[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:stajic:sr":3,"public-menus:all":38,"post:managed-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose:sr":205,"related:post:managed-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose:sr:1":1783},{"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","sr","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":1782},{"id":207,"title":208,"slug":209,"content":210,"contentJson":211,"excerpt":881,"featuredImage":882,"featuredImageAlt":883,"featuredImageCaption":10,"featuredImageTitle":10,"featuredImageCopyright":10,"featuredImageAuthor":10,"featuredImageSourceUrl":10,"featuredImageLicense":10,"featuredImageIsAiGenerated":43,"status":884,"publishedAt":885,"createdAt":886,"updatedAt":887,"seoLocalePaths":888,"categories":897,"author":910,"translations":915},"475","Upravljani harness za agente naspram samostalno hostovane petlje agenta: Šta dobijate, šta gubite","managed-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose","\u003Cnav class=\"editorjs-toc\" data-editorjs-toc=\"true\" aria-label=\"Sadržaj\">\u003Cstrong class=\"editorjs-toc__title\">Sadržaj\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\">Greška: tretiranje samostalnog hostovanja kao jedne odluke\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-10\" class=\"editorjs-toc__link\">Tri arhitekture koje se često nazivaju „samostalno hostovanim”\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-12\" class=\"editorjs-toc__link\">Model sa dva nivoa\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-15\" class=\"editorjs-toc__link\">Upravljani okvir: šta zapravo dobijate\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-19\" class=\"editorjs-toc__link\">Upravljano okruženje (harness): čega se odričete\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-23\" class=\"editorjs-toc__link\">Samo-hostovano okruženje za izvršavanje: srednja arhitektura\u003C\u002Fa>\u003Col class=\"editorjs-toc__list editorjs-toc__list--depth-1\">\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-27\" class=\"editorjs-toc__link\">Kada je samo-hostovano izvršavanje dovoljno\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-29\" class=\"editorjs-toc__link\">Kada ćete možda morati da posedujete i okruženje (harness)\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-33\" class=\"editorjs-toc__link\">Test eskalacije kontrole\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-36\" class=\"editorjs-toc__link\">Operativno opterećenje raste nelinearno kada posedujete okruženje (harness)\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-40\" class=\"editorjs-toc__link\">Bezbednosna granica: samostalno hostovanje računarskih resursa ne čini agenta automatski privatnim\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-44\" class=\"editorjs-toc__link\">Latencija i troškovi: kontrola može pomeriti uska grla umesto da ih ukloni\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-47\" class=\"editorjs-toc__link\">Matrica odluka za produkciju\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-49\" class=\"editorjs-toc__link\">Hibridni model nije kompromis — to je često čista arhitektura\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-53\" class=\"editorjs-toc__link\">Šta bi promenilo ovaj odgovor?\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-56\" class=\"editorjs-toc__link\">Ograničenja\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-59\" class=\"editorjs-toc__link\">Zaključak\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-62\" class=\"editorjs-toc__link\">Često postavljana pitanja\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-64\" class=\"editorjs-toc__link\">Rečnik pojmova\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"editorjs-toc__item\">\u003Ca href=\"#section-66\" class=\"editorjs-toc__link\">Primarni izvori i preporučena literatura\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Fol>\u003C\u002Fnav>\n\u003Cp>Izraz „samostalno hostovani agent” (self-hosted agent) danas krije najmanje tri različite arhitekture. Možete koristiti upravljani okvir (managed harness) sa računarskim resursima koje hostuje OpenAI, upravljani okvir povezan sa infrastrukturom kojom sami upravljate, ili sami pokretati i okvir i petlju agenta. Ovi izbori imaju veoma različite implikacije po pitanju kontrole, oporavka, upravljanja kontekstom, bezbednosti, kašnjenja i operativnog opterećenja.\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\">Direktan odgovor\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\"><strong>Nemojte birati između „upravljanog” i „samostalno hostovanog” kao da je to binarna odluka.<\u002Fstrong> Razdvojite <strong>nivo okvira (harness plane)<\u002Fstrong> od <strong>nivoa izvršavanja (execution plane)<\u002Fstrong>. Upravljani okvir i dalje može koristiti samostalno hostovane računarske resurse. Samostalno hostovano okruženje vam pruža kontrolu nad datotekama, paketima, mrežnim pristupom i izvršavanjem, a da pritom ne zahteva od vas da posedujete samu petlju agenta. Pokrećite okvir sami samo onda kada vam je potrebna kontrola nad orkestracijom, životnim ciklusom, pretpostavkama o usmeravanju modela ili ponašanjem u vremenu izvršavanja koje upravljani okvir ne može da pruži.\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\">Aktuelno na dan 25. septembra 2026.\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">OpenAI Agents API je u javnoj beta verziji i njegova arhitektura može evoluirati. Trenutna dokumentacija razdvaja Codex okvir koji hostuje OpenAI od okruženja za izvršavanje i eksplicitno podržava samostalno hostovana okruženja. OpenAI takođe nudi Codex okvir zasebno putem Codex SDK-a za infrastrukturu kojom sami upravljate.\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\">Model korišćen u ovom članku\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">Model nivoa okvira \u002F nivoa izvršavanja i Test eskalacije kontrole u nastavku predstavljaju praktične arhitektonske alate predložene ovde. Oni ne predstavljaju formalnu terminologiju vendora.\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-6\">Greška: tretiranje samostalnog hostovanja kao jedne odluke\u003C\u002Fh2>\n\u003Cp>U konvencionalnom softveru, „samostalno hostovano” obično znači da se aplikacija izvršava na infrastrukturi koju sami kontrolišete. Sistemi agenata komplikuju tu definiciju jer se okruženje za izvršavanje može podeliti. Petlja modela i alata može se izvršavati na jednom mestu, dok se izvršavanje koda, rad sa datotekama i pristup privatnoj mreži odvijaju negde drugde.\u003C\u002Fp>\n\u003Cp>Trenutna arhitektura OpenAI Agents API-ja čini ovu podelu eksplicitnom: OpenAI pokreće okvir, dok okruženje za izvršavanje može biti odsutno, hostovano od strane OpenAI-ja ili samostalno hostovano. Samostalno hostovano okruženje, dakle, ne znači da je i petlja agenta samostalno hostovana.\u003C\u002Fp>\n\u003Cp>Ova razlika je važna jer mnogi timovi biraju složenije runtime okruženje nego što im je potrebno. Žele pristup privatnoj mreži ili prilagođene pakete, zaključe da ceo agent mora biti samostalno hostovan i slučajno preuzmu vlasništvo nad upravljanjem kontekstom, orkestracijom, oporavkom i životnim ciklusom, što je sve moglo ostati pod upravljanjem platforme.\u003C\u002Fp>\n\u003Ch2 id=\"section-10\">Tri arhitekture koje se često nazivaju „samostalno hostovanim”\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\">Arhitektura\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Ko pokreće okvir?\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Gde se izvršavaju kod\u002Fdatoteke\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Šta primarno posedujete\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Upravljani okvir + upravljano okruženje\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Platforma\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Sandbox okruženje koje hostuje platforma\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Aplikacija, alati, logika proizvoda, autorizacija\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Upravljani okvir + samostalno hostovano okruženje\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Platforma\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Vaš kontejner, VM, laptop, privatni oblak ili drugi računarski resursi\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Obezbeđivanje okruženja, umrežavanje, datoteke i životni ciklus; platforma i dalje poseduje okvir\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Okvir \u002F petlja agenta pod sopstvenom upravom\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Vi\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Okruženje po vašem izboru\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Proces okvira, orkestracija, strategija konteksta, hosting, oporavak, izvršavanje i životni ciklus aplikacije\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-12\">Model sa dva nivoa\u003C\u002Fh2>\n\u003Csection class=\"editorjs-comparison my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">Razdvojite nivo okvira od nivoa izvršavanja\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\">Nivo\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\">Šta poseduje\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\">Pitanja koja treba postaviti\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\">Nivo okvira (Harness plane)\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\">Nivo izvršavanja (Execution plane)\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\">Nivo aplikacije (Application plane)\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\u003Caside class=\"editorjs-callout editorjs-callout--success my-6 rounded-xl border p-5 border-emerald-300 bg-emerald-50 dark:border-emerald-900 dark:bg-emerald-950\u002F20\" role=\"note\">\u003Cstrong class=\"block mb-2 text-gray-900 dark:text-gray-100\">Ključna arhitektonska posledica\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\">Možete samostalno hostovati <strong>nivo izvršavanja<\u002Fstrong> bez samostalnog hostovanja <strong>nivoa okvira<\u002Fstrong>. To je često pravi kompromis.\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-15\">Upravljani okvir: šta zapravo dobijate\u003C\u002Fh2>\n\u003Cp>Upravljani okvir uklanja više od same while petlje. OpenAI-jev trenutni Agents API upravlja sesijama, orkestracijom, sažimanjem konteksta i oporavkom. Rad kompanije Anthropic na upravljanim agentima opisuje istu širu motivaciju: okviri sadrže pretpostavke o ponašanju modela, a te pretpostavke moraju da evoluiraju kako se modeli poboljšavaju.\u003C\u002Fp>\n\u003Cp>To znači da korist nije samo u manjem broju linija koda. Platforma može da ažurira ponašanje u vremenu izvršavanja, upravljanje kontekstom na dugim horizontima, koordinaciju podagenata i oporavak, a da ne zahteva od svakog razvojnog tima da ponovo gradi te mehanizme.\u003C\u002Fp>\n\u003Cul>\u003Cli>Manje orkestracionog koda koji je u vlasništvu aplikacije.\u003C\u002Fli>\u003Cli>Upravljano ponašanje trajnih sesija.\u003C\u002Fli>\u003Cli>Upravljano sažimanje konteksta i oporavak.\u003C\u002Fli>\u003Cli>Runtime okruženje koje može da evoluira sa mogućnostima modela.\u003C\u002Fli>\u003Cli>Jednostavnije usvajanje funkcija podagenata i dugotrajnih agenata koje su izvorne za platformu.\u003C\u002Fli>\u003Cli>Potencijalno manje operativno opterećenje za timove čija diferencijacija nije sam okvir.\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch2 id=\"section-19\">Upravljano okruženje (harness): čega se odričete\u003C\u002Fh2>\n\u003Cp>Delegiranje okruženja takođe delegira i deo kontrole. Vaša aplikacija više ne poseduje svaki detalj iteracije, strategije konteksta, orkestracije i evolucije izvršnog okruženja. Ažuriranje platforme može poboljšati sistem, ali takođe može promeniti ponašanje od kojeg je vaš proizvod implicitno zavisio.\u003C\u002Fp>\n\u003Cp>Ovo stvara drugačiju vrstu inženjerskog zahteva: snažne evaluacije, eksplicitne granice proizvoda i integracioni sloj koji sprečava da ponašanje upravljane sesije postane vaš poslovni izvor istine.\u003C\u002Fp>\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\">Kompromis upravljanog okruženja\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Šta to operativno znači\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Manja kontrola petlje\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ne možete pretpostaviti da je svaki detalj orkestracije definisan aplikacijom\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Evolucija platforme\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ponašanje okruženja može se poboljšati ili promeniti bez promene vašeg koda\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Životni ciklus specifičan za dobavljača\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Sesije, događaji i semantika oporavka postaju deo integracione površine\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Granica opservabilnosti\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Tragovi platforme moraju se spojiti sa revizorskim podacima aplikacije\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Trošak portabilnosti\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Prelazak na drugo okruženje kasnije može zahtevati više od pukog menjanja krajnjih tačaka modela\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-23\">Samo-hostovano okruženje za izvršavanje: srednja arhitektura\u003C\u002Fh2>\n\u003Cp>OpenAI-jev model samo-hostovanog okruženja je važan jer razdvaja privatne računarske resurse od vlasništva nad okruženjem. Platforma i dalje pokreće Codex harness, dok se izvršilac pokreće unutar vašeg okruženja i prima komande preko odlazne veze.