":363,"type":42,"tunes":365},"h-responses",{"text":364,"level":219},"2. Responses API：掌控循环，使用平台原语",{},{"id":367,"data":368,"type":226,"tunes":370},"p-resp-1",{"text":369},"当你希望使用 OpenAI 模型和工具能力，而不想将整个智能体运行时委托出去时，Responses API 是更低层级的选择。OpenAI 将 Responses 描述为新项目推荐的 API 原语，以及 Chat Completions 的演进，具备内置工具、多轮状态选项、多模态输入和智能体工具使用能力。",{},{"id":372,"data":373,"type":226,"tunes":375},"p-resp-2",{"text":374},"一个 Responses 请求本身可以调用工具，但当你围绕它构建智能体时，你的应用程序仍需负责更大的工作流。这意味着你的代码决定如何持久化应用状态、何时继续、如何恢复、如何协调专家、如何压缩长历史记录，以及如何表示可恢复的工作。",{},{"id":377,"data":378,"type":226,"tunes":380},"p-resp-3",{"text":379},"这本身并不逊色。当智能体行为必须深度嵌入现有应用逻辑、当你需要自定义状态模型，或当托管式运行框架会隐藏你真正需要的控制权时，这就是正确的边界。",{},{"id":382,"data":383,"type":42,"tunes":385},"h-sdk",{"text":384,"level":219},"3. Agents SDK：仍受支持，但不再是默认的前进路径",{},{"id":387,"data":388,"type":226,"tunes":390},"p-sdk-1",{"text":389},"Agents SDK 仍然是一个开源框架，用于在你的应用程序中运行智能体工作流。它在 TypeScript 和 Python 中提供智能体定义、工具、交接、护栏、追踪、会话和运行器循环。",{},{"id":392,"data":393,"type":226,"tunes":395},"p-sdk-2",{"text":394},"但其战略地位已经改变。OpenAI 现在将 Agents SDK 标记为功能完备：维护、安全修复、关键错误修复和兼容性工作仍在继续，但不计划推出重大新功能。OpenAI 推荐新应用程序使用 Agents API。",{},{"id":397,"data":398,"type":226,"tunes":400},"p-sdk-3",{"text":399},"这并不意味着现有的 SDK 应用程序应立即重写。它意味着架构不应再假设 SDK 是下一个重大智能体运行时能力的落点。",{},{"id":402,"data":403,"type":234,"tunes":406},"sdk-warning",{"body":404,"title":405,"variant":241},"OpenAI 并未将 Agents SDK 描述为不受支持。现有应用程序可以继续使用它，维护、安全修复、关键错误修复和兼容性工作仍在继续。重要的变化是，不计划推出重大新功能，并且新应用程序被引导至 Agents API。","不要将“功能完备”翻译为“已弃用”",{},{"id":408,"data":409,"type":42,"tunes":411},"h-codex-sdk",{"text":410,"level":219},"Codex SDK 的定位",{},{"id":413,"data":414,"type":226,"tunes":416},"p-codex-1",{"text":415},"当前的选择并不是简单的三岔路口。OpenAI 的运行时概览将 Codex SDK 列为在你运营的基础设施中运行 Codex 运行框架的选项。这在架构上既不同于托管的 Agents API，也不同于 Agents SDK。",{},{"id":418,"data":419,"type":226,"tunes":421},"p-codex-2",{"text":420},"如果你的真实需求是“我想要 Codex 运行框架，但我需要自己运营它”，那么 Codex SDK 就是值得评估的界面。如果你的需求是“我想掌控围绕模型调用的循环”，请评估 Responses。如果你的需求是“我已经有一个可用的 Agents SDK 应用程序”，现有 SDK 可以继续有效，同时你可以根据其维护状态进行规划。",{},{"id":423,"data":424,"type":42,"tunes":426},"h-test",{"text":425,"level":219},"运行时所有权测试",{},{"id":428,"data":429,"type":226,"tunes":431},"p-test-intro",{"text":430},"有用的架构决策始于确定你的团队必须拥有什么。将每项需求评分为必须控制、倾向于控制或倾向于托管。",{},{"id":433,"data":434,"type":475,"tunes":476},"ownership-test",{"rows":435,"title":425,"layout":304,"columns":465},[436,441,445,449,453,457,461],{"id":437,"label":438,"values":439},"loop","智能体循环",[440,440,440],"",{"id":442,"label":443,"values":444},"session","持久会话",[440,440,440],{"id":446,"label":447,"values":448},"harness","运行框架运行时",[440,440,440],{"id":450,"label":451,"values":452},"environment","执行环境",[440,440,440],{"id":454,"label":455,"values":456},"orchestration","编排语义",[440,440,440],{"id":458,"label":459,"values":460},"portability","提供商 \u002F 传输灵活性",[440,440,440],{"id":462,"label":463,"values":464},"ops","运维负担",[440,440,440],[466,469,472],{"id":467,"label":468},"decision","决策",{"id":470,"label":471},"managed","如果你倾向于托管",{"id":473,"label":474},"control","如果你需要控制","comparison",{},{"id":478,"data":479,"type":42,"tunes":481},"h-tree",{"text":480,"level":219},"新系统的决策树",{},{"id":483,"data":484,"type":509,"tunes":510},"decision-flow",{"steps":485,"title":507,"orientation":508},[486,489,492,495,498,501,504],{"label":487,"description":488},"1. 这是一个新的智能体应用程序吗？","如果否，不要仅仅因为存在更新的界面就迁移。首先评估当前应用程序的实际约束。",{"label":490,"description":491},"2. 你想要一个托管的长运行智能体运行框架吗？","如果是，从 Agents API 开始，这是 OpenAI 推荐的新智能体应用程序路径。",{"label":493,"description":494},"3. 你需要 Codex 运行框架但必须自己运营它吗？","评估 Codex SDK，而不是在 Agents SDK 之上重建运行框架行为。",{"label":496,"description":497},"4. 你需要拥有智能体循环和状态模型吗？","使用 Responses API 作为更低层级的平台原语，并围绕它构建循环。",{"label":499,"description":500},"5. 你已经在使用 Agents SDK 了吗？","如果它满足需求，请继续使用；不计划推出重大新功能，因此将未来平台采用视为明确的路线图决策。",{"label":502,"description":503},"6. Agents API 是否缺少所需能力？","OpenAI 明确允许 Agents SDK 作为需要不受支持能力的新应用程序的短期选项。",{"label":505,"description":506},"7. 通过生产形态的验证性开发进行验证","在锁定架构之前，测试工具、环境、审批、延迟、可观测性、故障恢复和生命周期。","根据控制边界选择运行时","auto","processFlow",{},{"id":512,"data":513,"type":42,"tunes":515},"h-not-drive",{"text":514,"level":219},"什么不应驱动决策",{},{"id":517,"data":518,"type":304,"tunes":544},"weak-rules",{"content":519,"stretched":43,"withHeadings":14},[520,524,528,532,536,540],[521,522,523],"薄弱的决策规则","为何失败","更好的问题",[525,526,527],"“最新的 API 一定最好。”","较新的方案可能在战略上更受青睐，但仍可能缺少你所需的能力。","哪些运行时职责应由平台管理，哪些应由应用自行拥有？",[529,530,531],"“我们已经熟悉这个 SDK。”","团队熟悉度可能维护着一个路线图已经改变的架构。","在下一个产品周期内，留下与迁移的成本分别是多少？",[533,534,535],"“托管就意味着没有基础设施。”","Agents API 仍可使用自托管环境，你的应用仍拥有产品逻辑。","实际被委托出去的是哪一层基础设施？",[537,538,539],"“Responses 只适用于简单调用。”","Responses 提供内置工具和有状态原语；它可以作为自定义智能体循环的基础。","我们需要平台拥有整个执行框架，还是只需要模型\u002F工具原语？",[541,542,543],"“功能完备就意味着我们必须立即迁移。”","SDK 仍会为现有应用继续维护。","继续留下会阻碍哪些具体的未来需求？",{},{"id":546,"data":547,"type":42,"tunes":549},"h-migration",{"text":548,"level":219},"迁移是架构变更，而不是重命名导入",{},{"id":551,"data":552,"type":226,"tunes":554},"p-migration-1",{"text":553},"从 Agents SDK 迁移到 Agents API 会改变所有权归属。在 SDK 中，循环运行在你的应用内。在 Agents API 中，OpenAI 运行执行框架和会话，而你的应用通过任务、事件、工具和环境边界进行集成。",{},{"id":556,"data":557,"type":226,"tunes":559},"p-migration-2",{"text":558},"因此，真正的迁移计划需要梳理会话状态、自定义编排、交接、工具执行、审批、存储、追踪、重试、故障恢复、环境生命周期，以及任何特定于提供商的抽象。代码量可能下降，但运维假设会发生变化。",{},{"id":561,"data":562,"type":42,"tunes":564},"h-migration-inventory",{"text":563,"level":218},"迁移清单",{},{"id":566,"data":567,"type":343,"tunes":580},"migration-list",{"meta":568,"items":569,"style":342},{},[570,571,572,573,574,575,576,577,578,579],"智能体定义和指令所有权。","工具定义以及每个工具的执行位置。","交接、管理者\u002F专家模式以及子智能体行为。","会话标识符、对话状态、可恢复性和历史保留。","人工审批和中断语义。","自定义上下文裁剪或压缩逻辑。","追踪、评估、可观测性和生产调试。","自托管文件、容器、私有网络访问或其他执行依赖。","提供商抽象或非 OpenAI 模型依赖。","重试、超时、幂等性、恢复和生命周期假设。",{},{"id":582,"data":583,"type":42,"tunes":585},"h-beta",{"text":584,"level":219},"公开测试版改变了风险模型",{},{"id":587,"data":588,"type":226,"tunes":590},"p-beta-1",{"text":589},"Agents API 是新智能体应用的推荐方向，但它也处于公开测试阶段。这些事实并不矛盾。战略方向回答的是“平台将走向何方？”测试版状态回答的是“我应该为多少接口和运维变化预留预算？”",{},{"id":592,"data":593,"type":226,"tunes":595},"p-beta-2",{"text":594},"对于生产系统，应将集成隔离在应用边界之后。尽可能将领域状态、权限、审计数据和业务规则保留在供应商特定的会话对象之外。这样更容易吸收 API 演进，而不会让智能体运行时成为整个产品的真相来源。",{},{"id":597,"data":598,"type":42,"tunes":600},"h-default",{"text":599,"level":219},"一种实用的默认架构",{},{"id":602,"data":603,"type":226,"tunes":605},"p-default-1",{"text":604},"对于许多新的 OpenAI 原生应用，2026 年一个合理的默认方案是：使用 Agents API 提供托管执行框架和持久会话，应用拥有领域服务和授权，业务操作使用显式函数工具，并根据数据和计算需求选择 OpenAI 托管或自托管执行。",{},{"id":607,"data":608,"type":226,"tunes":610},"p-default-2",{"text":609},"这样可以让智能体运行时保持强大，同时不让它成为业务真相的所有者。应用仍然决定用户可以做什么、哪些数据具有权威性、哪些操作需要审批，以及如何验证结果。",{},{"id":612,"data":613,"type":618,"tunes":619},"ref-reliability",{"url":614,"title":615,"excerpt":616,"ctaLabel":617},"https:\u002F\u002Fstajic.de\u002Fzh\u002Fblog\u002Fai-agent-reliability-why-the-final-answer-is-not-enough","AI 智能体可靠性：为什么最终答案还不够","托管执行框架可以减轻运维负担，但可靠性仍然需要可追踪的执行、明确的验收标准，以及超越最终答案的评估。","阅读可靠性文章","referralArticle",{},{"id":621,"data":622,"type":42,"tunes":624},"h-change",{"text":623,"level":219},"什么会改变这个答案？",{},{"id":626,"data":627,"type":226,"tunes":629},"p-change-1",{"text":628},"如果 Agents API 增加或移除能力、以不同的契约退出测试版、改变环境或定价边界，或引入减少所有权差异的迁移工具，那么建议就会改变。如果你的应用依赖提供商可移植性、自定义编排语义、仅本地执行，或托管执行框架无法支持的能力，建议也会改变。",{},{"id":631,"data":632,"type":226,"tunes":634},"p-change-2",{"text":633},"对于现有的 Agents SDK 应用，答案还会随迁移成本而变化。如果系统稳定、经过充分评估，并且没有被 SDK 的功能完备状态所阻碍，那么立即迁移可能带来的风险大于价值。如果产品路线图依赖于仅在 Agents API 中提供的能力，那么延迟迁移可能会造成另一种债务。",{},{"id":636,"data":637,"type":42,"tunes":639},"h-limitations",{"text":638,"level":219},"局限性",{},{"id":641,"data":642,"type":226,"tunes":644},"p-limit-1",{"text":643},"本比较聚焦于运行时所有权和 OpenAI 所声明的平台方向。它不会针对特定工作负载对延迟、质量或总成本进行基准测试。这些属性取决于模型选择、工具使用、环境、任务长度、缓存、沙箱使用和应用架构。",{},{"id":646,"data":647,"type":226,"tunes":649},"p-limit-2",{"text":648},"Agents API 也足够新，生产经验仍在积累中。因此，设计应通过具有代表性的工作负载进行验证，而不是仅根据产品定位来选择。",{},{"id":651,"data":652,"type":42,"tunes":654},"h-conclusion",{"text":653,"level":219},"结论",{},{"id":656,"data":657,"type":226,"tunes":659},"p-conclusion-1",{"text":658},"2026 年 OpenAI 智能体决策的核心在于运行时所有权。Agents API 意味着 OpenAI 拥有更多 harness 和持久会话机制。Responses 意味着你的应用程序拥有围绕平台原语的循环。Agents SDK 对现有系统仍然有效，但已不再是重大新智能体运行时功能的默认目标。",{},{"id":661,"data":662,"type":226,"tunes":664},"p-conclusion-2",{"text":663},"对于新应用程序，除非有实际需求推动你向技术栈更低层移动，否则应遵循平台方向。从 Agents API 开始，当你需要拥有循环时转向 Responses，当你需要将 harness 部署在自己的基础设施中时评估 Codex SDK，并在现有投资或临时能力缺口使其合理时保留 Agents SDK。",{},{"id":666,"data":667,"type":42,"tunes":669},"h-faq",{"text":668,"level":219},"常见问题",{},{"id":671,"data":672,"type":671,"tunes":699},"faq",{"items":673,"title":698},[674,678,682,686,690,694],{"id":675,"answer":676,"question":677},"faq1","OpenAI 目前推荐新智能体应用程序使用 Agents API。Agents SDK 功能已完备，并继续为现有应用程序提供支持，维护、安全、关键错误修复和兼容性工作仍在继续。","对于新项目，我应该使用 OpenAI Agents API 还是 Agents SDK？",{"id":679,"answer":680,"question":681},"faq2","OpenAI 将其描述为功能已完备，而非不受支持。不计划推出重大新功能，但维护、安全修复、关键错误修复和兼容性工作仍在继续。","Agents SDK 是否已弃用？",{"id":683,"answer":684,"question":685},"faq3","当你的应用程序应拥有智能体循环、状态策略、编排和延续逻辑，同时仍使用 OpenAI 模型和平台工具时，请使用 Responses。","何时应使用 Responses API 而非 Agents API？",{"id":687,"answer":688,"question":689},"faq4","不需要。Agents API 可以使用无执行环境、OpenAI 托管环境，或连接到托管 harness 的自托管环境。","Agents API 是否需要 OpenAI 托管的计算资源？",{"id":691,"answer":692,"question":693},"faq5","OpenAI 将 Codex SDK 定位为在你运营的基础设施中运行 Codex harness。当你想要 harness 但不想要托管的 Agents API 运行时，它是相关选项。","Codex SDK 适用于什么场景？",{"id":695,"answer":696,"question":697},"faq6","不应自动迁移。