\u003C\u002Fp>\n\u003Cp>Vi kontrolišete obezbeđivanje resursa, fajlove, zavisnosti, mrežni pristup i čišćenje. Okruženje stoga može raditi sa privatnom infrastrukturom ili prilagođenim softverom bez potrebe da se celokupno izvršno okruženje agenta prebaci u vašu aplikaciju.\u003C\u002Fp>\n\u003Cp>Cena je odgovornost za životni ciklus. Vaša aplikacija mora mapirati sesije na računarske resurse, izbegavati duplicirano obezbeđivanje, ponovo povezivati okruženja, koordinirati gašenje i sačuvati sve fajlove koji moraju nadživeti okruženje.\u003C\u002Fp>\n\u003Ch3 id=\"section-27\">Kada je samo-hostovano izvršavanje dovoljno\u003C\u002Fh3>\n\u003Cul>\u003Cli>Agentu je potreban pristup privatnom VPC-u ili internom servisu.\u003C\u002Fli>\u003Cli>Agentu su potrebni prilagođeni binarni fajlovi, paketi, drajveri ili sistemski softver.\u003C\u002Fli>\u003Cli>Radno opterećenje mora da se izvršava na hardveru ili nalozima u oblaku koje vi kontrolišete.\u003C\u002Fli>\u003Cli>Fajlovi moraju ostati unutar kontrolisanog okruženja.\u003C\u002Fli>\u003Cli>Potreban vam je sopstveni sandbox provajder ili model izolacije.\u003C\u002Fli>\u003Cli>Želite orkestraciju kojom upravlja platforma, ali izvršavanje kontrolisano infrastrukturom.\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch2 id=\"section-29\">Kada ćete možda morati da posedujete i okruženje (harness)\u003C\u002Fh2>\n\u003Cp>Posedovanje okruženja postaje opravdano kada je samo okruženje deo diferencijacije vašeg proizvoda ili skupa ograničenja. OpenAI-jev trenutni pregled izvršnog okruženja pozicionira Codex SDK za pokretanje Codex okruženja u infrastrukturi kojom vi upravljate, dok je Responses opcija nižeg nivoa kada želite sami da posedujete agentsku petlju.\u003C\u002Fp>\n\u003Cp>Ključ je identifikovati zahtev koji suštinski pripada ravni okruženja (harness plane), a ne ravni izvršavanja (execution plane).\u003C\u002Fp>\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\">Zahtev\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Problem ravni izvršavanja ili problem ravni okruženja?\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Verovatan smer\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Pristup privatnoj bazi podataka\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ravan izvršavanja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Upravljano okruženje + samo-hostovano okruženje može biti dovoljno\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Prilagođeni Linux paketi\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ravan izvršavanja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Upravljano okruženje + samo-hostovano okruženje\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Prilagođeni GPU hardver\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ravan izvršavanja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Upravljano okruženje + samo-hostovano okruženje gde je podržano\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Prilagođena logika zaustavljanja agenta\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ravan okruženja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Samo-upravljano okruženje \u002F prilagođena petlja\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Rutiranje modela kroz više provajdera pri svakom koraku\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ravan okruženja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Prilagođena petlja ili okruženje kojim vi upravljate\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Prilagođeni algoritam za sažimanje konteksta\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ravan okruženja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Samo-upravljano okruženje ako upravljano izvršno okruženje to ne može da izloži\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Deterministička semantika orkestracije koju zahteva proizvod\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ravan okruženja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Samo-upravljano okruženje ili strogo kontrolisana prilagođena petlja\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Isključivo lokalno raspoređivanje proizvoda bez zavisnosti od upravljanog okruženja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ravan okruženja + ravan izvršavanja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Samo-upravljano izvršno okruženje\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-33\">Test eskalacije kontrole\u003C\u002Fh2>\n\u003Cp>Koristite najmanje samo-hostovanu arhitekturu koja zadovoljava stvarni zahtev. Eskalirajte kontrolu sloj po sloj.\u003C\u002Fp>\n\u003Csection class=\"editorjs-process my-6\">\u003Ch3 class=\"mb-3 text-lg font-semibold\">Test eskalacije kontrole\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. Počnite sa granicom aplikacije\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Zadržite istinu domena, autorizaciju i važne poslovne akcije u sopstvenom proizvodu, bez obzira na izvršno okruženje agenta.\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. Zapitajte se da li je agentu potrebno lokalno izvršavanje\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Ako nije, upravljano okruženje bez namenskog okruženja može biti dovoljno.\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. Zapitajte se da li su resursi hostovani na platformi prihvatljivi\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Ako jesu, koristite upravljano okruženje i izbegnite nepotrebno vlasništvo nad infrastrukturom.\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. Ako nisu, sami hostujte ravan izvršavanja\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Povežite sopstveno okruženje za privatnu mrežu, fajlove, pakete ili kontrolisane računarske resurse.\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. Ponovo procenite preostalo ograničenje\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Ako je zahtev sada zadovoljen, stanite. Nemojte sami hostovati okruženje (harness) samo radi arhitektonske simetrije.\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. Eskalirajte na vlasništvo nad okruženjem samo za zahteve okruženja\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Preuzmite vlasništvo nad Codex okruženjem ili prilagođenom agentskom petljom kada orkestracija, strategija konteksta, životni ciklus ili portabilnost to zaista zahtevaju.\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. Dokažite da dodatna kontrola vredi dodatnih operativnih troškova\u003C\u002Fdiv>\u003Cdiv class=\"mt-1 text-sm text-gray-600 dark:text-gray-300\">Uporedite pouzdanost, kašnjenje, troškove, oporavak, opservabilnost i inženjersko opterećenje pre nego što se obavežete.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-36\">Operativno opterećenje raste nelinearno kada posedujete okruženje (harness)\u003C\u002Fh2>\n\u003Cp>Petlja kojom sami upravljate zvuči jednostavno u demonstraciji: pozovi model, pregledaj poziv alatke, izvrši alatku, dodaj rezultat, ponovi. Produkcija dodaje trajno stanje, ponovne pokušaje, duplirane događaje, otkazivanje, odobravanje, prekoračenje konteksta, vremenska ograničenja alatki, ponovna pokretanja procesa, perzistenciju tragova, povratni pritisak, konkurentni rad i oporavak nakon delimičnih sporednih efekata.\u003C\u002Fp>\n\u003Cp>Istraživanja dugotrajnih agenata kompanije Anthropic iznova pokazuju da dizajn radnog okruženja (harness) značajno utiče na performanse. Njihov rad na razvoju dugotrajnih aplikacija koristi eksplicitno planiranje, strukturisane artefakte i evaluatorske agente jer naivne petlje obično gube napredak ili se prerano prekidaju. Radno okruženje je stoga produkciona logika, a ne samo prateća infrastruktura.\u003C\u002Fp>\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\">Ako sami upravljate radnim okruženjem, potreban vam je i odgovor na\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Zašto je to važno\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Trajno stanje sesije\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Procesi se ponovo pokreću; dugotrajni rad mora pravilno da se nastavi\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Sažimanje konteksta\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Istorija na kraju prevazilazi praktični radni kontekst\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Idempotentnost alatki\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ponovni pokušaji ne smeju ponavljati nepovratne sporedne efekte\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Otkazivanje i prekidanje\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Korisnicima i sistemima je potrebno da zaustave ili preusmere rad\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Oporavak nakon delimičnog izvršavanja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Alatka može uspeti čak i ako agent nikada ne primi rezultat\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Konkurentnost\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Više zadataka, radnika ili agenata može pristupati deljenom stanju\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Opservabilnost\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Konačni izlaz nije dovoljan za otklanjanje grešaka u izvršavanju\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Verzionisanje\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Ažuriranja radnog okruženja mogu promeniti ponašanje čak i kada upiti ostaju isti\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Evaluacija\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Promene u izvršnom okruženju zahtevaju regresiono testiranje kroz reprezentativne putanje\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-40\">Bezbednosna granica: samostalno hostovanje računarskih resursa ne čini agenta automatski privatnim\u003C\u002Fh2>\n\u003Cp>Samostalno hostovano okruženje za izvršavanje kontroliše gde se komande pokreću i gde se fajlovi nalaze, ali upravljano radno okruženje i interakcija sa modelom i dalje prelaze granicu servisa. Timovi bi zato trebalo eksplicitno da mapiraju tokove podataka umesto da koriste izraz „samostalno hostovano” kao sinonim za svojstvo privatnosti.\u003C\u002Fp>\n\u003Cp>OpenAI-jev samostalno hostovani izvršilac koristi ograničene akreditive okruženja i odlazne veze. To jeste korisna izolacija, ali vašoj aplikaciji su i dalje potrebna sopstvena pravila za tajne podatke, izloženost privatnoj mreži, izolaciju korisnika od okruženja, zadržavanje datoteka, autorizaciju alatki i klasifikaciju podataka.\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\">Važna razlika\u003C\u002Fstrong>\u003Cdiv class=\"text-gray-700 dark:text-gray-200\"><strong>Kontrola infrastrukture, izolacija izvršavanja i granice upravljanja podacima su povezane, ali nisu identične.<\u002Fstrong> Donosite odluke o njima zasebno.\u003C\u002Fdiv>\u003C\u002Faside>\n\u003Ch2 id=\"section-44\">Latencija i troškovi: kontrola može pomeriti uska grla umesto da ih ukloni\u003C\u002Fh2>\n\u003Cp>Samostalno hostovanje može smanjiti određene troškove putanje podataka ili pokretanja okruženja, ali takođe može povećati vreme raspoređivanja resursa, upravljanje životnim ciklusom WebSocket-a, hladna pokretanja, čišćenje sandbox okruženja, infrastrukturu za opservabilnost i inženjersko opterećenje. Upravljano okruženje može koštati više po jedinici računarskih resursa, a da istovremeno bude jeftinije za rad pri malom ili nepravilnom obimu posla.\u003C\u002Fp>\n\u003Cp>Pravo poređenje jeste ukupan trošak sistema: korišćenje modela i alatki, vreme okruženja, infrastruktura, inženjerski napor, dežurstva, oporavak od grešaka i trošak sporijih iteracija.\u003C\u002Fp>\n\u003Ch2 id=\"section-47\">Matrica odluka za produkciju\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\">Ograničenje\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Upravljano radno okruženje + upravljano okruženje\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Upravljano radno okruženje + samostalno hostovano okruženje\u003C\u002Fth>\u003Cth class=\"border border-gray-300 px-4 py-2 text-left font-semibold\">Radno okruženje \u002F petlja u sopstvenoj režiji\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Najbrži put do produkcije\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Jak\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Umeren\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Najslabiji\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Izvršavanje u privatnoj mreži\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Slabo \u002F zavisi od dizajna povezivanja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Jako\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Jako\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Prilagođeni paketi \u002F sistemski softver\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Umereno\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Jako\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Jako\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Kontrola na nivou radnog okruženja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Niska\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Niska\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Najviša\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Operativno opterećenje\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Najniže\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Srednje\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Najviše\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Portabilnost\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Najniža\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Srednja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Potencijalno najviša uz namensko projektovanje\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Kontrola strategije konteksta\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Upravlja platforma\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Upravlja platforma\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Kontroliše aplikacija\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Kontrola infrastrukture za izvršavanje\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Niska\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Visoka\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Visoka\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Mogućnost korišćenja ažuriranja upravljanog okruženja\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Najviša\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Najviša\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Sami ste odgovorni za usvajanje\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Najbolji izbor\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Timovi koji se diferenciraju na nivou proizvoda\u002Falatki\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Timovi kojima su potrebni privatni\u002Fprilagođeni računarski resursi bez vođenja orkestracije\u003C\u002Ftd>\u003Ctd class=\"border border-gray-300 px-4 py-2\">Timovi čija je semantika izvršnog okruženja sama po sebi zahtev\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"section-49\">Hibridni model nije kompromis — to je često čista arhitektura\u003C\u002Fh2>\n\u003Cp>Upravljano radno okruženje sa samostalno hostovanim izvršavanjem nije „polovično samostalno hostovano”. To je promišljeno razdvajanje odgovornosti. Platforma preuzima složenost dugotrajnog izvršavanja agenata, dok vaša infrastruktura upravlja izvršavanjem, privatnom povezanošću i datotekama.\u003C\u002Fp>\n\u003Cp>Ta granica podseća na druge cloud arhitekture: upravljana kontrolna ravan, ravan podataka ili izvršavanja pod kontrolom korisnika. Važan projektantski zadatak jeste definisanje ugovora između njih — identitet sesije, identitet okruženja, akreditivi, datoteke, dozvole za alatke, događaji životnog ciklusa i čišćenje resursa.\u003C\u002Fp>\n\u003Caside class=\"editorjs-referral my-6\">\u003Ca href=\"https:\u002F\u002Fstajic.de\u002Fsr\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026\" 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\">OpenAI Agents API vs Agents SDK vs Responses API: Na čemu treba graditi u 2026. godini?\u003C\u002Fstrong>\u003Cp class=\"mt-2 text-sm text-gray-600 dark:text-gray-300\">Poređenje vlasništva nad izvršnim okruženjem u trenutnim OpenAI platformama za agente i raspodela kontrolnih granica.\u003C\u002Fp>\u003Cspan class=\"mt-3 inline-flex text-sm font-medium text-primary-600 dark:text-primary-400\">Pročitajte poređenje izvršnih okruženja →\u003C\u002Fspan>\u003C\u002Fdiv>\u003C\u002Fa>\u003C\u002Faside>\n\u003Ch2 id=\"section-53\">Šta bi promenilo ovaj odgovor?\u003C\u002Fh2>\n\u003Cp>Preporuka se menja ukoliko upravljana radna okruženja ponude znatno veću kontrolu nad izvršavanjem, ukoliko samostalno hostovana okruženja dobiju jednostavnije primitive za trajne sesije i oporavak, ili ukoliko propisi zahtevaju da čitava agentska petlja i interakcija sa modelom ostanu unutar infrastrukture kojom sami upravljate.\u003C\u002Fp>\n\u003Cp>Takođe se menja sa sposobnostima modela. Anthropic eksplicitno napominje da pretpostavke o radnom okviru (harness) mogu zastareti kako se modeli poboljšavaju. Kontrolni mehanizam koji je danas od suštinskog značaja može kasnije postati nepotreban, dok nova sposobnost modela može stvoriti novi zahtev za upravljanjem.\u003C\u002Fp>\n\u003Ch2 id=\"section-56\">Ograničenja\u003C\u002Fh2>\n\u003Cp>Ovaj članak razdvaja arhitektonske odgovornosti; on ne tvrdi da je jedan model hostovanja univerzalno bezbedniji, jeftiniji ili pouzdaniji. Ti ishodi zavise od implementacije, radnog opterećenja, zahteva usklađenosti, stručnosti tima i ponašanja pružaoca usluga.\u003C\u002Fp>\n\u003Cp>OpenAI Agents API je još uvek u javnoj beta verziji, a proizvodi upravljanih agenata različitih dobavljača izlažu različite granice. Model sa dve ravni ima za cilj da pomogne u poređenju tih arhitektura bez pretpostavke da svaki dobavljač koristi identične termine.\u003C\u002Fp>\n\u003Ch2 id=\"section-59\">Zaključak\u003C\u002Fh2>\n\u003Cp>Korisno pitanje nije „Da li treba sami da hostujemo agenta?” Već glasi: Koju ravan zapravo moramo da kontrolišemo?\u003C\u002Fp>\n\u003Cp>Ako je zahtev privatno računarsko okruženje, prilagođeni paketi, lokalni fajlovi ili pristup internoj mreži, sami hostujte ravan izvršavanja, a radni okvir zadržite kao upravljan. Ako su zahtev semantika orkestracije, strategija konteksta, kontrola nad pružaocem usluga ili sam životni ciklus izvršavanja, onda preuzimanje vlasništva nad radnim okvirom može biti opravdano. Eskalirajte kontrolu samo onoliko koliko to zahtev nalaže.\u003C\u002Fp>\n\u003Ch2 id=\"section-62\">Često postavljana pitanja\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\">Upravljani radni okviri i samostalno hostovana okruženja za izvršavanje agenata\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\">Da li je samostalno hostovano okruženje za OpenAI Agents API ujedno i samostalno hostovan agent?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">Ne u potpunosti. OpenAI i dalje pokreće upravljani Codex radni okvir, dok vaša infrastruktura pokreće okruženje za izvršavanje koje se koristi za komande, fajlove i lokalne alate.\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\">Kada je samostalno hostovano okruženje dovoljno?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">Često je dovoljno kada se vaši zahtevi odnose na pristup privatnoj mreži, prilagođene pakete, kontrolisane fajlove, specifičan hardver ili infrastrukturne politike, a ne na kontrolu nad samom petljom agenta.\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\">Kada bi trebalo da samostalno pokrećem radni okvir?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">Razmotrite preuzimanje vlasništva nad radnim okvirom kada su vam potrebne prilagođena semantika orkestracije, prilagođeno upravljanje kontekstom, usmeravanje pružaoca usluga, ponašanje pri izvršavanju isključivo lokalno, ili neki drugi zahtev koji pripada petlji agenta, a ne okruženju za izvršavanje.\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\">Da li samostalno hostovanje automatski poboljšava bezbednost?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">Ne. Ono menja komponente koje kontrolišete. Bezbednost zavisi od protoka podataka, izolacije, akreditiva, dozvola za alate, umrežavanja, evidentiranja i dizajna životnog ciklusa kroz sve komponente.\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\">Koji je glavni operativni trošak posedovanja petlje agenta?\u003C\u002Fh4>\u003Cdiv class=\"mt-2 text-gray-600 dark:text-gray-300\">Postajete odgovorni za trajno stanje, upravljanje kontekstom, ponovne pokušaje, otkazivanje, oporavak, opservabilnost, konkurentnost, nadogradnje okruženja za izvršavanje i evaluaciju promena u radnom okviru.\u003C\u002Fdiv>\u003C\u002Fdiv>\u003C\u002Fsection>\n\u003Ch2 id=\"section-64\">Rečnik pojmova\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\">Ključni arhitektonski termini\u003C\u002Fh3>\u003Cdl>\u003Cdiv id=\"harness-plane\" 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\">Ravan radnog okvira (Harness plane)\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">Sloj izvršavanja agenta odgovoran za izvođenje petlje, orkestraciju, upravljanje kontekstom, kontinuitet sesije i oporavak.\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"execution-plane\" 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\">Ravan izvršavanja (Execution plane)\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">Okruženje u kojem se izvršavaju komande, pokreće kôd i pristupa fajlovima, paketima i lokalnim resursima.\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"managed-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\">Upravljani radni okvir (Managed harness)\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">Radni okvir agenta čijim izvršnim okruženjem, upravljanjem sesijama i orkestracijom upravlja pružalac platforme.\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"self-hosted-environment\" 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\">Samostalno hostovano okruženje (Self-hosted environment)\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">Računarski resursi i fajlovi kojima upravlja vlasnik aplikacije, dok zasebni radni okvir agenta može ostati pod upravljanjem negde drugde.\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"self-operated-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\">Samostalno vođen radni okvir (Self-operated harness)\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">Izvršno okruženje agenta čijom petljom, hostovanjem, strategijom konteksta i životnim ciklusom upravlja tim aplikacije.\u003C\u002Fdd>\u003C\u002Fdiv>\u003Cdiv id=\"control-escalation-test\" 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\">Test eskalacije kontrole (Control Escalation Test)\u003C\u002Fdt>\u003Cdd class=\"mt-1 text-gray-600 dark:text-gray-300\">Metoda odlučivanja koja povećava vlasništvo nad infrastrukturom i okruženjem za izvršavanje samo kada se zahtev ne može zadovoljiti na sloju niže kontrole.\u003C\u002Fdd>\u003C\u002Fdiv>\u003C\u002Fdl>\u003C\u002Fsection>\n\u003Ch2 id=\"section-66\">Primarni izvori i preporučena literatura\u003C\u002Fh2>\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 — Arhitektura Agents API-ja\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Trenutno razdvajanje između hostovanog radnog okvira, okruženja za izvršavanje i aplikativnog servera.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Fenvironments\u002Fself-hosted\" 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 — Samostalno hostovana izolovana okruženja (sandboxes)\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Kako se okruženja za izvršavanje kojima upravlja korisnik povezuju sa upravljanim radnim okvirom i koje odgovornosti za životni ciklus ostaju na aplikaciji.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Fenvironments\u002Flifecycle\" 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 — Životni ciklus izolovanog okruženja (sandbox)\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Priprema resursa, ponovno povezivanje, sprečavanje dupliranja okruženja i odgovornosti za čišćenje za samostalno hostovane računarske resurse.\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 — Opcije okruženja za izvršavanje agenata\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Trenutno poređenje Agents API-ja, Codex SDK-a i Responses API-ja prema odgovornostima kojima upravlja platforma u odnosu na aplikaciju.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fdevelopers.openai.com\u002Fblog\u002Fcodex-as-a-platform\" 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 — Codex kao platforma\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Codex radni okvir otvorenog koda i integracioni slojevi za aplikacije koje zahtevaju dublju kontrolu izvršavanja.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fmanaged-agents\" 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\">Anthropic — Skaliranje upravljanih agenata: Odvajanje mozga od ruku\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Diskusija o arhitekturi upravljanih agenata i zašto pretpostavke o radnom okviru moraju da evoluiraju u skladu sa sposobnostima modela.\u003C\u002Fp>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Feffective-harnesses-for-long-running-agents\" 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\">Anthropic — Efikasna radna okruženja za dugotrajne agente\u003C\u002Fstrong>\u003Cp class=\"text-sm text-gray-600 dark:text-gray-400\">Inženjerske lekcije koje pokazuju da performanse dugotrajnih agenata suštinski zavise od dizajna radnog okruženja i trajnih artefakata.\u003C\u002Fp>\u003C\u002Fa>",{"time":212,"blocks":213,"version":880},1790365416705,[214,222,228,236,243,249,254,259,264,269,274,299,304,334,341,346,351,356,370,375,380,385,408,413,418,423,428,433,445,450,455,460,496,501,506,534,539,544,549,584,589,594,599,605,610,615,620,625,676,681,686,691,700,705,710,715,720,725,730,735,740,745,750,776,781,811,816,826,835,844,853,862,871],{"id":215,"data":216,"type":220,"tunes":221},"-PDI7SJcNl",{"title":217,"maxLevel":218,"minLevel":219},"Sadržaj",3,2,"tableOfContents",{},{"id":223,"data":224,"type":226,"tunes":227},"intro",{"text":225},"Izraz „samostalno hostovani agent” (self-hosted agent) danas krije najmanje tri različite arhitekture. Možete koristiti upravljani okvir (managed harness) sa računarskim resursima koje hostuje OpenAI, upravljani okvir povezan sa infrastrukturom kojom sami upravljate, ili sami pokretati i okvir i petlju agenta. Ovi izbori imaju veoma različite implikacije po pitanju kontrole, oporavka, upravljanja kontekstom, bezbednosti, kašnjenja i operativnog opterećenja.","paragraph",{},{"id":229,"data":230,"type":234,"tunes":235},"direct",{"body":231,"title":232,"variant":233},"\u003Cstrong>Nemojte birati između „upravljanog” i „samostalno hostovanog” kao da je to binarna odluka.