评估当前系统是否因 SDK 的功能已完备状态而受阻，所需能力是否存在于 Agents API 中，以及迁移收益是否超过运营和架构变更成本。","现有的 Agents SDK 应用是否应立即迁移？","2026 年 OpenAI 智能体运行时选择",{},{"id":701,"data":702,"type":42,"tunes":704},"h-glossary",{"text":703,"level":219},"术语表",{},{"id":706,"data":707,"type":706,"tunes":729},"glossary",{"title":708,"entries":709},"关键运行时术语",[710,713,716,719,722,725],{"term":711,"anchor":446,"definition":712},"Harness","协调模型调用、工具、上下文、会话和智能体执行的运行时循环及支持机制。",{"term":281,"anchor":714,"definition":715},"agents-api","OpenAI 的托管 API，用于使用托管 Codex harness 的持久云智能体。",{"term":287,"anchor":717,"definition":718},"responses-api","OpenAI 的较低层 API 原语，用于模型响应、托管工具和有状态交互，应用程序可围绕其构建自己的智能体循环。",{"term":293,"anchor":720,"definition":721},"agents-sdk","OpenAI 的开源框架，用于在应用程序代码中运行智能体工作流；截至 2026 年 9 月功能已完备。",{"term":299,"anchor":723,"definition":724},"codex-sdk","OpenAI 列出的运行时选项，用于在你运营的基础设施中运行 Codex harness。",{"term":726,"anchor":727,"definition":728},"运行时所有权","runtime-ownership","描述智能体循环、会话状态、执行环境和生命周期中哪些部分由平台运营、哪些由应用程序运营的架构边界。",{},{"id":731,"data":732,"type":42,"tunes":734},"h-sources",{"text":733,"level":219},"主要来源与延伸阅读",{},{"id":736,"data":737,"type":743,"tunes":744},"src-launch",{"link":738,"meta":739},"https:\u002F\u002Fopenai.com\u002Findex\u002Fintroducing-the-agents-api\u002F",{"image":740,"title":741,"description":742},{"url":440},"OpenAI — 推出 Agents API","2026 年 9 月 10 日 Agents API 的发布公告，介绍了托管的 Codex harness 和公开测试版。","linkTool",{},{"id":746,"data":747,"type":743,"tunes":753},"src-agents-overview",{"link":748,"meta":749},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents",{"image":750,"title":751,"description":752},{"url":440},"OpenAI — Agents 运行时概述","当前对 Agents API、Codex SDK 和 Responses API 的比较，包括 Agents SDK 的支持状态。",{},{"id":755,"data":756,"type":743,"tunes":762},"src-agents-api",{"link":757,"meta":758},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Foverview",{"image":759,"title":760,"description":761},{"url":440},"OpenAI — Agents API 概述","关于持久云智能体、会话、编排、上下文压缩、恢复和环境选择的文档。",{},{"id":764,"data":765,"type":743,"tunes":771},"src-architecture",{"link":766,"meta":767},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Farchitecture",{"image":768,"title":769,"description":770},{"url":440},"OpenAI — Agents API 架构","托管 harness、应用服务器以及无\u002FOpenAI 托管\u002F自托管执行环境之间的架构边界。",{},{"id":773,"data":774,"type":743,"tunes":780},"src-sdk",{"link":775,"meta":776},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\u002Fsdk",{"image":777,"title":778,"description":779},{"url":440},"OpenAI — Agents SDK","当前 Agents SDK 支持通知以及对应用程序拥有的智能体循环的说明。",{},{"id":782,"data":783,"type":743,"tunes":789},"src-responses",{"link":784,"meta":785},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fmigrate-to-responses",{"image":786,"title":787,"description":788},{"url":440},"OpenAI — 迁移到 Responses API","当前 Responses API 的定位、内置工具、有状态上下文和智能体原语。",{},{"id":791,"data":792,"type":220,"tunes":793},"toc",{"title":217,"maxLevel":218,"minLevel":219},{},"2.31","OpenAI 的智能体技术栈在 2026 年 9 月发生了变化。本架构指南按运行时归属将 Agents API、Agents SDK、Responses API 和 Codex SDK 区分开来——以便团队能够选择正确的控制边界，而不是比较产品名称。","\u002Fuploads\u002F2026\u002F09\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026-1790351846714-zi7lus.webp","openai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026-1790351846714-zi7lus","PUBLISHED","2026-09-25T11:56:00.000Z","2026-09-25T15:56:16.274Z","2026-09-25T19:58:41.459Z",{"en":803,"de":804,"sr":805,"es":806,"fr":807,"it":808,"ru":809,"zh":810},"\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fde\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fsr\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fes\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Ffr\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fit\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fru\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","\u002Fzh\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026",[812,815,819],{"id":101,"name":813,"slug":814},"Overview","overview-digital-platform",{"id":816,"name":817,"slug":818},48,"能力","capabilities",{"id":820,"name":821,"slug":822},46,"概览","overview",{"id":824,"login":825,"email":826,"displayName":827},"20","rooth8233","aleksandar@stajic.de","Aleksandar Stajić",[829,1299],{"lang":830,"title":831,"content":832,"contentJson":833,"excerpt":1298},"en","OpenAI Agents API vs Agents SDK vs Responses API: What Should You Build On in 2026?","{\"time\":1790352010644,\"blocks\":[{\"id\":\"75qcDhG72q\",\"type\":\"tableOfContents\",\"data\":{\"title\":\"Contents\",\"minLevel\":2,\"maxLevel\":3},\"tunes\":{}},{\"id\":\"intro\",\"type\":\"paragraph\",\"data\":{\"text\":\"OpenAI's agent stack changed materially in September 2026. The new Agents API introduced a managed Codex harness for durable cloud agents, while the older Agents SDK moved into a feature-complete maintenance phase. The Responses API remains the lower-level surface for applications that want direct model calls or to own the agent loop themselves. These are not three interchangeable wrappers around the same thing: they place the runtime boundary in different places.\"},\"tunes\":{}},{\"id\":\"direct\",\"type\":\"callout\",\"data\":{\"variant\":\"info\",\"title\":\"Direct answer\",\"body\":\"\u003Cstrong>For a new OpenAI-native agent application in September 2026, start with the Agents API unless you have a concrete reason to own more of the runtime.\u003C\u002Fstrong> Use the Responses API when your application should own the loop, state strategy, and orchestration. Keep the Agents SDK for existing applications or short-term gaps where the Agents API does not yet support a required capability. If you want the Codex harness itself but need to run it in infrastructure you operate, evaluate the Codex SDK rather than forcing the Agents SDK into that role.\"},\"tunes\":{}},{\"id\":\"date-warning\",\"type\":\"callout\",\"data\":{\"variant\":\"warning\",\"title\":\"Time-sensitive architecture decision\",\"body\":\"This article reflects OpenAI's platform state on \u003Cstrong>25 September 2026\u003C\u002Fstrong>. The Agents API launched in public beta on 10 September 2026. Beta surfaces, feature coverage, pricing, and migration guidance can change; verify current documentation before committing a production architecture.\"},\"tunes\":{}},{\"id\":\"model-note\",\"type\":\"callout\",\"data\":{\"variant\":\"note\",\"title\":\"The decision model used here\",\"body\":\"The Runtime Ownership Test below is a practical architecture model, not an OpenAI product taxonomy. It asks which parts of the agent runtime your team must control versus which parts you want OpenAI to operate.\"},\"tunes\":{}},{\"id\":\"h-boundary\",\"type\":\"header\",\"data\":{\"text\":\"The architecture changed: choose a runtime boundary, not a library\",\"level\":2},\"tunes\":{}},{\"id\":\"p-boundary-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The important decision is no longer simply “Which SDK should I install?” It is who owns the harness, the agent loop, durable session state, context compaction, recovery, execution environment, and application lifecycle.\"},\"tunes\":{}},{\"id\":\"p-boundary-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"OpenAI's current Agents overview makes this boundary explicit. The Agents API runs a hosted Codex harness and manages orchestration plus durable session state. The Responses API provides model responses and hosted capabilities while your application owns the surrounding agent loop. The Agents SDK runs the loop in your application and is now feature complete rather than the forward path for major new agent features.\"},\"tunes\":{}},{\"id\":\"h-short\",\"type\":\"header\",\"data\":{\"text\":\"The short comparison\",\"level\":2},\"tunes\":{}},{\"id\":\"short-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Option\",\"Best fit in 2026\",\"Who owns the agent loop?\",\"Session \u002F context responsibility\",\"Strategic status\"],[\"Agents API\",\"New OpenAI-native durable agents\",\"OpenAI-managed Codex harness\",\"OpenAI manages sessions, orchestration, compaction and recovery\",\"Recommended starting point for new agent apps; public beta\"],[\"Responses API\",\"Direct model integrations and custom agent runtimes\",\"Your application\",\"You choose response chaining, Conversations, storage and loop logic\",\"Core API primitive; recommended over Chat Completions for new projects\"],[\"Agents SDK\",\"Existing SDK applications or temporary capability gaps\",\"Your application via the SDK runner\",\"Your application operates deployment, storage and runtime behaviour\",\"Feature complete; maintenance and compatibility continue, major new features not planned\"],[\"Codex SDK\",\"Codex harness in infrastructure you operate\",\"Codex harness in your environment\",\"You operate harness hosting and lifecycle\",\"Separate option when you want the harness without the hosted Agents API runtime\"]]},\"tunes\":{}},{\"id\":\"h-agents-api\",\"type\":\"header\",\"data\":{\"text\":\"1. Agents API: managed harness, durable cloud agent\",\"level\":2},\"tunes\":{}},{\"id\":\"p-api-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Agents API exposes the Codex harness through an OpenAI-managed service. OpenAI manages sessions, orchestration, context compaction, and recovery. Your application still provides tools and chooses the execution environment.\"},\"tunes\":{}},{\"id\":\"p-api-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"That last distinction matters. “Managed agent” does not necessarily mean “all compute runs inside OpenAI.” The Agents API architecture supports no environment, an OpenAI-hosted environment, or a self-hosted environment connected to the hosted harness. With a self-hosted environment, your application owns provisioning, reconnection, shutdown, and persistent files while the harness remains managed.\"},\"tunes\":{}},{\"id\":\"p-api-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Agents API is therefore a runtime service, not merely a request format. Sessions can persist, stream progress, receive additional tasks, use tools, work with files, and recover across long-running work.\"},\"tunes\":{}},{\"id\":\"h-api-gain\",\"type\":\"header\",\"data\":{\"text\":\"What you gain with the Agents API\",\"level\":3},\"tunes\":{}},{\"id\":\"api-gains\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"A managed Codex harness instead of building and operating the main agent loop yourself.\",\"Durable sessions for work that spans turns and long-running tasks.\",\"Managed orchestration, context compaction, and recovery.\",\"OpenAI-hosted or self-hosted execution environments, depending on workload requirements.\",\"Streaming and webhooks for progress and lifecycle events.\",\"A platform direction that OpenAI explicitly recommends for new agent applications.\"]},\"tunes\":{}},{\"id\":\"h-api-own\",\"type\":\"header\",\"data\":{\"text\":\"What you still own\",\"level\":3},\"tunes\":{}},{\"id\":\"api-own\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Your product and application server.\",\"Function-tool implementations and business logic.\",\"Authorization and policy decisions around your own systems.\",\"The execution environment lifecycle when you choose self-hosted compute.\",\"Evaluation, acceptance criteria, domain-specific guardrails, and the decision of what the agent is allowed to do.\"]},\"tunes\":{}},{\"id\":\"h-responses\",\"type\":\"header\",\"data\":{\"text\":\"2. Responses API: own the loop, use the platform primitives\",\"level\":2},\"tunes\":{}},{\"id\":\"p-resp-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Responses API is the lower-level choice when you want OpenAI model and tool capabilities without delegating the overall agent runtime. OpenAI describes Responses as the recommended API primitive for new projects and as an evolution of Chat Completions with built-in tools, multi-turn state options, multimodal input, and agentic tool use.\"},\"tunes\":{}},{\"id\":\"p-resp-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"A Responses request can itself invoke tools, but your application remains responsible for the larger workflow when you build an agent around it. That means your code decides how to persist application state, when to continue, how to recover, how to coordinate specialists, how to compact long histories, and how to represent resumable work.\"},\"tunes\":{}},{\"id\":\"p-resp-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"This is not inherently inferior. It is the right boundary when agent behaviour must be deeply embedded in existing application logic, when you need a custom state model, or when a managed harness would hide control you actually need.\"},\"tunes\":{}},{\"id\":\"h-sdk\",\"type\":\"header\",\"data\":{\"text\":\"3. Agents SDK: supported, but no longer the default forward path\",\"level\":2},\"tunes\":{}},{\"id\":\"p-sdk-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Agents SDK remains an open-source framework for running agent workflows in your application. It provides agent definitions, tools, handoffs, guardrails, tracing, sessions, and the runner loop in TypeScript and Python.\"},\"tunes\":{}},{\"id\":\"p-sdk-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"But its strategic status changed. OpenAI now labels the Agents SDK feature complete: maintenance, security fixes, critical bug fixes, and compatibility work continue, but major new features are not planned. OpenAI recommends the Agents API for new applications.\"},\"tunes\":{}},{\"id\":\"p-sdk-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"That does not mean an existing SDK application should be rewritten immediately. It means the architecture should stop assuming that the SDK is where the next major agent-runtime capabilities will land.\"},\"tunes\":{}},{\"id\":\"sdk-warning\",\"type\":\"callout\",\"data\":{\"variant\":\"warning\",\"title\":\"Do not translate “feature complete” into “deprecated”\",\"body\":\"OpenAI has not described the Agents SDK as unsupported. Existing applications can continue to use it, and maintenance, security, critical bug fixes, and compatibility work continue. The important change is that major new features are not planned and new applications are directed to the Agents API.\"},\"tunes\":{}},{\"id\":\"h-codex-sdk\",\"type\":\"header\",\"data\":{\"text\":\"Where the Codex SDK fits\",\"level\":2},\"tunes\":{}},{\"id\":\"p-codex-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The current choice is not a simple three-way fork. OpenAI's runtime overview includes the Codex SDK as the option for running the Codex harness in infrastructure you operate. That is architecturally different from both the hosted Agents API and the Agents SDK.\"},\"tunes\":{}},{\"id\":\"p-codex-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"If your real requirement is “I want the Codex harness, but I need to operate it myself,” the Codex SDK is the surface to evaluate. If your requirement is “I want to own the loop around model calls,” evaluate Responses. If your requirement is “I already have a working Agents SDK application,” the existing SDK can remain valid while you plan against its maintenance status.\"},\"tunes\":{}},{\"id\":\"h-test\",\"type\":\"header\",\"data\":{\"text\":\"The Runtime Ownership Test\",\"level\":2},\"tunes\":{}},{\"id\":\"p-test-intro\",\"type\":\"paragraph\",\"data\":{\"text\":\"A useful architecture decision starts by deciding what your team must own. Score each requirement as either must control, prefer to control, or prefer managed.\"},\"tunes\":{}},{\"id\":\"ownership-test\",\"type\":\"comparison\",\"data\":{\"title\":\"Runtime Ownership Test\",\"layout\":\"table\",\"columns\":[{\"id\":\"decision\",\"label\":\"Decision\"},{\"id\":\"managed\",\"label\":\"If you prefer managed\"},{\"id\":\"control\",\"label\":\"If you require control\"}],\"rows\":[{\"id\":\"loop\",\"label\":\"Agent loop\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"session\",\"label\":\"Durable sessions\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"harness\",\"label\":\"Harness runtime\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"environment\",\"label\":\"Execution environment\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"orchestration\",\"label\":\"Orchestration semantics\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"portability\",\"label\":\"Provider \u002F transport flexibility\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"ops\",\"label\":\"Operational burden\",\"values\":[\"\",\"\",\"\"]}]},\"tunes\":{}},{\"id\":\"h-tree\",\"type\":\"header\",\"data\":{\"text\":\"A decision tree for new systems\",\"level\":2},\"tunes\":{}},{\"id\":\"decision-flow\",\"type\":\"processFlow\",\"data\":{\"title\":\"Choose the runtime by the control boundary\",\"orientation\":\"auto\",\"steps\":[{\"label\":\"1. Is this a new agent application?\",\"description\":\"If no, do not migrate solely because a newer surface exists. Evaluate the current application's actual constraints first.\"},{\"label\":\"2. Do you want a managed long-running agent harness?\",\"description\":\"If yes, start with the Agents API as OpenAI's recommended path for new agent applications.\"},{\"label\":\"3. Do you need the Codex harness but must operate it yourself?\",\"description\":\"Evaluate the Codex SDK rather than rebuilding harness behaviour on top of the Agents SDK.\"},{\"label\":\"4. Do you need to own the agent loop and state model?\",\"description\":\"Use the Responses API as the lower-level platform primitive and build the loop around it.\"},{\"label\":\"5. Are you already on the Agents SDK?\",\"description\":\"Continue if it meets requirements; major new features are not planned, so treat future platform adoption as an explicit roadmap decision.\"},{\"label\":\"6. Is a required capability missing from the Agents API?\",\"description\":\"OpenAI explicitly allows the Agents SDK as a short-term option for new applications that need unsupported capabilities.\"},{\"label\":\"7. Validate with a production-shaped spike\",\"description\":\"Test tools, environment, approvals, latency, observability, failure recovery, and lifecycle before locking the architecture.\"}]},\"tunes\":{}},{\"id\":\"h-not-drive\",\"type\":\"header\",\"data\":{\"text\":\"What should not drive the decision\",\"level\":2},\"tunes\":{}},{\"id\":\"weak-rules\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Weak decision rule\",\"Why it fails\",\"Better question\"],[\"“The newest API must be best.”\",\"Newer can be strategically preferred while still lacking a capability you need.\",\"Which runtime responsibilities should be managed versus application-owned?\"],[\"“We already know the SDK.”\",\"Team familiarity can preserve an architecture whose roadmap has changed.\",\"What is the cost of staying versus moving over the next product cycle?\"],[\"“Managed means no infrastructure.”\",\"Agents API can still use self-hosted environments and your application still owns product logic.\",\"Which infrastructure layer is actually being delegated?\"],[\"“Responses is only for simple calls.”\",\"Responses provides built-in tools and stateful primitives; it can be the foundation of custom agent loops.\",\"Do we need the platform to own the harness, or only model\u002Ftool primitives?\"],[\"“Feature complete means we must migrate now.”\",\"The SDK remains maintained for existing applications.\",\"What concrete future requirement is blocked by staying?