\u003C\u002Fstrong> Razdvojite \u003Cstrong>nivo okvira (harness plane)\u003C\u002Fstrong> od \u003Cstrong>nivoa izvršavanja (execution plane)\u003C\u002Fstrong>. Upravljani okvir i dalje može koristiti samostalno hostovane računarske resurse. Samostalno hostovano okruženje vam pruža kontrolu nad datotekama, paketima, mrežnim pristupom i izvršavanjem, a da pritom ne zahteva od vas da posedujete samu petlju agenta. Pokrećite okvir sami samo onda kada vam je potrebna kontrola nad orkestracijom, životnim ciklusom, pretpostavkama o usmeravanju modela ili ponašanjem u vremenu izvršavanja koje upravljani okvir ne može da pruži.","Direktan odgovor","info","callout",{},{"id":237,"data":238,"type":234,"tunes":242},"current",{"body":239,"title":240,"variant":241},"OpenAI Agents API je u javnoj beta verziji i njegova arhitektura može evoluirati. Trenutna dokumentacija razdvaja Codex okvir koji hostuje OpenAI od okruženja za izvršavanje i eksplicitno podržava samostalno hostovana okruženja. OpenAI takođe nudi Codex okvir zasebno putem Codex SDK-a za infrastrukturu kojom sami upravljate.","Aktuelno na dan 25. septembra 2026.","warning",{},{"id":244,"data":245,"type":234,"tunes":248},"note",{"body":246,"title":247,"variant":244},"Model nivoa okvira \u002F nivoa izvršavanja i Test eskalacije kontrole u nastavku predstavljaju praktične arhitektonske alate predložene ovde. Oni ne predstavljaju formalnu terminologiju vendora.","Model korišćen u ovom članku",{},{"id":250,"data":251,"type":42,"tunes":253},"h-mistake",{"text":252,"level":219},"Greška: tretiranje samostalnog hostovanja kao jedne odluke",{},{"id":255,"data":256,"type":226,"tunes":258},"p-mistake-1",{"text":257},"U konvencionalnom softveru, „samostalno hostovano” obično znači da se aplikacija izvršava na infrastrukturi koju sami kontrolišete. Sistemi agenata komplikuju tu definiciju jer se okruženje za izvršavanje može podeliti. Petlja modela i alata može se izvršavati na jednom mestu, dok se izvršavanje koda, rad sa datotekama i pristup privatnoj mreži odvijaju negde drugde.",{},{"id":260,"data":261,"type":226,"tunes":263},"p-mistake-2",{"text":262},"Trenutna arhitektura OpenAI Agents API-ja čini ovu podelu eksplicitnom: OpenAI pokreće okvir, dok okruženje za izvršavanje može biti odsutno, hostovano od strane OpenAI-ja ili samostalno hostovano. Samostalno hostovano okruženje, dakle, ne znači da je i petlja agenta samostalno hostovana.",{},{"id":265,"data":266,"type":226,"tunes":268},"p-mistake-3",{"text":267},"Ova razlika je važna jer mnogi timovi biraju složenije runtime okruženje nego što im je potrebno. Žele pristup privatnoj mreži ili prilagođene pakete, zaključe da ceo agent mora biti samostalno hostovan i slučajno preuzmu vlasništvo nad upravljanjem kontekstom, orkestracijom, oporavkom i životnim ciklusom, što je sve moglo ostati pod upravljanjem platforme.",{},{"id":270,"data":271,"type":42,"tunes":273},"h-three",{"text":272,"level":219},"Tri arhitekture koje se često nazivaju „samostalno hostovanim”",{},{"id":275,"data":276,"type":297,"tunes":298},"three-table",{"content":277,"stretched":43,"withHeadings":14},[278,283,288,292],[279,280,281,282],"Arhitektura","Ko pokreće okvir?","Gde se izvršavaju kod\u002Fdatoteke","Šta primarno posedujete",[284,285,286,287],"Upravljani okvir + upravljano okruženje","Platforma","Sandbox okruženje koje hostuje platforma","Aplikacija, alati, logika proizvoda, autorizacija",[289,285,290,291],"Upravljani okvir + samostalno hostovano okruženje","Vaš kontejner, VM, laptop, privatni oblak ili drugi računarski resursi","Obezbeđivanje okruženja, umrežavanje, datoteke i životni ciklus; platforma i dalje poseduje okvir",[293,294,295,296],"Okvir \u002F petlja agenta pod sopstvenom upravom","Vi","Okruženje po vašem izboru","Proces okvira, orkestracija, strategija konteksta, hosting, oporavak, izvršavanje i životni ciklus aplikacije","table",{},{"id":300,"data":301,"type":42,"tunes":303},"h-two-plane",{"text":302,"level":219},"Model sa dva nivoa",{},{"id":305,"data":306,"type":332,"tunes":333},"plane-comparison",{"rows":307,"title":321,"layout":297,"columns":322},[308,313,317],{"id":309,"label":310,"values":311},"harness","Nivo okvira (Harness plane)",[312,312,312],"",{"id":314,"label":315,"values":316},"execution","Nivo izvršavanja (Execution plane)",[312,312,312],{"id":318,"label":319,"values":320},"application","Nivo aplikacije (Application plane)",[312,312,312],"Razdvojite nivo okvira od nivoa izvršavanja",[323,326,329],{"id":324,"label":325},"plane","Nivo",{"id":327,"label":328},"owns","Šta poseduje",{"id":330,"label":331},"questions","Pitanja koja treba postaviti","comparison",{},{"id":335,"data":336,"type":234,"tunes":340},"key-consequence",{"body":337,"title":338,"variant":339},"Možete samostalno hostovati \u003Cstrong>nivo izvršavanja\u003C\u002Fstrong> bez samostalnog hostovanja \u003Cstrong>nivoa okvira\u003C\u002Fstrong>. To je često pravi kompromis.","Ključna arhitektonska posledica","success",{},{"id":342,"data":343,"type":42,"tunes":345},"h-managed",{"text":344,"level":219},"Upravljani okvir: šta zapravo dobijate",{},{"id":347,"data":348,"type":226,"tunes":350},"p-managed-1",{"text":349},"Upravljani okvir uklanja više od same while petlje. OpenAI-jev trenutni Agents API upravlja sesijama, orkestracijom, sažimanjem konteksta i oporavkom. Rad kompanije Anthropic na upravljanim agentima opisuje istu širu motivaciju: okviri sadrže pretpostavke o ponašanju modela, a te pretpostavke moraju da evoluiraju kako se modeli poboljšavaju.",{},{"id":352,"data":353,"type":226,"tunes":355},"p-managed-2",{"text":354},"To znači da korist nije samo u manjem broju linija koda. Platforma može da ažurira ponašanje u vremenu izvršavanja, upravljanje kontekstom na dugim horizontima, koordinaciju podagenata i oporavak, a da ne zahteva od svakog razvojnog tima da ponovo gradi te mehanizme.",{},{"id":357,"data":358,"type":368,"tunes":369},"managed-list",{"meta":359,"items":360,"style":367},{},[361,362,363,364,365,366],"Manje orkestracionog koda koji je u vlasništvu aplikacije.","Upravljano ponašanje trajnih sesija.","Upravljano sažimanje konteksta i oporavak.","Runtime okruženje koje može da evoluira sa mogućnostima modela.","Jednostavnije usvajanje funkcija podagenata i dugotrajnih agenata koje su izvorne za platformu.","Potencijalno manje operativno opterećenje za timove čija diferencijacija nije sam okvir.","unordered","list",{},{"id":371,"data":372,"type":42,"tunes":374},"h-managed-cost",{"text":373,"level":219},"Upravljano okruženje (harness): čega se odričete",{},{"id":376,"data":377,"type":226,"tunes":379},"p-managed-cost-1",{"text":378},"Delegiranje okruženja takođe delegira i deo kontrole. Vaša aplikacija više ne poseduje svaki detalj iteracije, strategije konteksta, orkestracije i evolucije izvršnog okruženja. Ažuriranje platforme može poboljšati sistem, ali takođe može promeniti ponašanje od kojeg je vaš proizvod implicitno zavisio.",{},{"id":381,"data":382,"type":226,"tunes":384},"p-managed-cost-2",{"text":383},"Ovo stvara drugačiju vrstu inženjerskog zahteva: snažne evaluacije, eksplicitne granice proizvoda i integracioni sloj koji sprečava da ponašanje upravljane sesije postane vaš poslovni izvor istine.",{},{"id":386,"data":387,"type":297,"tunes":407},"managed-cost-table",{"content":388,"stretched":43,"withHeadings":14},[389,392,395,398,401,404],[390,391],"Kompromis upravljanog okruženja","Šta to operativno znači",[393,394],"Manja kontrola petlje","Ne možete pretpostaviti da je svaki detalj orkestracije definisan aplikacijom",[396,397],"Evolucija platforme","Ponašanje okruženja može se poboljšati ili promeniti bez promene vašeg koda",[399,400],"Životni ciklus specifičan za dobavljača","Sesije, događaji i semantika oporavka postaju deo integracione površine",[402,403],"Granica opservabilnosti","Tragovi platforme moraju se spojiti sa revizorskim podacima aplikacije",[405,406],"Trošak portabilnosti","Prelazak na drugo okruženje kasnije može zahtevati više od pukog menjanja krajnjih tačaka modela",{},{"id":409,"data":410,"type":42,"tunes":412},"h-self-env",{"text":411,"level":219},"Samo-hostovano okruženje za izvršavanje: srednja arhitektura",{},{"id":414,"data":415,"type":226,"tunes":417},"p-self-env-1",{"text":416},"OpenAI-jev model samo-hostovanog okruženja je važan jer razdvaja privatne računarske resurse od vlasništva nad okruženjem. Platforma i dalje pokreće Codex harness, dok se izvršilac pokreće unutar vašeg okruženja i prima komande preko odlazne veze.",{},{"id":419,"data":420,"type":226,"tunes":422},"p-self-env-2",{"text":421},"Vi kontrolišete obezbeđivanje resursa, fajlove, zavisnosti, mrežni pristup i čišćenje. Okruženje stoga može raditi sa privatnom infrastrukturom ili prilagođenim softverom bez potrebe da se celokupno izvršno okruženje agenta prebaci u vašu aplikaciju.",{},{"id":424,"data":425,"type":226,"tunes":427},"p-self-env-3",{"text":426},"Cena je odgovornost za životni ciklus. Vaša aplikacija mora mapirati sesije na računarske resurse, izbegavati duplicirano obezbeđivanje, ponovo povezivati okruženja, koordinirati gašenje i sačuvati sve fajlove koji moraju nadživeti okruženje.",{},{"id":429,"data":430,"type":42,"tunes":432},"h-enough",{"text":431,"level":218},"Kada je samo-hostovano izvršavanje dovoljno",{},{"id":434,"data":435,"type":368,"tunes":444},"enough-list",{"meta":436,"items":437,"style":367},{},[438,439,440,441,442,443],"Agentu je potreban pristup privatnom VPC-u ili internom servisu.","Agentu su potrebni prilagođeni binarni fajlovi, paketi, drajveri ili sistemski softver.","Radno opterećenje mora da se izvršava na hardveru ili nalozima u oblaku koje vi kontrolišete.","Fajlovi moraju ostati unutar kontrolisanog okruženja.","Potreban vam je sopstveni sandbox provajder ili model izolacije.","Želite orkestraciju kojom upravlja platforma, ali izvršavanje kontrolisano infrastrukturom.",{},{"id":446,"data":447,"type":42,"tunes":449},"h-own-harness",{"text":448,"level":219},"Kada ćete možda morati da posedujete i okruženje (harness)",{},{"id":451,"data":452,"type":226,"tunes":454},"p-own-1",{"text":453},"Posedovanje okruženja postaje opravdano kada je samo okruženje deo diferencijacije vašeg proizvoda ili skupa ograničenja. OpenAI-jev trenutni pregled izvršnog okruženja pozicionira Codex SDK za pokretanje Codex okruženja u infrastrukturi kojom vi upravljate, dok je Responses opcija nižeg nivoa kada želite sami da posedujete agentsku petlju.",{},{"id":456,"data":457,"type":226,"tunes":459},"p-own-2",{"text":458},"Ključ je identifikovati zahtev koji suštinski pripada ravni okruženja (harness plane), a ne ravni izvršavanja (execution plane).",{},{"id":461,"data":462,"type":297,"tunes":495},"requirements-table",{"content":463,"stretched":43,"withHeadings":14},[464,468,472,475,478,482,485,488,491],[465,466,467],"Zahtev","Problem ravni izvršavanja ili problem ravni okruženja?","Verovatan smer",[469,470,471],"Pristup privatnoj bazi podataka","Ravan izvršavanja","Upravljano okruženje + samo-hostovano okruženje može biti dovoljno",[473,470,474],"Prilagođeni Linux paketi","Upravljano okruženje + samo-hostovano okruženje",[476,470,477],"Prilagođeni GPU hardver","Upravljano okruženje + samo-hostovano okruženje gde je podržano",[479,480,481],"Prilagođena logika zaustavljanja agenta","Ravan okruženja","Samo-upravljano okruženje \u002F prilagođena petlja",[483,480,484],"Rutiranje modela kroz više provajdera pri svakom koraku","Prilagođena petlja ili okruženje kojim vi upravljate",[486,480,487],"Prilagođeni algoritam za sažimanje konteksta","Samo-upravljano okruženje ako upravljano izvršno okruženje to ne može da izloži",[489,480,490],"Deterministička semantika orkestracije koju zahteva proizvod","Samo-upravljano okruženje ili strogo kontrolisana prilagođena petlja",[492,493,494],"Isključivo lokalno raspoređivanje proizvoda bez zavisnosti od upravljanog okruženja","Ravan okruženja + ravan izvršavanja","Samo-upravljano izvršno okruženje",{},{"id":497,"data":498,"type":42,"tunes":500},"h-control-test",{"text":499,"level":219},"Test eskalacije kontrole",{},{"id":502,"data":503,"type":226,"tunes":505},"p-control-intro",{"text":504},"Koristite najmanje samo-hostovanu arhitekturu koja zadovoljava stvarni zahtev. Eskalirajte kontrolu sloj po sloj.",{},{"id":507,"data":508,"type":532,"tunes":533},"control-flow",{"steps":509,"title":499,"orientation":531},[510,513,516,519,522,525,528],{"label":511,"description":512},"1. Počnite sa granicom aplikacije","Zadržite istinu domena, autorizaciju i važne poslovne akcije u sopstvenom proizvodu, bez obzira na izvršno okruženje agenta.",{"label":514,"description":515},"2. Zapitajte se da li je agentu potrebno lokalno izvršavanje","Ako nije, upravljano okruženje bez namenskog okruženja može biti dovoljno.",{"label":517,"description":518},"3. Zapitajte se da li su resursi hostovani na platformi prihvatljivi","Ako jesu, koristite upravljano okruženje i izbegnite nepotrebno vlasništvo nad infrastrukturom.",{"label":520,"description":521},"4. Ako nisu, sami hostujte ravan izvršavanja","Povežite sopstveno okruženje za privatnu mrežu, fajlove, pakete ili kontrolisane računarske resurse.",{"label":523,"description":524},"5. Ponovo procenite preostalo ograničenje","Ako je zahtev sada zadovoljen, stanite. Nemojte sami hostovati okruženje (harness) samo radi arhitektonske simetrije.",{"label":526,"description":527},"6. Eskalirajte na vlasništvo nad okruženjem samo za zahteve okruženja","Preuzmite vlasništvo nad Codex okruženjem ili prilagođenom agentskom petljom kada orkestracija, strategija konteksta, životni ciklus ili portabilnost to zaista zahtevaju.",{"label":529,"description":530},"7. Dokažite da dodatna kontrola vredi dodatnih operativnih troškova","Uporedite pouzdanost, kašnjenje, troškove, oporavak, opservabilnost i inženjersko opterećenje pre nego što se obavežete.","auto","processFlow",{},{"id":535,"data":536,"type":42,"tunes":538},"h-ops",{"text":537,"level":219},"Operativno opterećenje raste nelinearno kada posedujete okruženje (harness)",{},{"id":540,"data":541,"type":226,"tunes":543},"p-ops-1",{"text":542},"Petlja kojom sami upravljate zvuči jednostavno u demonstraciji: pozovi model, pregledaj poziv alatke, izvrši