\"]]},\"tunes\":{}},{\"id\":\"h-migration\",\"type\":\"header\",\"data\":{\"text\":\"Migration is an architectural change, not an import rename\",\"level\":2},\"tunes\":{}},{\"id\":\"p-migration-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Moving from the Agents SDK to the Agents API changes ownership. In the SDK, the loop runs in your application. In the Agents API, OpenAI runs the harness and session while your application integrates through tasks, events, tools, and environment boundaries.\"},\"tunes\":{}},{\"id\":\"p-migration-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"A real migration plan therefore needs to map session state, custom orchestration, handoffs, tool execution, approvals, storage, tracing, retries, failure recovery, environment lifecycle, and any provider-specific abstractions. Code volume may fall while operational assumptions change.\"},\"tunes\":{}},{\"id\":\"h-migration-inventory\",\"type\":\"header\",\"data\":{\"text\":\"The migration inventory\",\"level\":3},\"tunes\":{}},{\"id\":\"migration-list\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Agent definitions and instruction ownership.\",\"Tool definitions and where each tool executes.\",\"Handoffs, manager\u002Fspecialist patterns, and subagent behaviour.\",\"Session identifiers, conversation state, resumability, and history retention.\",\"Human approvals and interruption semantics.\",\"Custom context trimming or compaction logic.\",\"Tracing, evals, observability, and production debugging.\",\"Self-hosted files, containers, private-network access, or other execution dependencies.\",\"Provider abstraction or non-OpenAI model dependencies.\",\"Retry, timeout, idempotency, recovery, and lifecycle assumptions.\"]},\"tunes\":{}},{\"id\":\"h-beta\",\"type\":\"header\",\"data\":{\"text\":\"Public beta changes the risk model\",\"level\":2},\"tunes\":{}},{\"id\":\"p-beta-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Agents API is the recommended direction for new agent applications, but it is also in public beta. Those facts are not contradictory. Strategic direction answers “where is the platform going?” Beta status answers “how much interface and operational change should I budget for?”\"},\"tunes\":{}},{\"id\":\"p-beta-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"For production systems, isolate the integration behind an application boundary. Keep domain state, permissions, audit data, and business rules outside vendor-specific session objects where possible. This makes it easier to absorb API evolution without turning the agent runtime into the source of truth for your entire product.\"},\"tunes\":{}},{\"id\":\"h-default\",\"type\":\"header\",\"data\":{\"text\":\"A practical default architecture\",\"level\":2},\"tunes\":{}},{\"id\":\"p-default-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"For many new OpenAI-native applications, a reasonable 2026 default is: Agents API for the managed harness and durable session, application-owned domain services and authorization, explicit function tools for business actions, and either OpenAI-hosted or self-hosted execution depending on data and compute requirements.\"},\"tunes\":{}},{\"id\":\"p-default-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"That keeps the agent runtime powerful without making it the owner of business truth. The application still decides what a user may do, which data is authoritative, which actions require approval, and how results are validated.\"},\"tunes\":{}},{\"id\":\"ref-reliability\",\"type\":\"referralArticle\",\"data\":{\"url\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fai-agent-reliability-why-the-final-answer-is-not-enough\",\"title\":\"AI Agent Reliability: Why the Final Answer Is Not Enough\",\"excerpt\":\"A managed harness reduces operational burden, but reliability still requires traceable execution, explicit acceptance criteria and evaluation beyond the final answer.\",\"ctaLabel\":\"Read the reliability article\"},\"tunes\":{}},{\"id\":\"h-change\",\"type\":\"header\",\"data\":{\"text\":\"What would change this answer?\",\"level\":2},\"tunes\":{}},{\"id\":\"p-change-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The recommendation changes if the Agents API adds or removes capabilities, exits beta with different contracts, changes environment or pricing boundaries, or introduces migration tooling that reduces ownership differences. It also changes if your application depends on provider portability, custom orchestration semantics, local-only execution, or a capability the hosted harness cannot support.\"},\"tunes\":{}},{\"id\":\"p-change-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"For an existing Agents SDK application, the answer also changes with migration cost. If the system is stable, well evaluated, and not blocked by the SDK's feature-complete status, immediate migration may create more risk than value. If the product roadmap depends on capabilities landing only in the Agents API, delaying migration can create a different kind of debt.\"},\"tunes\":{}},{\"id\":\"h-limitations\",\"type\":\"header\",\"data\":{\"text\":\"Limitations\",\"level\":2},\"tunes\":{}},{\"id\":\"p-limit-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"This comparison focuses on runtime ownership and OpenAI's stated platform direction. It does not benchmark latency, quality, or total cost for a specific workload. Those properties depend on model choice, tool usage, environment, task length, caching, sandbox usage, and application architecture.\"},\"tunes\":{}},{\"id\":\"p-limit-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"The Agents API is also new enough that production experience is still accumulating. A design should therefore be validated with representative workloads rather than chosen solely from product positioning.\"},\"tunes\":{}},{\"id\":\"h-conclusion\",\"type\":\"header\",\"data\":{\"text\":\"Conclusion\",\"level\":2},\"tunes\":{}},{\"id\":\"p-conclusion-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The 2026 OpenAI agent decision is fundamentally about runtime ownership. Agents API means OpenAI owns more of the harness and durable-session machinery. Responses means your application owns the loop around platform primitives. The Agents SDK remains valid for existing systems but is no longer the default destination for major new agent-runtime features.\"},\"tunes\":{}},{\"id\":\"p-conclusion-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"For a new application, follow the platform direction unless a real requirement pushes you lower in the stack. Start with the Agents API, move to Responses when you need to own the loop, evaluate Codex SDK when you need the harness in your infrastructure, and keep the Agents SDK where existing investments or temporary capability gaps justify it.\"},\"tunes\":{}},{\"id\":\"h-faq\",\"type\":\"header\",\"data\":{\"text\":\"FAQ\",\"level\":2},\"tunes\":{}},{\"id\":\"faq\",\"type\":\"faq\",\"data\":{\"title\":\"OpenAI agent runtime choices in 2026\",\"items\":[{\"id\":\"faq1\",\"question\":\"Should I use the OpenAI Agents API or Agents SDK for a new project?\",\"answer\":\"OpenAI currently recommends the Agents API for new agent applications. The Agents SDK is feature complete and remains supported for existing applications, with maintenance, security, critical bug fixes and compatibility work continuing.\"},{\"id\":\"faq2\",\"question\":\"Is the Agents SDK deprecated?\",\"answer\":\"OpenAI describes it as feature complete, not as unsupported. Major new features are not planned, but maintenance, security fixes, critical bug fixes and compatibility work continue.\"},{\"id\":\"faq3\",\"question\":\"When should I use the Responses API instead of the Agents API?\",\"answer\":\"Use Responses when your application should own the agent loop, state strategy, orchestration and continuation logic while still using OpenAI models and platform tools.\"},{\"id\":\"faq4\",\"question\":\"Does the Agents API require OpenAI-hosted compute?\",\"answer\":\"No. The Agents API can use no execution environment, an OpenAI-hosted environment, or a self-hosted environment connected to the managed harness.\"},{\"id\":\"faq5\",\"question\":\"Where does the Codex SDK fit?\",\"answer\":\"OpenAI positions the Codex SDK for running the Codex harness in infrastructure you operate. It is the relevant option when you want the harness but not the hosted Agents API runtime.\"},{\"id\":\"faq6\",\"question\":\"Should an existing Agents SDK app migrate immediately?\",\"answer\":\"Not automatically. Evaluate whether the current system is blocked by the SDK's feature-complete status, whether required capabilities exist in the Agents API, and whether the migration benefit exceeds the operational and architectural change.\"}]},\"tunes\":{}},{\"id\":\"h-glossary\",\"type\":\"header\",\"data\":{\"text\":\"Glossary\",\"level\":2},\"tunes\":{}},{\"id\":\"glossary\",\"type\":\"glossary\",\"data\":{\"title\":\"Key runtime terms\",\"entries\":[{\"term\":\"Harness\",\"definition\":\"The runtime loop and supporting machinery that coordinates model calls, tools, context, sessions and agent execution.\",\"anchor\":\"harness\"},{\"term\":\"Agents API\",\"definition\":\"OpenAI's managed API for durable cloud agents using a hosted Codex harness.\",\"anchor\":\"agents-api\"},{\"term\":\"Responses API\",\"definition\":\"OpenAI's lower-level API primitive for model responses, hosted tools and stateful interactions around which applications can build their own agent loop.\",\"anchor\":\"responses-api\"},{\"term\":\"Agents SDK\",\"definition\":\"OpenAI's open-source framework for running agent workflows in application code; feature complete as of September 2026.\",\"anchor\":\"agents-sdk\"},{\"term\":\"Codex SDK\",\"definition\":\"The runtime option OpenAI lists for running the Codex harness in infrastructure you operate.\",\"anchor\":\"codex-sdk\"},{\"term\":\"Runtime ownership\",\"definition\":\"The architectural boundary describing which parts of the agent loop, session state, execution environment and lifecycle are operated by the platform versus the application.\",\"anchor\":\"runtime-ownership\"}]},\"tunes\":{}},{\"id\":\"h-sources\",\"type\":\"header\",\"data\":{\"text\":\"Primary sources and further reading\",\"level\":2},\"tunes\":{}},{\"id\":\"src-launch\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fopenai.com\u002Findex\u002Fintroducing-the-agents-api\u002F\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Introducing the Agents API\",\"description\":\"Launch announcement for the Agents API on 10 September 2026, describing the managed Codex harness and public beta.\"}},\"tunes\":{}},{\"id\":\"src-agents-overview\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Agents runtime overview\",\"description\":\"Current comparison of Agents API, Codex SDK and Responses API, including the support status of the Agents SDK.\"}},\"tunes\":{}},{\"id\":\"src-agents-api\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Foverview\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Agents API overview\",\"description\":\"Documentation for durable cloud agents, sessions, orchestration, context compaction, recovery and environment choices.\"}},\"tunes\":{}},{\"id\":\"src-architecture\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Farchitecture\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Agents API architecture\",\"description\":\"Architecture boundary between the hosted harness, application server and no\u002Fopenai-hosted\u002Fself-hosted execution environments.\"}},\"tunes\":{}},{\"id\":\"src-sdk\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\u002Fsdk\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Agents SDK\",\"description\":\"Current Agents SDK support notice and explanation of the application-owned agent loop.\"}},\"tunes\":{}},{\"id\":\"src-responses\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fmigrate-to-responses\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Migrate to the Responses API\",\"description\":\"Current Responses API positioning, built-in tools, stateful context and agentic primitives.\"}},\"tunes\":{}},{\"id\":\"toc\",\"type\":\"tableOfContents\",\"data\":{\"title\":\"Contents\",\"minLevel\":2,\"maxLevel\":3},\"tunes\":{}}],\"version\":\"2.31.6\"}",{"time":834,"blocks":835,"version":1297},1790352010644,[836,840,844,849,854,859,863,867,871,875,905,909,913,917,921,925,936,940,950,954,958,962,966,970,974,978,982,987,991,995,999,1003,1007,1040,1044,1070,1074,1102,1106,1110,1114,1118,1133,1137,1141,1145,1149,1153,1157,1164,1168,1172,1176,1180,1184,1188,1192,1196,1200,1204,1227,1231,1249,1253,1260,1267,1274,1281,1287,1294],{"id":215,"data":837,"type":220,"tunes":839},{"title":838,"maxLevel":218,"minLevel":219},"Contents",{},{"id":223,"data":841,"type":226,"tunes":843},{"text":842},"OpenAI's agent stack changed materially in September 2026. The new Agents API introduced a managed Codex harness for durable cloud agents, while the older Agents SDK moved into a feature-complete maintenance phase. The Responses API remains the lower-level surface for applications that want direct model calls or to own the agent loop themselves. These are not three interchangeable wrappers around the same thing: they place the runtime boundary in different places.",{},{"id":229,"data":845,"type":234,"tunes":848},{"body":846,"title":847,"variant":233},"\u003Cstrong>For a new OpenAI-native agent application in September 2026, start with the Agents API unless you have a concrete reason to own more of the runtime.\u003C\u002Fstrong> Use the Responses API when your application should own the loop, state strategy, and orchestration. Keep the Agents SDK for existing applications or short-term gaps where the Agents API does not yet support a required capability. If you want the Codex harness itself but need to run it in infrastructure you operate, evaluate the Codex SDK rather than forcing the Agents SDK into that role.","Direct answer",{},{"id":237,"data":850,"type":234,"tunes":853},{"body":851,"title":852,"variant":241},"This article reflects OpenAI's platform state on \u003Cstrong>25 September 2026\u003C\u002Fstrong>. The Agents API launched in public beta on 10 September 2026. Beta surfaces, feature coverage, pricing, and migration guidance can change; verify current documentation before committing a production architecture.","Time-sensitive architecture decision",{},{"id":244,"data":855,"type":234,"tunes":858},{"body":856,"title":857,"variant":248},"The Runtime Ownership Test below is a practical architecture model, not an OpenAI product taxonomy. It asks which parts of the agent runtime your team must control versus which parts you want OpenAI to operate.","The decision model used here",{},{"id":251,"data":860,"type":42,"tunes":862},{"text":861,"level":219},"The architecture changed: choose a runtime boundary, not a library",{},{"id":256,"data":864,"type":226,"tunes":866},{"text":865},"The important decision is no longer simply “Which SDK should I install?” It is who owns the harness, the agent loop, durable session state, context compaction, recovery, execution environment, and application lifecycle.",{},{"id":261,"data":868,"type":226,"tunes":870},{"text":869},"OpenAI's current Agents overview makes this boundary explicit. The Agents API runs a hosted Codex harness and manages orchestration plus durable session state. The Responses API provides model responses and hosted capabilities while your application owns the surrounding agent loop. The Agents SDK runs the loop in your application and is now feature complete rather than the forward path for major new agent features.",{},{"id":266,"data":872,"type":42,"tunes":874},{"text":873,"level":219},"The short comparison",{},{"id":271,"data":876,"type":304,"tunes":904},{"content":877,"stretched":43,"withHeadings":14},[878,884,889,894,899],[879,880,881,882,883],"Option","Best fit in 2026","Who owns the agent loop?","Session \u002F context responsibility","Strategic status",[281,885,886,887,888],"New OpenAI-native durable agents","OpenAI-managed Codex harness","OpenAI manages sessions, orchestration, compaction and recovery","Recommended starting point for new agent apps; public beta",[287,890,891,892,893],"Direct model integrations and custom agent runtimes","Your application","You choose response chaining, Conversations, storage and loop logic","Core API primitive; recommended over Chat Completions for new projects",[293,895,896,897,898],"Existing SDK applications or temporary capability gaps","Your application via the SDK runner","Your application operates deployment, storage and runtime behaviour","Feature complete; maintenance and compatibility continue, major new features not planned",[299,900,901,902,903],"Codex harness in infrastructure you operate","Codex harness in your environment","You operate harness hosting and lifecycle","Separate option when you want the harness without the hosted Agents API runtime",{},{"id":307,"data":906,"type":42,"tunes":908},{"text":907,"level":219},"1. Agents API: managed harness, durable cloud agent",{},{"id":312,"data":910,"type":226,"tunes":912},{"text":911},"The Agents API exposes the Codex harness through an OpenAI-managed service. OpenAI manages sessions, orchestration, context compaction, and recovery. Your application still provides tools and chooses the execution environment.",{},{"id":317,"data":914,"type":226,"tunes":916},{"text":915},"That last distinction matters. “Managed agent” does not necessarily mean “all compute runs inside OpenAI.” The Agents API architecture supports no environment, an OpenAI-hosted environment, or a self-hosted environment connected to the hosted harness. With a self-hosted environment, your application owns provisioning, reconnection, shutdown, and persistent files while the harness remains managed.",{},{"id":322,"data":918,"type":226,"tunes":920},{"text":919},"The Agents API is therefore a runtime service, not merely a request format. Sessions can persist, stream progress, receive additional tasks, use tools, work with files, and recover across long-running work.",{},{"id":327,"data":922,"type":42,"tunes":924},{"text":923,"level":218},"What you gain with the Agents API",{},{"id":332,"data":926,"type":343,"tunes":935},{"meta":927,"items":928,"style":342},{},[929,930,931,932,933,934],"A managed Codex harness instead of building and operating the main agent loop yourself.","Durable sessions for work that spans turns and long-running tasks.","Managed orchestration, context compaction, and recovery.","OpenAI-hosted or self-hosted execution environments, depending on workload requirements.","Streaming and webhooks for progress and lifecycle events.","A platform direction that OpenAI explicitly recommends for new agent applications.",{},{"id":346,"data":937,"type":42,"tunes":939},{"text":938,"level":218},"What you still own",{},{"id":351,"data":941,"type":343,"tunes":949},{"meta":942,"items":943,"style":342},{},[944,945,946,947,948],"Your product and application server.","Function-tool implementations and business logic.","Authorization and policy decisions around your own systems.","The execution environment lifecycle when you choose self-hosted compute.","Evaluation, acceptance criteria, domain-specific guardrails, and the decision of what the agent is allowed to do.",{},{"id":362,"data":951,"type":42,"tunes":953},{"text":952,"level":219},"2. Responses API: own