alatku, dodaj rezultat, ponovi. Produkcija dodaje trajno stanje, ponovne pokušaje, duplirane događaje, otkazivanje, odobravanje, prekoračenje konteksta, vremenska ograničenja alatki, ponovna pokretanja procesa, perzistenciju tragova, povratni pritisak, konkurentni rad i oporavak nakon delimičnih sporednih efekata.",{},{"id":545,"data":546,"type":226,"tunes":548},"p-ops-2",{"text":547},"Istraživanja dugotrajnih agenata kompanije Anthropic iznova pokazuju da dizajn radnog okruženja (harness) značajno utiče na performanse. Njihov rad na razvoju dugotrajnih aplikacija koristi eksplicitno planiranje, strukturisane artefakte i evaluatorske agente jer naivne petlje obično gube napredak ili se prerano prekidaju. Radno okruženje je stoga produkciona logika, a ne samo prateća infrastruktura.",{},{"id":550,"data":551,"type":297,"tunes":583},"ops-table",{"content":552,"stretched":43,"withHeadings":14},[553,556,559,562,565,568,571,574,577,580],[554,555],"Ako sami upravljate radnim okruženjem, potreban vam je i odgovor na","Zašto je to važno",[557,558],"Trajno stanje sesije","Procesi se ponovo pokreću; dugotrajni rad mora pravilno da se nastavi",[560,561],"Sažimanje konteksta","Istorija na kraju prevazilazi praktični radni kontekst",[563,564],"Idempotentnost alatki","Ponovni pokušaji ne smeju ponavljati nepovratne sporedne efekte",[566,567],"Otkazivanje i prekidanje","Korisnicima i sistemima je potrebno da zaustave ili preusmere rad",[569,570],"Oporavak nakon delimičnog izvršavanja","Alatka može uspeti čak i ako agent nikada ne primi rezultat",[572,573],"Konkurentnost","Više zadataka, radnika ili agenata može pristupati deljenom stanju",[575,576],"Opservabilnost","Konačni izlaz nije dovoljan za otklanjanje grešaka u izvršavanju",[578,579],"Verzionisanje","Ažuriranja radnog okruženja mogu promeniti ponašanje čak i kada upiti ostaju isti",[581,582],"Evaluacija","Promene u izvršnom okruženju zahtevaju regresiono testiranje kroz reprezentativne putanje",{},{"id":585,"data":586,"type":42,"tunes":588},"h-security",{"text":587,"level":219},"Bezbednosna granica: samostalno hostovanje računarskih resursa ne čini agenta automatski privatnim",{},{"id":590,"data":591,"type":226,"tunes":593},"p-sec-1",{"text":592},"Samostalno hostovano okruženje za izvršavanje kontroliše gde se komande pokreću i gde se fajlovi nalaze, ali upravljano radno okruženje i interakcija sa modelom i dalje prelaze granicu servisa. Timovi bi zato trebalo eksplicitno da mapiraju tokove podataka umesto da koriste izraz „samostalno hostovano” kao sinonim za svojstvo privatnosti.",{},{"id":595,"data":596,"type":226,"tunes":598},"p-sec-2",{"text":597},"OpenAI-jev samostalno hostovani izvršilac koristi ograničene akreditive okruženja i odlazne veze. To jeste korisna izolacija, ali vašoj aplikaciji su i dalje potrebna sopstvena pravila za tajne podatke, izloženost privatnoj mreži, izolaciju korisnika od okruženja, zadržavanje datoteka, autorizaciju alatki i klasifikaciju podataka.",{},{"id":600,"data":601,"type":234,"tunes":604},"sec-callout",{"body":602,"title":603,"variant":241},"\u003Cstrong>Kontrola infrastrukture, izolacija izvršavanja i granice upravljanja podacima su povezane, ali nisu identične.\u003C\u002Fstrong> Donosite odluke o njima zasebno.","Važna razlika",{},{"id":606,"data":607,"type":42,"tunes":609},"h-cost",{"text":608,"level":219},"Latencija i troškovi: kontrola može pomeriti uska grla umesto da ih ukloni",{},{"id":611,"data":612,"type":226,"tunes":614},"p-cost-1",{"text":613},"Samostalno hostovanje može smanjiti određene troškove putanje podataka ili pokretanja okruženja, ali takođe može povećati vreme raspoređivanja resursa, upravljanje životnim ciklusom WebSocket-a, hladna pokretanja, čišćenje sandbox okruženja, infrastrukturu za opservabilnost i inženjersko opterećenje. Upravljano okruženje može koštati više po jedinici računarskih resursa, a da istovremeno bude jeftinije za rad pri malom ili nepravilnom obimu posla.",{},{"id":616,"data":617,"type":226,"tunes":619},"p-cost-2",{"text":618},"Pravo poređenje jeste ukupan trošak sistema: korišćenje modela i alatki, vreme okruženja, infrastruktura, inženjerski napor, dežurstva, oporavak od grešaka i trošak sporijih iteracija.",{},{"id":621,"data":622,"type":42,"tunes":624},"h-matrix",{"text":623,"level":219},"Matrica odluka za produkciju",{},{"id":626,"data":627,"type":297,"tunes":675},"decision-matrix",{"content":628,"stretched":43,"withHeadings":14},[629,634,639,643,646,650,655,660,664,667,670],[630,631,632,633],"Ograničenje","Upravljano radno okruženje + upravljano okruženje","Upravljano radno okruženje + samostalno hostovano okruženje","Radno okruženje \u002F petlja u sopstvenoj režiji",[635,636,637,638],"Najbrži put do produkcije","Jak","Umeren","Najslabiji",[640,641,642,642],"Izvršavanje u privatnoj mreži","Slabo \u002F zavisi od dizajna povezivanja","Jako",[644,645,642,642],"Prilagođeni paketi \u002F sistemski softver","Umereno",[647,648,648,649],"Kontrola na nivou radnog okruženja","Niska","Najviša",[651,652,653,654],"Operativno opterećenje","Najniže","Srednje","Najviše",[656,657,658,659],"Portabilnost","Najniža","Srednja","Potencijalno najviša uz namensko projektovanje",[661,662,662,663],"Kontrola strategije konteksta","Upravlja platforma","Kontroliše aplikacija",[665,648,666,666],"Kontrola infrastrukture za izvršavanje","Visoka",[668,649,649,669],"Mogućnost korišćenja ažuriranja upravljanog okruženja","Sami ste odgovorni za usvajanje",[671,672,673,674],"Najbolji izbor","Timovi koji se diferenciraju na nivou proizvoda\u002Falatki","Timovi kojima su potrebni privatni\u002Fprilagođeni računarski resursi bez vođenja orkestracije","Timovi čija je semantika izvršnog okruženja sama po sebi zahtev",{},{"id":677,"data":678,"type":42,"tunes":680},"h-hybrid",{"text":679,"level":219},"Hibridni model nije kompromis — to je često čista arhitektura",{},{"id":682,"data":683,"type":226,"tunes":685},"p-hybrid-1",{"text":684},"Upravljano radno okruženje sa samostalno hostovanim izvršavanjem nije „polovično samostalno hostovano”. To je promišljeno razdvajanje odgovornosti. Platforma preuzima složenost dugotrajnog izvršavanja agenata, dok vaša infrastruktura upravlja izvršavanjem, privatnom povezanošću i datotekama.",{},{"id":687,"data":688,"type":226,"tunes":690},"p-hybrid-2",{"text":689},"Ta granica podseća na druge cloud arhitekture: upravljana kontrolna ravan, ravan podataka ili izvršavanja pod kontrolom korisnika. Važan projektantski zadatak jeste definisanje ugovora između njih — identitet sesije, identitet okruženja, akreditivi, datoteke, dozvole za alatke, događaji životnog ciklusa i čišćenje resursa.",{},{"id":692,"data":693,"type":698,"tunes":699},"ref-runtime",{"url":694,"title":695,"excerpt":696,"ctaLabel":697},"https:\u002F\u002Fstajic.de\u002Fsr\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","OpenAI Agents API vs Agents SDK vs Responses API: Na čemu treba graditi u 2026. godini?","Poređenje vlasništva nad izvršnim okruženjem u trenutnim OpenAI platformama za agente i raspodela kontrolnih granica.","Pročitajte poređenje izvršnih okruženja","referralArticle",{},{"id":701,"data":702,"type":42,"tunes":704},"h-change",{"text":703,"level":219},"Šta bi promenilo ovaj odgovor?",{},{"id":706,"data":707,"type":226,"tunes":709},"p-change-1",{"text":708},"Preporuka se menja ukoliko upravljana radna okruženja ponude znatno veću kontrolu nad izvršavanjem, ukoliko samostalno hostovana okruženja dobiju jednostavnije primitive za trajne sesije i oporavak, ili ukoliko propisi zahtevaju da čitava agentska petlja i interakcija sa modelom ostanu unutar infrastrukture kojom sami upravljate.",{},{"id":711,"data":712,"type":226,"tunes":714},"p-change-2",{"text":713},"Takođe se menja sa sposobnostima modela. Anthropic eksplicitno napominje da pretpostavke o radnom okviru (harness) mogu zastareti kako se modeli poboljšavaju. Kontrolni mehanizam koji je danas od suštinskog značaja može kasnije postati nepotreban, dok nova sposobnost modela može stvoriti novi zahtev za upravljanjem.",{},{"id":716,"data":717,"type":42,"tunes":719},"h-limit",{"text":718,"level":219},"Ograničenja",{},{"id":721,"data":722,"type":226,"tunes":724},"p-limit-1",{"text":723},"Ovaj članak razdvaja arhitektonske odgovornosti; on ne tvrdi da je jedan model hostovanja univerzalno bezbedniji, jeftiniji ili pouzdaniji. Ti ishodi zavise od implementacije, radnog opterećenja, zahteva usklađenosti, stručnosti tima i ponašanja pružaoca usluga.",{},{"id":726,"data":727,"type":226,"tunes":729},"p-limit-2",{"text":728},"OpenAI Agents API je još uvek u javnoj beta verziji, a proizvodi upravljanih agenata različitih dobavljača izlažu različite granice. Model sa dve ravni ima za cilj da pomogne u poređenju tih arhitektura bez pretpostavke da svaki dobavljač koristi identične termine.",{},{"id":731,"data":732,"type":42,"tunes":734},"h-conclusion",{"text":733,"level":219},"Zaključak",{},{"id":736,"data":737,"type":226,"tunes":739},"p-conclusion-1",{"text":738},"Korisno pitanje nije „Da li treba sami da hostujemo agenta?” Već glasi: Koju ravan zapravo moramo da kontrolišemo?",{},{"id":741,"data":742,"type":226,"tunes":744},"p-conclusion-2",{"text":743},"Ako je zahtev privatno računarsko okruženje, prilagođeni paketi, lokalni fajlovi ili pristup internoj mreži, sami hostujte ravan izvršavanja, a radni okvir zadržite kao upravljan. Ako su zahtev semantika orkestracije, strategija konteksta, kontrola nad pružaocem usluga ili sam životni ciklus izvršavanja, onda preuzimanje vlasništva nad radnim okvirom može biti opravdano. Eskalirajte kontrolu samo onoliko koliko to zahtev nalaže.",{},{"id":746,"data":747,"type":42,"tunes":749},"h-faq",{"text":748,"level":219},"Često postavljana pitanja",{},{"id":751,"data":752,"type":751,"tunes":775},"faq",{"items":753,"title":774},[754,758,762,766,770],{"id":755,"answer":756,"question":757},"faq1","Ne u potpunosti. OpenAI i dalje pokreće upravljani Codex radni okvir, dok vaša infrastruktura pokreće okruženje za izvršavanje koje se koristi za komande, fajlove i lokalne alate.","Da li je samostalno hostovano okruženje za OpenAI Agents API ujedno i samostalno hostovan agent?",{"id":759,"answer":760,"question":761},"faq2","Često je dovoljno kada se vaši zahtevi odnose na pristup privatnoj mreži, prilagođene pakete, kontrolisane fajlove, specifičan hardver ili infrastrukturne politike, a ne na kontrolu nad samom petljom agenta.","Kada je samostalno hostovano okruženje dovoljno?",{"id":763,"answer":764,"question":765},"faq3","Razmotrite preuzimanje vlasništva nad radnim okvirom kada su vam potrebne prilagođena semantika orkestracije, prilagođeno upravljanje kontekstom, usmeravanje pružaoca usluga, ponašanje pri izvršavanju isključivo lokalno, ili neki drugi zahtev koji pripada petlji agenta, a ne okruženju za izvršavanje.","Kada bi trebalo da samostalno pokrećem radni okvir?",{"id":767,"answer":768,"question":769},"faq4","Ne. Ono menja komponente koje kontrolišete. Bezbednost zavisi od protoka podataka, izolacije, akreditiva, dozvola za alate, umrežavanja, evidentiranja i dizajna životnog ciklusa kroz sve komponente.","Da li samostalno hostovanje automatski poboljšava bezbednost?",{"id":771,"answer":772,"question":773},"faq5","Postajete odgovorni za trajno stanje, upravljanje kontekstom, ponovne pokušaje, otkazivanje, oporavak, opservabilnost, konkurentnost, nadogradnje okruženja za izvršavanje i evaluaciju promena u radnom okviru.","Koji je glavni operativni trošak posedovanja petlje agenta?","Upravljani radni okviri i samostalno hostovana okruženja za izvršavanje agenata",{},{"id":777,"data":778,"type":42,"tunes":780},"h-glossary",{"text":779,"level":219},"Rečnik pojmova",{},{"id":782,"data":783,"type":782,"tunes":810},"glossary",{"title":784,"entries":785},"Ključni arhitektonski termini",[786,790,794,798,802,806],{"term":787,"anchor":788,"definition":789},"Ravan radnog okvira (Harness plane)","harness-plane","Sloj izvršavanja agenta odgovoran za izvođenje petlje, orkestraciju, upravljanje kontekstom, kontinuitet sesije i oporavak.",{"term":791,"anchor":792,"definition":793},"Ravan izvršavanja (Execution plane)","execution-plane","Okruženje u kojem se izvršavaju komande, pokreće kôd i pristupa fajlovima, paketima i lokalnim resursima.",{"term":795,"anchor":796,"definition":797},"Upravljani radni okvir (Managed harness)","managed-harness","Radni okvir agenta čijim izvršnim okruženjem, upravljanjem sesijama i orkestracijom upravlja pružalac platforme.",{"term":799,"anchor":800,"definition":801},"Samostalno hostovano okruženje (Self-hosted environment)","self-hosted-environment","Računarski resursi i fajlovi kojima upravlja vlasnik aplikacije, dok zasebni radni okvir agenta može ostati pod upravljanjem negde drugde.",{"term":803,"anchor":804,"definition":805},"Samostalno vođen radni okvir (Self-operated harness)","self-operated-harness","Izvršno okruženje agenta čijom petljom, hostovanjem, strategijom konteksta i životnim ciklusom upravlja tim aplikacije.",{"term":807,"anchor":808,"definition":809},"Test eskalacije kontrole (Control Escalation Test)","control-escalation-test","Metoda odlučivanja koja povećava vlasništvo nad infrastrukturom i okruženjem za izvršavanje samo kada se zahtev ne može zadovoljiti na sloju niže kontrole.",{},{"id":812,"data":813,"type":42,"tunes":815},"h-sources",{"text":814,"level":219},"Primarni izvori i preporučena literatura",{},{"id":817,"data":818,"type":824,"tunes":825},"src-openai-architecture",{"link":819,"meta":820},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Farchitecture",{"image":821,"title":822,"description":823},{"url":312},"OpenAI — Arhitektura Agents API-ja","Trenutno razdvajanje između hostovanog radnog okvira, okruženja za izvršavanje i aplikativnog servera.","linkTool",{},{"id":827,"data":828,"type":824,"tunes":834},"src-openai-selfhosted",{"link":829,"meta":830},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Fenvironments\u002Fself-hosted",{"image":831,"title":832,"description":833},{"url":312},"OpenAI — Samostalno hostovana izolovana okruženja (sandboxes)","Kako se okruženja za izvršavanje kojima upravlja korisnik povezuju sa upravljanim radnim okvirom i koje odgovornosti za životni ciklus ostaju na aplikaciji.",{},{"id":836,"data":837,"type":824,"tunes":843},"src-openai-lifecycle",{"link":838,"meta":839},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Fenvironments\u002Flifecycle",{"image":840,"title":841,"description":842},{"url":312},"OpenAI — Životni ciklus izolovanog okruženja (sandbox)","Priprema resursa, ponovno povezivanje, sprečavanje dupliranja okruženja i odgovornosti za čišćenje za samostalno hostovane računarske resurse.",{},{"id":845,"data":846,"type":824,"tunes":852},"src-openai-runtime",{"link":847,"meta":848},"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents",{"image":849,"title":850,"description":851},{"url":312},"OpenAI — Opcije okruženja za izvršavanje agenata","Trenutno poređenje Agents API-ja, Codex SDK-a i Responses API-ja prema odgovornostima kojima upravlja platforma u odnosu na aplikaciju.",{},{"id":854,"data":855,"type":824,"tunes":861},"src-openai-platform",{"link":856,"meta":857},"https:\u002F\u002Fdevelopers.openai.com\u002Fblog\u002Fcodex-as-a-platform",{"image":858,"title":859,"description":860},{"url":312},"OpenAI — Codex kao platforma","Codex radni okvir otvorenog koda i integracioni slojevi za aplikacije koje zahtevaju dublju kontrolu izvršavanja.",{},{"id":863,"data":864,"type":824,"tunes":870},"src-anthropic-managed",{"link":865,"meta":866},"https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fmanaged-agents",{"image":867,"title":868,"description":869},{"url":312},"Anthropic — Skaliranje upravljanih agenata: Odvajanje mozga od ruku","Diskusija o arhitekturi upravljanih agenata i zašto pretpostavke o radnom okviru moraju da evoluiraju u skladu sa sposobnostima modela.",{},{"id":872,"data":873,"type":824,"tunes":879},"src-anthropic-harness",{"link":874,"meta":875},"https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Feffective-harnesses-for-long-running-agents",{"image":876,"title":877,"description":878},{"url":312},"Anthropic — Efikasna radna okruženja za dugotrajne agente","Inženjerske lekcije koje pokazuju da performanse dugotrajnih agenata suštinski zavise od dizajna radnog okruženja i trajnih artefakata.",{},"2.31","“Samostalno hostovani agent” može značiti veoma različite arhitekture. Ovaj vodič razgraničava upravljani harness, samostalno hostovano okruženje za izvršavanje i potpuno samostalno upravljanu petlju agenta—i pokazuje koja je granica kontrole timovima zapravo potrebna.","\u002Fuploads\u002F2026\u002F09\u002Fmanaged-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose-1790352403475-kj10jh.webp","managed-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose-1790352403475-kj10jh","PUBLISHED","2026-09-25T12:05:00.000Z","2026-09-25T16:05:38.279Z","2026-09-25T19:49:39.069Z",{"en":889,"de":890,"sr":891,"es":892,"fr":893,"it":894,"ru":895,"zh":896},"\u002Fblog\u002Fmanaged-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose","\u002Fde\u002Fblog\u002Fmanaged-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose","\u002Fsr\u002Fblog\u002Fmanaged-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose","\u002Fes\u002Fblog\u002Fmanaged-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose","\u002Ffr\u002Fblog\u002Fmanaged-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose","\u002Fit\u002Fblog\u002Fmanaged-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose","\u002Fru\u002Fblog\u002Fmanaged-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose","\u002Fzh\u002Fblog\u002Fmanaged-agent-harness-vs-self-hosted-agent-loop-what-you-gain-what-you-lose",[898,902,906],{"id":899,"name":900,"slug":901},89,"Eval harness","evaluation-harness",{"id":903,"name":904,"slug":905},78,"Petlja optimizacije","optimization-loop",{"id":907,"name":908,"slug":909},91,"Monitoring (kvalitet, drift)","monitoring",{"id":911,"login":912,"email":913,"displayName":914},"20","rooth8233","aleksandar@stajic.de","Aleksandar Stajić",[916,1465],{"lang":917,"title":918,"content":919,"contentJson":920,"excerpt":1464},"en","Managed Agent Harness vs Self-Hosted Agent Loop: What You Gain, What You Lose","{\"time\":1790352404508,\"blocks\":[{\"id\":\"-PDI7SJcNl\",\"type\":\"tableOfContents\",\"data\":{\"title\":\"Contents\",\"minLevel\":2,\"maxLevel\":3},\"tunes\":{}},{\"id\":\"intro\",\"type\":\"paragraph\",\"data\":{\"text\":\"The phrase “self-hosted agent” now hides at least three different architectures. You can use a managed harness with OpenAI-hosted compute, a managed harness connected to infrastructure you operate, or run the harness and agent loop yourself. Those choices have very different implications for control, recovery, context management, security, latency, and operational burden.\"},\"tunes\":{}},{\"id\":\"direct\",\"type\":\"callout\",\"data\":{\"variant\":\"info\",\"title\":\"Direct answer\",\"body\":\"\u003Cstrong>Do not choose between “managed” and “self-hosted” as if they were one binary decision.\u003C\u002Fstrong> Separate the \u003Cstrong>harness plane\u003C\u002Fstrong> from the \u003Cstrong>execution plane\u003C\u002Fstrong>. A managed harness can still use self-hosted compute. A self-hosted environment gives you control over files, packages, network access and execution without requiring you to own the agent loop. Run the harness yourself only when you need control over orchestration, lifecycle, model-routing assumptions, or runtime behaviour that a managed harness cannot expose.\"},\"tunes\":{}},{\"id\":\"current\",\"type\":\"callout\",\"data\":{\"variant\":\"warning\",\"title\":\"Current as of 25 September 2026\",\"body\":\"OpenAI's Agents API is in public beta and its architecture can evolve. Current documentation separates the OpenAI-hosted Codex harness from the execution environment and explicitly supports self-hosted environments. OpenAI also exposes the Codex harness separately through the Codex SDK for infrastructure you operate.\"},\"tunes\":{}},{\"id\":\"note\",\"type\":\"callout\",\"data\":{\"variant\":\"note\",\"title\":\"The model used in this article\",\"body\":\"The Harness Plane \u002F Execution Plane model and Control Escalation Test below are practical architecture tools proposed here. They are not formal vendor terminology.\"},\"tunes\":{}},{\"id\":\"h-mistake\",\"type\":\"header\",\"data\":{\"text\":\"The mistake: treating self-hosting as one decision\",\"level\":2},\"tunes\":{}},{\"id\":\"p-mistake-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"In conventional software, “self-hosted” usually means the application runs on infrastructure you control. Agent systems complicate that definition because the runtime can be split. The model-and-tool loop can run in one place while code execution, files and private-network access happen somewhere else.\"},\"tunes\":{}},{\"id\":\"p-mistake-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"OpenAI's current Agents API architecture makes this split explicit: OpenAI runs the harness, while the execution environment can be absent, OpenAI-hosted, or self-hosted. A self-hosted environment therefore does not mean the agent loop is self-hosted.\"},\"tunes\":{}},{\"id\":\"p-mistake-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"This distinction matters because many teams choose a more complex runtime than they need. They want private-network access or custom packages, conclude that the entire agent must be self-hosted, and accidentally take ownership of context management, orchestration, recovery and lifecycle that could have remained managed.\"},\"tunes\":{}},{\"id\":\"h-three\",\"type\":\"header\",\"data\":{\"text\":\"Three architectures that are often called “self-hosted”\",\"level\":2},\"tunes\":{}},{\"id\":\"three-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Architecture\",\"Who runs the harness?\",\"Where code\u002Ffiles execute\",\"What you primarily own\"],[\"Managed harness + managed environment\",\"Platform\",\"Platform-hosted sandbox\",\"Application, tools, product logic, authorization\"],[\"Managed harness + self-hosted environment\",\"Platform\",\"Your container, VM, laptop, private cloud or other compute\",\"Environment provisioning, networking, files and lifecycle; platform still owns harness\"],[\"Self-operated harness \u002F agent loop\",\"You\",\"Your chosen environment\",\"Harness process, orchestration, context strategy, hosting, recovery, execution and application lifecycle\"]]},\"tunes\":{}},{\"id\":\"h-two-plane\",\"type\":\"header\",\"data\":{\"text\":\"The two-plane model\",\"level\":2},\"tunes\":{}},{\"id\":\"plane-comparison\",\"type\":\"comparison\",\"data\":{\"title\":\"Separate the harness plane from the execution plane\",\"layout\":\"table\",\"columns\":[{\"id\":\"plane\",\"label\":\"Plane\"},{\"id\":\"owns\",\"label\":\"What it owns\"},{\"id\":\"questions\",\"label\":\"Questions to ask\"}],\"rows\":[{\"id\":\"harness\",\"label\":\"Harness plane\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"execution\",\"label\":\"Execution plane\",\"values\":[\"\",\"\",\"\"]},{\"id\":\"application\",\"label\":\"Application plane\",\"values\":[\"\",\"\",\"\"]}]},\"tunes\":{}},{\"id\":\"key-consequence\",\"type\":\"callout\",\"data\":{\"variant\":\"success\",\"title\":\"Key architectural consequence\",\"body\":\"You can self-host the \u003Cstrong>execution plane\u003C\u002Fstrong> without self-hosting the \u003Cstrong>harness plane\u003C\u002Fstrong>. That is often the right middle ground.\"},\"tunes\":{}},{\"id\":\"h-managed\",\"type\":\"header\",\"data\":{\"text\":\"Managed harness: what you actually gain\",\"level\":2},\"tunes\":{}},{\"id\":\"p-managed-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"A managed harness removes more than a while-loop. OpenAI's current Agents API manages sessions, orchestration, context compaction and recovery. Anthropic's work on managed agents describes the same broader motivation: harnesses contain assumptions about model behaviour, and those assumptions need to evolve as models improve.\"},\"tunes\":{}},{\"id\":\"p-managed-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"That means the benefit is not only fewer lines of code. The platform can update runtime behaviour, long-horizon context handling, subagent coordination and recovery without requiring every application team to rebuild those mechanisms.\"},\"tunes\":{}},{\"id\":\"managed-list\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"Less application-owned orchestration code.\",\"Managed durable-session behaviour.\",\"Managed context compaction and recovery.\",\"A runtime that can evolve with model capabilities.\",\"Simpler adoption of platform-native subagent and long-running-agent features.\",\"Potentially lower operational burden for teams whose differentiation is not the harness itself.\"]},\"tunes\":{}},{\"id\":\"h-managed-cost\",\"type\":\"header\",\"data\":{\"text\":\"Managed harness: what you give up\",\"level\":2},\"tunes\":{}},{\"id\":\"p-managed-cost-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Delegating the harness also delegates some control. Your application no longer owns every detail of iteration, context strategy, orchestration and runtime evolution. A platform update can improve the system, but it can also change behaviour your product implicitly depended on.\"},\"tunes\":{}},{\"id\":\"p-managed-cost-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"This creates a different kind of engineering requirement: strong evals, explicit product boundaries and an integration layer that prevents managed-session behaviour from becoming your business source of truth.\"},\"tunes\":{}},{\"id\":\"managed-cost-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Managed-harness trade-off\",\"What it means operationally\"],[\"Less loop control\",\"You cannot assume every orchestration detail is application-defined\"],[\"Platform evolution\",\"Harness behaviour can improve or change without your code changing\"],[\"Vendor-specific lifecycle\",\"Sessions, events and recovery semantics become part of the integration surface\"],[\"Observability boundary\",\"Platform traces must be joined with application audit data\"],[\"Portability cost\",\"Moving to another harness later may require more than swapping model endpoints\"]]},\"tunes\":{}},{\"id\":\"h-self-env\",\"type\":\"header\",\"data\":{\"text\":\"Self-hosted execution environment: the middle architecture\",\"level\":2},\"tunes\":{}},{\"id\":\"p-self-env-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"OpenAI's self-hosted environment model is important because it decouples private compute from harness ownership. The platform still runs the Codex harness, while an executor runs inside your environment and receives commands over an outbound connection.\"},\"tunes\":{}},{\"id\":\"p-self-env-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"You control provisioning, files, dependencies, network access and cleanup. The harness can therefore work against private infrastructure or custom software without requiring the entire agent runtime to move into your application.\"},\"tunes\":{}},{\"id\":\"p-self-env-3\",\"type\":\"paragraph\",\"data\":{\"text\":\"The cost is lifecycle responsibility. Your application must map sessions to compute, avoid duplicate provisioning, reconnect environments, coordinate shutdown and preserve any files that must outlive the environment.