the loop, use the platform primitives",{},{"id":367,"data":955,"type":226,"tunes":957},{"text":956},"The Responses API is the lower-level choice when you want OpenAI model and tool capabilities without delegating the overall agent runtime. OpenAI describes Responses as the recommended API primitive for new projects and as an evolution of Chat Completions with built-in tools, multi-turn state options, multimodal input, and agentic tool use.",{},{"id":372,"data":959,"type":226,"tunes":961},{"text":960},"A Responses request can itself invoke tools, but your application remains responsible for the larger workflow when you build an agent around it. That means your code decides how to persist application state, when to continue, how to recover, how to coordinate specialists, how to compact long histories, and how to represent resumable work.",{},{"id":377,"data":963,"type":226,"tunes":965},{"text":964},"This is not inherently inferior. It is the right boundary when agent behaviour must be deeply embedded in existing application logic, when you need a custom state model, or when a managed harness would hide control you actually need.",{},{"id":382,"data":967,"type":42,"tunes":969},{"text":968,"level":219},"3. Agents SDK: supported, but no longer the default forward path",{},{"id":387,"data":971,"type":226,"tunes":973},{"text":972},"The Agents SDK remains an open-source framework for running agent workflows in your application. It provides agent definitions, tools, handoffs, guardrails, tracing, sessions, and the runner loop in TypeScript and Python.",{},{"id":392,"data":975,"type":226,"tunes":977},{"text":976},"But its strategic status changed. OpenAI now labels the Agents SDK feature complete: maintenance, security fixes, critical bug fixes, and compatibility work continue, but major new features are not planned. OpenAI recommends the Agents API for new applications.",{},{"id":397,"data":979,"type":226,"tunes":981},{"text":980},"That does not mean an existing SDK application should be rewritten immediately. It means the architecture should stop assuming that the SDK is where the next major agent-runtime capabilities will land.",{},{"id":402,"data":983,"type":234,"tunes":986},{"body":984,"title":985,"variant":241},"OpenAI has not described the Agents SDK as unsupported. Existing applications can continue to use it, and maintenance, security, critical bug fixes, and compatibility work continue. The important change is that major new features are not planned and new applications are directed to the Agents API.","Do not translate “feature complete” into “deprecated”",{},{"id":408,"data":988,"type":42,"tunes":990},{"text":989,"level":219},"Where the Codex SDK fits",{},{"id":413,"data":992,"type":226,"tunes":994},{"text":993},"The current choice is not a simple three-way fork. OpenAI's runtime overview includes the Codex SDK as the option for running the Codex harness in infrastructure you operate. That is architecturally different from both the hosted Agents API and the Agents SDK.",{},{"id":418,"data":996,"type":226,"tunes":998},{"text":997},"If your real requirement is “I want the Codex harness, but I need to operate it myself,” the Codex SDK is the surface to evaluate. If your requirement is “I want to own the loop around model calls,” evaluate Responses. If your requirement is “I already have a working Agents SDK application,” the existing SDK can remain valid while you plan against its maintenance status.",{},{"id":423,"data":1000,"type":42,"tunes":1002},{"text":1001,"level":219},"The Runtime Ownership Test",{},{"id":428,"data":1004,"type":226,"tunes":1006},{"text":1005},"A useful architecture decision starts by deciding what your team must own. Score each requirement as either must control, prefer to control, or prefer managed.",{},{"id":433,"data":1008,"type":475,"tunes":1039},{"rows":1009,"title":1031,"layout":304,"columns":1032},[1010,1013,1016,1019,1022,1025,1028],{"id":437,"label":1011,"values":1012},"Agent loop",[440,440,440],{"id":442,"label":1014,"values":1015},"Durable sessions",[440,440,440],{"id":446,"label":1017,"values":1018},"Harness runtime",[440,440,440],{"id":450,"label":1020,"values":1021},"Execution environment",[440,440,440],{"id":454,"label":1023,"values":1024},"Orchestration semantics",[440,440,440],{"id":458,"label":1026,"values":1027},"Provider \u002F transport flexibility",[440,440,440],{"id":462,"label":1029,"values":1030},"Operational burden",[440,440,440],"Runtime Ownership Test",[1033,1035,1037],{"id":467,"label":1034},"Decision",{"id":470,"label":1036},"If you prefer managed",{"id":473,"label":1038},"If you require control",{},{"id":478,"data":1041,"type":42,"tunes":1043},{"text":1042,"level":219},"A decision tree for new systems",{},{"id":483,"data":1045,"type":509,"tunes":1069},{"steps":1046,"title":1068,"orientation":508},[1047,1050,1053,1056,1059,1062,1065],{"label":1048,"description":1049},"1. Is this a new agent application?","If no, do not migrate solely because a newer surface exists. Evaluate the current application's actual constraints first.",{"label":1051,"description":1052},"2. Do you want a managed long-running agent harness?","If yes, start with the Agents API as OpenAI's recommended path for new agent applications.",{"label":1054,"description":1055},"3. Do you need the Codex harness but must operate it yourself?","Evaluate the Codex SDK rather than rebuilding harness behaviour on top of the Agents SDK.",{"label":1057,"description":1058},"4. Do you need to own the agent loop and state model?","Use the Responses API as the lower-level platform primitive and build the loop around it.",{"label":1060,"description":1061},"5. Are you already on the Agents SDK?","Continue if it meets requirements; major new features are not planned, so treat future platform adoption as an explicit roadmap decision.",{"label":1063,"description":1064},"6. Is a required capability missing from the Agents API?","OpenAI explicitly allows the Agents SDK as a short-term option for new applications that need unsupported capabilities.",{"label":1066,"description":1067},"7. Validate with a production-shaped spike","Test tools, environment, approvals, latency, observability, failure recovery, and lifecycle before locking the architecture.","Choose the runtime by the control boundary",{},{"id":512,"data":1071,"type":42,"tunes":1073},{"text":1072,"level":219},"What should not drive the decision",{},{"id":517,"data":1075,"type":304,"tunes":1101},{"content":1076,"stretched":43,"withHeadings":14},[1077,1081,1085,1089,1093,1097],[1078,1079,1080],"Weak decision rule","Why it fails","Better question",[1082,1083,1084],"“The newest API must be best.”","Newer can be strategically preferred while still lacking a capability you need.","Which runtime responsibilities should be managed versus application-owned?",[1086,1087,1088],"“We already know the SDK.”","Team familiarity can preserve an architecture whose roadmap has changed.","What is the cost of staying versus moving over the next product cycle?",[1090,1091,1092],"“Managed means no infrastructure.”","Agents API can still use self-hosted environments and your application still owns product logic.","Which infrastructure layer is actually being delegated?",[1094,1095,1096],"“Responses is only for simple calls.”","Responses provides built-in tools and stateful primitives; it can be the foundation of custom agent loops.","Do we need the platform to own the harness, or only model\u002Ftool primitives?",[1098,1099,1100],"“Feature complete means we must migrate now.”","The SDK remains maintained for existing applications.","What concrete future requirement is blocked by staying?",{},{"id":546,"data":1103,"type":42,"tunes":1105},{"text":1104,"level":219},"Migration is an architectural change, not an import rename",{},{"id":551,"data":1107,"type":226,"tunes":1109},{"text":1108},"Moving from the Agents SDK to the Agents API changes ownership. In the SDK, the loop runs in your application. In the Agents API, OpenAI runs the harness and session while your application integrates through tasks, events, tools, and environment boundaries.",{},{"id":556,"data":1111,"type":226,"tunes":1113},{"text":1112},"A real migration plan therefore needs to map session state, custom orchestration, handoffs, tool execution, approvals, storage, tracing, retries, failure recovery, environment lifecycle, and any provider-specific abstractions. Code volume may fall while operational assumptions change.",{},{"id":561,"data":1115,"type":42,"tunes":1117},{"text":1116,"level":218},"The migration inventory",{},{"id":566,"data":1119,"type":343,"tunes":1132},{"meta":1120,"items":1121,"style":342},{},[1122,1123,1124,1125,1126,1127,1128,1129,1130,1131],"Agent definitions and instruction ownership.","Tool definitions and where each tool executes.","Handoffs, manager\u002Fspecialist patterns, and subagent behaviour.","Session identifiers, conversation state, resumability, and history retention.","Human approvals and interruption semantics.","Custom context trimming or compaction logic.","Tracing, evals, observability, and production debugging.","Self-hosted files, containers, private-network access, or other execution dependencies.","Provider abstraction or non-OpenAI model dependencies.","Retry, timeout, idempotency, recovery, and lifecycle assumptions.",{},{"id":582,"data":1134,"type":42,"tunes":1136},{"text":1135,"level":219},"Public beta changes the risk model",{},{"id":587,"data":1138,"type":226,"tunes":1140},{"text":1139},"The Agents API is the recommended direction for new agent applications, but it is also in public beta. Those facts are not contradictory. Strategic