\"},\"tunes\":{}},{\"id\":\"h-enough\",\"type\":\"header\",\"data\":{\"text\":\"When self-hosted execution is enough\",\"level\":3},\"tunes\":{}},{\"id\":\"enough-list\",\"type\":\"list\",\"data\":{\"style\":\"unordered\",\"meta\":{},\"items\":[\"The agent needs access to a private VPC or internal service.\",\"The agent needs custom binaries, packages, drivers or system software.\",\"The workload must run on hardware or cloud accounts you control.\",\"Files must remain inside a controlled environment.\",\"You need your own sandbox provider or isolation model.\",\"You want platform-managed orchestration but infrastructure-controlled execution.\"]},\"tunes\":{}},{\"id\":\"h-own-harness\",\"type\":\"header\",\"data\":{\"text\":\"When you may need to own the harness too\",\"level\":2},\"tunes\":{}},{\"id\":\"p-own-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Owning the harness becomes justified when the harness itself is part of your product differentiation or constraint set. OpenAI's current runtime overview positions the Codex SDK for running the Codex harness in infrastructure you operate, while Responses is the lower-level option when you want to own the agent loop yourself.\"},\"tunes\":{}},{\"id\":\"p-own-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"The key is to identify a requirement that genuinely lives in the harness plane, not the execution plane.\"},\"tunes\":{}},{\"id\":\"requirements-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Requirement\",\"Execution-plane problem or harness-plane problem?\",\"Likely direction\"],[\"Private database access\",\"Execution plane\",\"Managed harness + self-hosted environment may be sufficient\"],[\"Custom Linux packages\",\"Execution plane\",\"Managed harness + self-hosted environment\"],[\"Custom GPU hardware\",\"Execution plane\",\"Managed harness + self-hosted environment where supported\"],[\"Custom agent stopping logic\",\"Harness plane\",\"Self-operated harness \u002F custom loop\"],[\"Cross-provider model routing at every step\",\"Harness plane\",\"Custom loop or harness you operate\"],[\"Custom context-compaction algorithm\",\"Harness plane\",\"Self-operated harness if the managed runtime cannot expose it\"],[\"Deterministic orchestration semantics required by product\",\"Harness plane\",\"Self-operated harness or tightly controlled custom loop\"],[\"Local-only product deployment with no managed harness dependency\",\"Harness plane + execution plane\",\"Self-operated runtime\"]]},\"tunes\":{}},{\"id\":\"h-control-test\",\"type\":\"header\",\"data\":{\"text\":\"The Control Escalation Test\",\"level\":2},\"tunes\":{}},{\"id\":\"p-control-intro\",\"type\":\"paragraph\",\"data\":{\"text\":\"Use the least self-hosted architecture that satisfies the actual requirement. Escalate control one layer at a time.\"},\"tunes\":{}},{\"id\":\"control-flow\",\"type\":\"processFlow\",\"data\":{\"title\":\"Control Escalation Test\",\"orientation\":\"auto\",\"steps\":[{\"label\":\"1. Start with the application boundary\",\"description\":\"Keep domain truth, authorization and consequential business actions in your own product regardless of agent runtime.\"},{\"label\":\"2. Ask whether the agent needs local execution\",\"description\":\"If not, a managed harness without a dedicated environment may be enough.\"},{\"label\":\"3. Ask whether platform-hosted compute is acceptable\",\"description\":\"If yes, use a managed environment and avoid unnecessary infrastructure ownership.\"},{\"label\":\"4. If not, self-host the execution plane\",\"description\":\"Connect your own environment for private network, files, packages or controlled compute.\"},{\"label\":\"5. Re-evaluate the remaining constraint\",\"description\":\"If the requirement is now satisfied, stop. Do not self-host the harness simply for architectural symmetry.\"},{\"label\":\"6. Escalate to harness ownership only for harness requirements\",\"description\":\"Own the Codex harness or custom agent loop when orchestration, context strategy, lifecycle or portability genuinely requires it.\"},{\"label\":\"7. Prove the extra control is worth the extra operations\",\"description\":\"Benchmark reliability, latency, cost, recovery, observability and engineering burden before committing.\"}]},\"tunes\":{}},{\"id\":\"h-ops\",\"type\":\"header\",\"data\":{\"text\":\"Operational burden grows nonlinearly when you own the harness\",\"level\":2},\"tunes\":{}},{\"id\":\"p-ops-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"A self-operated loop sounds simple in a demo: call model, inspect tool call, execute tool, append result, repeat. Production adds durable state, retries, duplicate events, cancellation, approval, context overflow, tool timeouts, process restarts, trace persistence, backpressure, concurrent work and recovery after partial side effects.\"},\"tunes\":{}},{\"id\":\"p-ops-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"Long-running-agent research from Anthropic repeatedly shows that harness design materially affects performance. Their work on long-running application development uses explicit planning, structured artifacts and evaluator agents because naïve loops tend to lose progress or terminate prematurely. The harness is therefore production logic, not plumbing.\"},\"tunes\":{}},{\"id\":\"ops-table\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"If you own the harness, you also need an answer for\",\"Why it matters\"],[\"Durable session state\",\"Processes restart; long-running work must resume correctly\"],[\"Context compaction\",\"History eventually exceeds practical working context\"],[\"Tool idempotency\",\"Retries must not repeat irreversible side effects\"],[\"Cancellation and interruption\",\"Users and systems need to stop or redirect work\"],[\"Recovery after partial execution\",\"A tool may succeed even if the agent never receives the result\"],[\"Concurrency\",\"Multiple tasks, workers or agents can touch shared state\"],[\"Observability\",\"Final output is insufficient for debugging runtime failures\"],[\"Versioning\",\"Harness updates can change behaviour even when prompts remain constant\"],[\"Evaluation\",\"Runtime changes need regression testing across representative trajectories\"]]},\"tunes\":{}},{\"id\":\"h-security\",\"type\":\"header\",\"data\":{\"text\":\"Security boundary: self-hosting compute does not automatically make the agent private\",\"level\":2},\"tunes\":{}},{\"id\":\"p-sec-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"A self-hosted execution environment controls where commands run and where files live, but the managed harness and model interaction still cross the service boundary. Teams should therefore map data flows explicitly rather than use “self-hosted” as shorthand for a privacy property.\"},\"tunes\":{}},{\"id\":\"p-sec-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"OpenAI's self-hosted executor uses restricted environment credentials and outbound connections. That is useful isolation, but your application still needs its own rules for secrets, private-network exposure, user-to-environment isolation, file retention, tool authorization and data classification.\"},\"tunes\":{}},{\"id\":\"sec-callout\",\"type\":\"callout\",\"data\":{\"variant\":\"warning\",\"title\":\"Important distinction\",\"body\":\"\u003Cstrong>Infrastructure control, execution isolation and data-governance boundaries are related but not identical.\u003C\u002Fstrong> Decide them separately.\"},\"tunes\":{}},{\"id\":\"h-cost\",\"type\":\"header\",\"data\":{\"text\":\"Latency and cost: control can move bottlenecks rather than remove them\",\"level\":2},\"tunes\":{}},{\"id\":\"p-cost-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"Self-hosting may reduce some data-path or environment-startup costs, but it can also add provisioning time, WebSocket lifecycle, cold starts, sandbox cleanup, observability infrastructure and engineering overhead. A managed environment may cost more per unit of compute while being cheaper to operate at low or irregular volume.\"},\"tunes\":{}},{\"id\":\"p-cost-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"The correct comparison is total system cost: model and tool usage, environment time, infrastructure, engineering effort, on-call burden, failure recovery and the cost of slower iteration.\"},\"tunes\":{}},{\"id\":\"h-matrix\",\"type\":\"header\",\"data\":{\"text\":\"A production decision matrix\",\"level\":2},\"tunes\":{}},{\"id\":\"decision-matrix\",\"type\":\"table\",\"data\":{\"withHeadings\":true,\"stretched\":false,\"content\":[[\"Constraint\",\"Managed harness + managed environment\",\"Managed harness + self-hosted environment\",\"Self-operated harness \u002F loop\"],[\"Fastest path to production\",\"Strong\",\"Moderate\",\"Weakest\"],[\"Private-network execution\",\"Weak \u002F depends on connectivity design\",\"Strong\",\"Strong\"],[\"Custom packages \u002F system software\",\"Moderate\",\"Strong\",\"Strong\"],[\"Harness-level control\",\"Low\",\"Low\",\"Highest\"],[\"Operational burden\",\"Lowest\",\"Medium\",\"Highest\"],[\"Portability\",\"Lowest\",\"Medium\",\"Potentially highest if intentionally designed\"],[\"Context strategy control\",\"Platform-managed\",\"Platform-managed\",\"Application-controlled\"],[\"Execution infrastructure control\",\"Low\",\"High\",\"High\"],[\"Ability to benefit from managed harness updates\",\"Highest\",\"Highest\",\"You own adoption\"],[\"Best fit\",\"Teams differentiating at product\u002Ftool layer\",\"Teams needing private\u002Fcustom compute without owning orchestration\",\"Teams whose runtime semantics are themselves a requirement\"]]},\"tunes\":{}},{\"id\":\"h-hybrid\",\"type\":\"header\",\"data\":{\"text\":\"Hybrid is not a compromise — it is often the clean architecture\",\"level\":2},\"tunes\":{}},{\"id\":\"p-hybrid-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"A managed harness with self-hosted execution is not “half self-hosted.” It is an intentional separation of concerns. The platform owns long-horizon agent-runtime complexity, while your infrastructure owns execution, private connectivity and files.\"},\"tunes\":{}},{\"id\":\"p-hybrid-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"That boundary resembles other cloud architectures: managed control plane, customer-controlled data or execution plane. The important design work is defining the contract between them — session identity, environment identity, credentials, files, tool permissions, lifecycle events and cleanup.\"},\"tunes\":{}},{\"id\":\"ref-runtime\",\"type\":\"referralArticle\",\"data\":{\"url\":\"https:\u002F\u002Fstajic.de\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026\",\"title\":\"OpenAI Agents API vs Agents SDK vs Responses API: What Should You Build On in 2026?\",\"excerpt\":\"A runtime-ownership comparison of OpenAI's current agent surfaces and where each control boundary belongs.\",\"ctaLabel\":\"Read the runtime comparison\"},\"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 managed harnesses expose substantially more runtime control, if self-hosted harnesses gain simpler durable-session and recovery primitives, or if regulation requires the entire agent loop and model interaction to remain inside infrastructure you operate.\"},\"tunes\":{}},{\"id\":\"p-change-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"It also changes with model capability. Anthropic explicitly notes that harness assumptions can become stale as models improve. A control mechanism that is essential today may become unnecessary later, while a new model capability can create a new governance requirement.\"},\"tunes\":{}},{\"id\":\"h-limit\",\"type\":\"header\",\"data\":{\"text\":\"Limitations\",\"level\":2},\"tunes\":{}},{\"id\":\"p-limit-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"This article separates architecture responsibilities; it does not claim that one hosting model is universally more secure, cheaper or more reliable. Those outcomes depend on implementation, workload, compliance requirements, team skill and provider behaviour.\"},\"tunes\":{}},{\"id\":\"p-limit-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"The OpenAI Agents API is still in public beta, and managed-agent products from different vendors expose different boundaries. The two-plane model is intended to help compare those architectures without assuming that every vendor uses identical terms.\"},\"tunes\":{}},{\"id\":\"h-conclusion\",\"type\":\"header\",\"data\":{\"text\":\"Conclusion\",\"level\":2},\"tunes\":{}},{\"id\":\"p-conclusion-1\",\"type\":\"paragraph\",\"data\":{\"text\":\"The useful question is not “Should we self-host the agent?” It is: Which plane do we actually need to control?\"},\"tunes\":{}},{\"id\":\"p-conclusion-2\",\"type\":\"paragraph\",\"data\":{\"text\":\"If the requirement is private compute, custom packages, local files or internal-network access, self-host the execution plane and keep the harness managed. If the requirement is orchestration semantics, context strategy, provider control or runtime lifecycle itself, then harness ownership may be justified. Escalate control only as far as the requirement demands.\"},\"tunes\":{}},{\"id\":\"h-faq\",\"type\":\"header\",\"data\":{\"text\":\"FAQ\",\"level\":2},\"tunes\":{}},{\"id\":\"faq\",\"type\":\"faq\",\"data\":{\"title\":\"Managed harnesses and self-hosted agent runtimes\",\"items\":[{\"id\":\"faq1\",\"question\":\"Is an OpenAI Agents API self-hosted environment a self-hosted agent?\",\"answer\":\"Not completely. OpenAI still runs the managed Codex harness, while your infrastructure runs the execution environment used for commands, files and local tools.\"},{\"id\":\"faq2\",\"question\":\"When is a self-hosted environment enough?