direction answers “where is the platform going?” Beta status answers “how much interface and operational change should I budget for?”",{},{"id":592,"data":1142,"type":226,"tunes":1144},{"text":1143},"For production systems, isolate the integration behind an application boundary. Keep domain state, permissions, audit data, and business rules outside vendor-specific session objects where possible. This makes it easier to absorb API evolution without turning the agent runtime into the source of truth for your entire product.",{},{"id":597,"data":1146,"type":42,"tunes":1148},{"text":1147,"level":219},"A practical default architecture",{},{"id":602,"data":1150,"type":226,"tunes":1152},{"text":1151},"For many new OpenAI-native applications, a reasonable 2026 default is: Agents API for the managed harness and durable session, application-owned domain services and authorization, explicit function tools for business actions, and either OpenAI-hosted or self-hosted execution depending on data and compute requirements.",{},{"id":607,"data":1154,"type":226,"tunes":1156},{"text":1155},"That keeps the agent runtime powerful without making it the owner of business truth. The application still decides what a user may do, which data is authoritative, which actions require approval, and how results are validated.",{},{"id":612,"data":1158,"type":618,"tunes":1163},{"url":1159,"title":1160,"excerpt":1161,"ctaLabel":1162},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fai-agent-reliability-why-the-final-answer-is-not-enough","AI Agent Reliability: Why the Final Answer Is Not Enough","A managed harness reduces operational burden, but reliability still requires traceable execution, explicit acceptance criteria and evaluation beyond the final answer.","Read the reliability article",{},{"id":621,"data":1165,"type":42,"tunes":1167},{"text":1166,"level":219},"What would change this answer?",{},{"id":626,"data":1169,"type":226,"tunes":1171},{"text":1170},"The recommendation changes if the Agents API adds or removes capabilities, exits beta with different contracts, changes environment or pricing boundaries, or introduces migration tooling that reduces ownership differences. It also changes if your application depends on provider portability, custom orchestration semantics, local-only execution, or a capability the hosted harness cannot support.",{},{"id":631,"data":1173,"type":226,"tunes":1175},{"text":1174},"For an existing Agents SDK application, the answer also changes with migration cost. If the system is stable, well evaluated, and not blocked by the SDK's feature-complete status, immediate migration may create more risk than value. If the product roadmap depends on capabilities landing only in the Agents API, delaying migration can create a different kind of debt.",{},{"id":636,"data":1177,"type":42,"tunes":1179},{"text":1178,"level":219},"Limitations",{},{"id":641,"data":1181,"type":226,"tunes":1183},{"text":1182},"This comparison focuses on runtime ownership and OpenAI's stated platform direction. It does not benchmark latency, quality, or total cost for a specific workload. Those properties depend on model choice, tool usage, environment, task length, caching, sandbox usage, and application architecture.",{},{"id":646,"data":1185,"type":226,"tunes":1187},{"text":1186},"The Agents API is also new enough that production experience is still accumulating. A design should therefore be validated with representative workloads rather than chosen solely from product positioning.",{},{"id":651,"data":1189,"type":42,"tunes":1191},{"text":1190,"level":219},"Conclusion",{},{"id":656,"data":1193,"type":226,"tunes":1195},{"text":1194},"The 2026 OpenAI agent decision is fundamentally about runtime ownership. Agents API means OpenAI owns more of the harness and durable-session machinery. Responses means your application owns the loop around platform primitives. The Agents SDK remains valid for existing systems but is no longer the default destination for major new agent-runtime features.",{},{"id":661,"data":1197,"type":226,"tunes":1199},{"text":1198},"For a new application, follow the platform direction unless a real requirement pushes you lower in the stack. Start with the Agents API, move to Responses when you need to own the loop, evaluate Codex SDK when you need the harness in your infrastructure, and keep the Agents SDK where existing investments or temporary capability gaps justify it.",{},{"id":666,"data":1201,"type":42,"tunes":1203},{"text":1202,"level":219},"FAQ",{},{"id":671,"data":1205,"type":671,"tunes":1226},{"items":1206,"title":1225},[1207,1210,1213,1216,1219,1222],{"id":675,"answer":1208,"question":1209},"OpenAI currently recommends the Agents API for new agent applications. The Agents SDK is feature complete and remains supported for existing applications, with maintenance, security, critical bug fixes and compatibility work continuing.","Should I use the OpenAI Agents API or Agents SDK for a new project?",{"id":679,"answer":1211,"question":1212},"OpenAI describes it as feature complete, not as unsupported. Major new features are not planned, but maintenance, security fixes, critical bug fixes and compatibility work continue.","Is the Agents SDK deprecated?",{"id":683,"answer":1214,"question":1215},"Use Responses when your application should own the agent loop, state strategy, orchestration and continuation logic while still using OpenAI models and platform tools.","When should I use the Responses API instead of the Agents API?",{"id":687,"answer":1217,"question":1218},"No. The Agents API can use no execution environment, an OpenAI-hosted environment, or a self-hosted environment connected to the managed harness.","Does the Agents API require OpenAI-hosted compute?",{"id":691,"answer":1220,"question":1221},"OpenAI positions the Codex SDK for running the Codex harness in infrastructure you operate. It is the relevant option when you want the harness but not the hosted Agents API runtime.","Where does the Codex SDK fit?",{"id":695,"answer":1223,"question":1224},"Not automatically. Evaluate whether the current system is blocked by the SDK's feature-complete status, whether required capabilities exist in the Agents API, and whether the migration benefit exceeds the operational and architectural change.","Should an existing Agents SDK app migrate immediately?","OpenAI agent runtime choices in 2026",{},{"id":701,"data":1228,"type":42,"tunes":1230},{"text":1229,"level":219},"Glossary",{},{"id":706,"data":1232,"type":706,"tunes":1248},{"title":1233,"entries":1234},"Key runtime terms",[1235,1237,1239,1241,1243,1245],{"term":711,"anchor":446,"definition":1236},"The runtime loop and supporting machinery that coordinates model calls, tools, context, sessions and agent execution.",{"term":281,"anchor":714,"definition":1238},"OpenAI's managed API for durable cloud agents using a hosted Codex harness.",{"term":287,"anchor":717,"definition":1240},"OpenAI's lower-level API primitive for model responses, hosted tools and stateful interactions around which applications can build their own agent loop.",{"term":293,"anchor":720,"definition":1242},"OpenAI's open-source framework for running agent workflows in application code; feature complete as of September 2026.",{"term":299,"anchor":723,"definition":1244},"The runtime option OpenAI lists for running the Codex harness in infrastructure you operate.",{"term":1246,"anchor":727,"definition":1247},"Runtime ownership","The architectural boundary describing which parts of the agent loop, session state, execution environment and lifecycle are operated by the platform versus the application.",{},{"id":731,"data":1250,"type":42,"tunes":1252},{"text":1251,"level":219},"Primary sources and further reading",{},{"id":736,"data":1254,"type":743,"tunes":1259},{"link":738,"meta":1255},{"image":1256,"title":1257,"description":1258},{"url":440},"OpenAI — Introducing the Agents API","Launch announcement for the Agents API on 10 September 2026, describing the managed Codex harness and public beta.",{},{"id":746,"data":1261,"type":743,"tunes":1266},{"link":748,"meta":1262},{"image":1263,"title":1264,"description":1265},{"url":440},"OpenAI — Agents runtime overview","Current comparison of Agents API, Codex SDK and Responses API, including the support status of the Agents SDK.",{},{"id":755,"data":1268,"type":743,"tunes":1273},{"link":757,"meta":1269},{"image":1270,"title":1271,"description":1272},{"url":440},"OpenAI — Agents API overview","Documentation for durable cloud agents, sessions, orchestration, context compaction, recovery and environment choices.",{},{"id":764,"data":1275,"type":743,"tunes":1280},{"link":766,"meta":1276},{"image":1277,"title":1278,"description":1279},{"url":440},"OpenAI — Agents API architecture","Architecture boundary between the hosted harness, application server and no\u002Fopenai-hosted\u002Fself-hosted execution environments.",{},{"id":773,"data":1282,"type":743,"tunes":1286},{"link":775,"meta":1283},{"image":1284,"title":778,"description":1285},{"url":440},"Current Agents SDK support notice and explanation of the application-owned agent loop.",{},{"id":782,"data":1288,"type":743,"tunes":1293},{"link":784,"meta":1289},{"image":1290,"title":1291,"description":1292},{"url":440},"OpenAI — Migrate to the Responses API","Current Responses API positioning, built-in tools, stateful context and agentic primitives.",{},{"id":791,"data":1295,"type":220,"tunes":1296},{"title":838,"maxLevel":218,"minLevel":219},{},"2.31.6","OpenAI’s agent stack changed in September 2026. This architecture guide separates the Agents API, Agents SDK, Responses API, and Codex SDK by runtime ownership—so teams can choose the right control boundary instead of comparing product names.",{"lang":7,"title":208,"content":210,"contentJson":1300,"excerpt":795},{"time":212,"blocks":1301,"version":794},[1302,1305,1308,1311,1314,1317,1320,1323,1326,1329,1338,1341,1344,1347,1350,1353,1358,1361,1366,1369,1372,1375,1378,1381,1384,1387,1390,1393,1396,1399,1402,1405,1408,1430,1433,1444,1447,1457,1460,1463,1466,1469,1474,1477,1480,1483,1486,1489,1492,1495,1498,1501,1504,1507,1510,1513,1516,1519,1522,1525,1535,1538,1548,1551,1556,1561,1566,1571,1576,1581],{"id":215,"data":1303,"type":220,"tunes":1304},{"title":217,"maxLevel":218,"minLevel":219},{},{"id":223,"data":1306,"type":226,"tunes":1307},{"text":225},{},{"id":229,"data":1309,"type":234,"tunes":1310},{"body":231,"title":232,"variant":233},{},{"id":237,"data":1312,"type":234,"tunes":1313},{"body":239,"title":240,"variant":241},{},{"id":244,"data":1315,"type":234,"tunes":1316},{"body":246,"title":247,"variant":248},{},{"id":251,"data":1318,"type":42,"tunes":1319},{"text":253,"level":219},{},{"id":256,"data":1321,"type":226,"tunes":1322},{"text":258},{},{"id":261,"data":1324,"type":226,"tunes":1325},{"text":263},{},{"id":266,"data":1327,"type":42,"tunes":1328},{"text":268,"level":219},{},{"id":271,"data":1330,"type":304,"tunes":1337},{"content":1331,"stretched":43,"withHeadings":14},[1332,1333,1334,1335,1336],[275,276,277,278,279],[281,282,283,284,285],[287,288,289,290,291],[293,294,295,296,297],[299,300,301,302,303],{},{"id":307,"data":1339,"type":42,"tunes":1340},{"text":309,"level":219},{},{"id":312,"data":1342,"type":226,"tunes":1343},{"text":314},{},{"id":317,"data":1345,"type":226,"tunes":1346},{"text":319},{},{"id":322,"data":1348,"type":226,"tunes":1349},{"text":324},{},{"id":327,"data":1351,"type":42,"tunes":1352},{"text":329,"level":218},{},{"id":332,"data":1354,"type":343,"tunes":1357},{"meta":1355,"items":1356,"style":342},{},[336,337,338,339,340,341],{},{"id":346,"data":1359,"type":42,"tunes":1360},{"text":348,"level":218},{},{"id":351,"data":1362,"type":343,"tunes":1365},{"meta":1363,"items":1364,"style":342},{},[355,356,357,358,359],{},{"id":362,"data":1367,"type":42,"tunes":1368},{"text":364,"level":219},{},{"id":367,"data":1370,"type":226,"tunes":1371},{"text":369},{},{"id":372,"data":1373,"type":226,"tunes":1374},{"text":374},{},{"id":377,"data":1376,"type":226,"tunes":1377},{"text":379},{},{"id":382,"data":1379,"type":42,"tunes":1380},{"text":384,"level":219},{},{"id":387,"data":1382,"type":226,"tunes":1383},{"text":389},{},{"id":392,"data":1385,"type":226,"tunes":1386},{"text":394},{},{"id":397,"data":1388,"type":226,"tunes":1389},{"text":399},{},{"id":402,"data":1391,"type":234,"tunes":1392},{"body":404,"title":405,"variant":241},{},{"id":408,"data":1394,"type":42,"tunes":1395},{"text":410,"level":219},{},{"id":413,"data":1397,"type":226,"tunes":1398},{"text":415},{},{"id":418,"data":1400,"type":226,"tunes":1401},{"text":420},{},{"id":423,"data":1403,"type":42,"tunes":1404},{"text":425,"level":219},{},{"id":428,"data":1406,"type":226,"tunes":1407},{"text":430},{},{"id":433,"data":1409,"type":475,"tunes":1429},{"rows":1410,"title":425,"layout":304,"columns":1425},[1411,1413,1415,1417,1419,1421,1423],{"id":437,"label":438,"values":1412},[440,440,440],{"id":442,"label":443,"values":1414},[440,440,440],{"id":446,"label":447,"values":1416},[440,440,440],{"id":450,"label":451,"values":1418},[440,440,440],{"id":454,"label":455,"values":1420},[440,440,440],{"id":458,"label":459,"values":1422},[440,440,440],{"id":462,"label":463,"values":1424},[440,440,440],[1426,1427,1428],{"id":467,"label":468},{"id":470,"label":471},{"id":473,"label":474},{},{"id":478,"data":1431,"type":42,"tunes":1432},{"text":480,"level":219},{},{"id":483,"data":1434,"type":509,"tunes":1443},{"steps":1435,"title":507,"orientation":508},[1436,1437,1438,1439,1440,1441,1442],{"label":487,"description":488},{"label":490,"description":491},{"label":493,"description":494},{"label":496,"description":497},{"label":499,"description":500},{"label":502,"description":503},{"label":505,"description":506},{},{"id":512,"data":1445,"type":42,"tunes":1446},{"text":514,"level":219},{},{"id":517,"data":1448,"type":304,"tunes":1456},{"content":1449,"stretched":43,"withHeadings":14},[1450,1451,1452,1453,1454,1455],[521,522,523],[525,526,527],[529,530,531],[533,534,535],[537,538,539],[541,542,543],{},{"id":546,"data":1458,"type":42,"tunes":1459},{"text":548,"level":219},{},{"id":551,"data":1461,"type":226,"tunes":1462},{"text":553},{},{"id":556,"data":1464,"type":226,"tunes":1465},{"text":558},{},{"id":561,"data":1467,"type":42,"tunes":1468},{"text":563,"level":218},{},{"id":566,"data":1470,"type":343,"tunes":1473},{"meta":1471,"items":1472,"style":342},{},[570,571,572,573,574,575,576,577,578,579],{},{"id":582,"data":1475,"type":42,"tunes":1476},{"text":584,"level":219},{},{"id":587,"data":1478,"type":226,"tunes":1479},{"text":589},{},{"id":592,"data":1481,"type":226,"tunes":1482},{"text":594},{},{"id":597,"data":1484,"type":42,"tunes":1485},{"text":599,"level":219},{},{"id":602,"data":1487,"type":226,"tunes":1488},{"text":604},{},{"id":607,"data":1490,"type":226,"tunes":1491},{"text":609},{},{"id":612,"data":1493,"type":618,"tunes":1494},{"url":614,"title":615,"excerpt":616,"ctaLabel":617},{},{"id":621,"data":1496,"type":42,"tunes":1497},{"text":623,"level":219},{},{"id":626,"data":1499,"type":226,"tunes":1500},{"text":628},{},{"id":631,"data":1502,"type":226,"tunes":1503},{"text":633},{},{"id":636,"data":1505,"type":42,"tunes":1506},{"text":638,"level":219},{},{"id":641,"data":1508,"type":226,"tunes":1509},{"text":643},{},{"id":646,"data":1511,"type":226,"tunes":1512},{"text":648},{},{"id":651,"data":1514,"type":42,"tunes":1515},{"text":653,"level":219},{},{"id":656,"data":1517,"type":226,"tunes":1518},{"text":658},{},{"id":661,"data":1520,"type":226,"tunes":1521},{"text":663},{},{"id":666,"data":1523,"type":42,"tunes":1524},{"text":668,"level":219},{},{"id":671,"data":1526,"type":671,"tunes":1534},{"items":1527,"title":698},[1528,1529,1530,1531,1532,1533],{"id":675,"answer":676,"question":677},{"id":679,"answer":680,"question":681},{"id":683,"answer":684,"question":685},{"id":687,"answer":688,"question":689},{"id":691,"answer":692,"question":693},{"id":695,"answer":696,"question":697},{},{"id":701,"data":1536,"type":42,"tunes":1537},{"text":703,"level":219},{},{"id":706,"data":1539,"type":706,"tunes":1547},{"title":708,"entries":1540},[1541,1542,1543,1544,1545,1546],{"term":711,"anchor":446,"definition":712},{"term":281,"anchor":714,"definition":715},{"term":287,"anchor":717,"definition":718},{"term":293,"anchor":720,"definition":721},{"term":299,"anchor":723,"definition":724},{"term":726,"anchor":727,"definition":728},{},{"id":731,"data":1549,"type":42,"tunes":1550},{"text":733,"level":219},{},{"id":736,"data":1552,"type":743,"tunes":1555},{"link":738,"meta":1553},{"image":1554,"title":741,"description":742},{"url":440},{},{"id":746,"data":1557,"type":743,"tunes":1560},{"link":748,"meta":1558},{"image":1559,"title":751,"description":752},{"url":440},{},{"id":755,"data":1562,"type":743,"tunes":1565},{"link":757,"meta":1563},{"image":1564,"title":760,"description":761},{"url":440},{},{"id":764,"data":1567,"type":743,"tunes":1570},{"link":766,"meta":1568},{"image":1569,"title":769,"description":770},{"url":440},{},{"id":773,"data":1572,"type":743,"tunes":1575},{"link":775,"meta":1573},{"image":1574,"title":778,"description":779},{"url":440},{},{"id":782,"data":1577,"type":743,"tunes":1580},{"link":784,"meta":1578},{"image":1579,"title":787,"description":788},{"url":440},{},{"id":791,"data":1582,"type":220,"tunes":1583},{"title":217,"maxLevel":218,"minLevel":219},{},"Post erfolgreich abgerufen",{"items":1586,"source":1615,"manualIds":1616,"manualMatchedIds":1617},[1587,1594,1601,1608],{"id":1588,"slug":1589,"title":1590,"excerpt":1591,"featuredImage":1592,"publishedAt":1593},"478","what-is-rag-the-simplest-explanation-of-how-it-works","什么是RAG？对其工作原理的最简单解释","RAG听起来很复杂，但想法很简单：在AI回答之前，它先从知识源查找有用的信息，并将该信息提供给语言模型。本指南使用一个简单的思维模型来解释RAG、LLM、状态、记忆和工具。","\u002Fuploads\u002F2026\u002F09\u002Fwhat-is-rag-the-simplest-explanation-of-how-it-works-1790377492124-khjagt.webp","2026-09-25T19:03:00.000Z",{"id":1595,"slug":1596,"title":1597,"excerpt":1598,"featuredImage":1599,"publishedAt":1600},"382","a-practical-monorepo-architecture-next-js-platform-admin-fastify-api-prisma-and-nginx","基于Next.js、Fastify、Prisma和NGINX的实用单体仓库架构","探索一种实用的单体仓库架构，结合Next.js、Fastify、Prisma与NGINX，重点展示实际集成与工作流程。","\u002Fuploads\u002F2026\u002F01\u002Fa-practical-monorepo-architecture-next-js-platform-admin-fastify-api-prisma-and-nginx-1769885526116-q1100n.webp","2026-01-31T08:18:00.000Z",{"id":1602,"slug":1603,"title":1604,"excerpt":1605,"featuredImage":1606,"publishedAt":1607},"476","mcp-vs-a2a-vs-ucp-vs-ap2-vs-a2ui-the-agent-protocol-stack-explained","MCP vs A2A vs UCP vs AP2 vs A2UI：智能体协议栈详解","MCP、A2A、UCP、AP2 和 A2UI 常被描述为相互竞争的智能体标准。它们大多解决的是不同的互操作性问题。本指南将每个协议映射到其实际标准化的边界，并展示它们如何在同一个生产系统中协同工作。","\u002Fuploads\u002F2026\u002F09\u002Fmcp-vs-a2a-vs-ucp-vs-ap2-vs-a2ui-the-agent-protocol-stack-explained-1790352625869-2ezle0.webp","2026-09-25T12:09:00.000Z",{"id":1609,"slug":1610,"title":1611,"excerpt":1612,"featuredImage":1613,"publishedAt":1614},"381","enterprise-grade-multi-tenant-architecture-for-an-international-platform","企业级多租户架构，适用于国际平台","Loving Rocks 是一款企业级婚礼平台，采用真正的多租户架构设计，实现租户间数据库隔离，并内置国际化支持，以确保全球可扩展性、安全性及长期运营稳定性。","\u002Fuploads\u002F2026\u002F01\u002Fenterprise-grade-multi-tenant-architecture-for-an-international-platform-1769789121298-b6v7ak.webp","2026-01-30T12:04:00.000Z","fallback",[],[]]