\",\"answer\":\"It is often enough when your requirements concern private-network access, custom packages, controlled files, specific hardware or infrastructure policy rather than control over the agent loop itself.\"},{\"id\":\"faq3\",\"question\":\"When should I run the harness myself?\",\"answer\":\"Consider harness ownership when you need custom orchestration semantics, custom context management, provider routing, local-only runtime behaviour, or another requirement that lives in the agent loop rather than the execution environment.\"},{\"id\":\"faq4\",\"question\":\"Does self-hosting automatically improve security?\",\"answer\":\"No. It changes which components you control. Security depends on data flow, isolation, credentials, tool permissions, networking, logging and lifecycle design across all components.\"},{\"id\":\"faq5\",\"question\":\"What is the main operational cost of owning the agent loop?\",\"answer\":\"You become responsible for durable state, context management, retries, cancellation, recovery, observability, concurrency, runtime upgrades and evaluation of harness changes.\"}]},\"tunes\":{}},{\"id\":\"h-glossary\",\"type\":\"header\",\"data\":{\"text\":\"Glossary\",\"level\":2},\"tunes\":{}},{\"id\":\"glossary\",\"type\":\"glossary\",\"data\":{\"title\":\"Key architecture terms\",\"entries\":[{\"term\":\"Harness plane\",\"definition\":\"The agent-runtime layer responsible for loop execution, orchestration, context management, session continuity and recovery.\",\"anchor\":\"harness-plane\"},{\"term\":\"Execution plane\",\"definition\":\"The environment in which commands run, code executes and files, packages and local resources are accessed.\",\"anchor\":\"execution-plane\"},{\"term\":\"Managed harness\",\"definition\":\"An agent harness whose runtime, session management and orchestration are operated by a platform provider.\",\"anchor\":\"managed-harness\"},{\"term\":\"Self-hosted environment\",\"definition\":\"Compute and files operated by the application owner while a separate agent harness may remain managed elsewhere.\",\"anchor\":\"self-hosted-environment\"},{\"term\":\"Self-operated harness\",\"definition\":\"An agent runtime whose loop, hosting, context strategy and lifecycle are operated by the application team.\",\"anchor\":\"self-operated-harness\"},{\"term\":\"Control Escalation Test\",\"definition\":\"A decision method that increases infrastructure and runtime ownership only when a requirement cannot be satisfied at a lower-control layer.\",\"anchor\":\"control-escalation-test\"}]},\"tunes\":{}},{\"id\":\"h-sources\",\"type\":\"header\",\"data\":{\"text\":\"Primary sources and further reading\",\"level\":2},\"tunes\":{}},{\"id\":\"src-openai-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\":\"Current separation between the hosted harness, execution environment and application server.\"}},\"tunes\":{}},{\"id\":\"src-openai-selfhosted\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Fenvironments\u002Fself-hosted\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Self-hosted sandboxes\",\"description\":\"How customer-operated execution environments connect to the managed harness and which lifecycle responsibilities remain with the application.\"}},\"tunes\":{}},{\"id\":\"src-openai-lifecycle\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents-api\u002Fenvironments\u002Flifecycle\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Sandbox lifecycle\",\"description\":\"Provisioning, reconnection, duplicate-environment prevention and cleanup responsibilities for self-hosted compute.\"}},\"tunes\":{}},{\"id\":\"src-openai-runtime\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Fagents\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Agent runtime options\",\"description\":\"Current comparison of Agents API, Codex SDK and Responses API by managed versus application-operated responsibilities.\"}},\"tunes\":{}},{\"id\":\"src-openai-platform\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fdevelopers.openai.com\u002Fblog\u002Fcodex-as-a-platform\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"OpenAI — Codex as a platform\",\"description\":\"Open-source Codex harness and integration layers for applications that want deeper runtime control.\"}},\"tunes\":{}},{\"id\":\"src-anthropic-managed\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fmanaged-agents\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"Anthropic — Scaling Managed Agents: Decoupling the brain from the hands\",\"description\":\"Discussion of managed-agent architecture and why harness assumptions need to evolve with model capability.\"}},\"tunes\":{}},{\"id\":\"src-anthropic-harness\",\"type\":\"linkTool\",\"data\":{\"link\":\"https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Feffective-harnesses-for-long-running-agents\",\"meta\":{\"image\":{\"url\":\"\"},\"title\":\"Anthropic — Effective harnesses for long-running agents\",\"description\":\"Engineering lessons showing that long-running agent performance depends materially on harness design and persistent artifacts.\"}},\"tunes\":{}}],\"version\":\"2.31.6\"}",{"time":921,"blocks":922,"version":1463},1790352404508,[923,927,931,936,941,946,950,954,958,962,966,989,993,1014,1019,1023,1027,1031,1042,1046,1050,1054,1076,1080,1084,1088,1092,1096,1107,1111,1115,1119,1151,1155,1159,1185,1189,1193,1197,1231,1235,1239,1243,1248,1252,1256,1260,1264,1307,1311,1315,1319,1326,1330,1334,1338,1342,1346,1350,1354,1358,1362,1366,1386,1390,1410,1414,1421,1428,1435,1442,1449,1456],{"id":215,"data":924,"type":220,"tunes":926},{"title":925,"maxLevel":218,"minLevel":219},"Contents",{},{"id":223,"data":928,"type":226,"tunes":930},{"text":929},"The phrase “self-hosted agent” now hides at least three different architectures. You can use a managed harness with OpenAI-hosted compute, a managed harness connected to infrastructure you operate, or run the harness and agent loop yourself. Those choices have very different implications for control, recovery, context management, security, latency, and operational burden.",{},{"id":229,"data":932,"type":234,"tunes":935},{"body":933,"title":934,"variant":233},"\u003Cstrong>Do not choose between “managed” and “self-hosted” as if they were one binary decision.\u003C\u002Fstrong> Separate the \u003Cstrong>harness plane\u003C\u002Fstrong> from the \u003Cstrong>execution plane\u003C\u002Fstrong>. A managed harness can still use self-hosted compute. A self-hosted environment gives you control over files, packages, network access and execution without requiring you to own the agent loop. Run the harness yourself only when you need control over orchestration, lifecycle, model-routing assumptions, or runtime behaviour that a managed harness cannot expose.","Direct answer",{},{"id":237,"data":937,"type":234,"tunes":940},{"body":938,"title":939,"variant":241},"OpenAI's Agents API is in public beta and its architecture can evolve. Current documentation separates the OpenAI-hosted Codex harness from the execution environment and explicitly supports self-hosted environments. OpenAI also exposes the Codex harness separately through the Codex SDK for infrastructure you operate.","Current as of 25 September 2026",{},{"id":244,"data":942,"type":234,"tunes":945},{"body":943,"title":944,"variant":244},"The Harness Plane \u002F Execution Plane model and Control Escalation Test below are practical architecture tools proposed here. They are not formal vendor terminology.","The model used in this article",{},{"id":250,"data":947,"type":42,"tunes":949},{"text":948,"level":219},"The mistake: treating self-hosting as one decision",{},{"id":255,"data":951,"type":226,"tunes":953},{"text":952},"In conventional software, “self-hosted” usually means the application runs on infrastructure you control. Agent systems complicate that definition because the runtime can be split. The model-and-tool loop can run in one place while code execution, files and private-network access happen somewhere else.",{},{"id":260,"data":955,"type":226,"tunes":957},{"text":956},"OpenAI's current Agents API architecture makes this split explicit: OpenAI runs the harness, while the execution environment can be absent, OpenAI-hosted, or self-hosted. A self-hosted environment therefore does not mean the agent loop is self-hosted.",{},{"id":265,"data":959,"type":226,"tunes":961},{"text":960},"This distinction matters because many teams choose a more complex runtime than they need. They want private-network access or custom packages, conclude that the entire agent must be self-hosted, and accidentally take ownership of context management, orchestration, recovery and lifecycle that could have remained managed.",{},{"id":270,"data":963,"type":42,"tunes":965},{"text":964,"level":219},"Three architectures that are often called “self-hosted”",{},{"id":275,"data":967,"type":297,"tunes":988},{"content":968,"stretched":43,"withHeadings":14},[969,974,979,983],[970,971,972,973],"Architecture","Who runs the harness?","Where code\u002Ffiles execute","What you primarily own",[975,976,977,978],"Managed harness + managed environment","Platform","Platform-hosted sandbox","Application, tools, product logic, authorization",[980,976,981,982],"Managed harness + self-hosted environment","Your container, VM, laptop, private cloud or other compute","Environment provisioning, networking, files and lifecycle; platform still owns harness",[984,985,986,987],"Self-operated harness \u002F agent loop","You","Your chosen environment","Harness process, orchestration, context strategy, hosting, recovery, execution and application lifecycle",{},{"id":300,"data":990,"type":42,"tunes":992},{"text":991,"level":219},"The two-plane model",{},{"id":305,"data":994,"type":332,"tunes":1013},{"rows":995,"title":1005,"layout":297,"columns":1006},[996,999,1002],{"id":309,"label":997,"values":998},"Harness plane",[312,312,312],{"id":314,"label":1000,"values":1001},"Execution plane",[312,312,312],{"id":318,"label":1003,"values":1004},"Application plane",[312,312,312],"Separate the harness plane from the execution plane",[1007,1009,1011],{"id":324,"label":1008},"Plane",{"id":327,"label":1010},"What it owns",{"id":330,"label":1012},"Questions to ask",{},{"id":335,"data":1015,"type":234,"tunes":1018},{"body":1016,"title":1017,"variant":339},"You can self-host the \u003Cstrong>execution plane\u003C\u002Fstrong> without self-hosting the \u003Cstrong>harness plane\u003C\u002Fstrong>. That is often the right middle ground.","Key architectural consequence",{},{"id":342,"data":1020,"type":42,"tunes":1022},{"text":1021,"level":219},"Managed harness: what you actually gain",{},{"id":347,"data":1024,"type":226,"tunes":1026},{"text":1025},"A managed harness removes more than a while-loop. OpenAI's current Agents API manages sessions, orchestration, context compaction and recovery. Anthropic's work on managed agents describes the same broader motivation: harnesses contain assumptions about model behaviour, and those assumptions need to evolve as models improve.",{},{"id":352,"data":1028,"type":226,"tunes":1030},{"text":1029},"That means the benefit is not only fewer lines of code. The platform can update runtime behaviour, long-horizon context handling, subagent coordination and recovery without requiring every application team to rebuild those mechanisms.",{},{"id":357,"data":1032,"type":368,"tunes":1041},{"meta":1033,"items":1034,"style":367},{},[1035,1036,1037,1038,1039,1040],"Less application-owned orchestration code.","Managed durable-session behaviour.","Managed context compaction and recovery.","A runtime that can evolve with model capabilities.","Simpler adoption of platform-native subagent and long-running-agent features.","Potentially lower operational burden for teams whose differentiation is not the harness itself.",{},{"id":371,"data":1043,"type":42,"tunes":1045},{"text":1044,"level":219},"Managed harness: what you give up",{},{"id":376,"data":1047,"type":226,"tunes":1049},{"text":1048},"Delegating the harness also delegates some control. Your application no longer owns every detail of iteration, context strategy, orchestration and runtime evolution. A platform update can improve the system, but it can also change behaviour your product implicitly depended on.",{},{"id":381,"data":1051,"type":226,"tunes":1053},{"text":1052},"This creates a different kind of engineering requirement: strong evals, explicit product boundaries and an integration layer that prevents managed-session behaviour from becoming your business source of truth.",{},{"id":386,"data":1055,"type":297,"tunes":1075},{"content":1056,"stretched":43,"withHeadings":14},[1057,1060,1063,1066,1069,1072],[1058,1059],"Managed-harness trade-off","What it means operationally",[1061,1062],"Less loop control","You cannot assume every orchestration detail is application-defined",[1064,1065],"Platform evolution","Harness behaviour can improve or change without your code changing",[1067,1068],"Vendor-specific lifecycle","Sessions, events and recovery semantics become part of the integration surface",[1070,1071],"Observability boundary","Platform traces must be joined with application audit data",[1073,1074],"Portability cost","Moving to another harness later may require more than swapping model endpoints",{},{"id":409,"data":1077,"type":42,"tunes":1079},{"text":1078,"level":219},"Self-hosted execution environment: the middle architecture",{},{"id":414,"data":1081,"type":226,"tunes":1083},{"text":1082},"OpenAI's self-hosted environment model is important because it decouples private compute from harness ownership. The platform still runs the Codex harness, while an executor runs inside your environment and receives commands over an outbound connection.",{},{"id":419,"data":1085,"type":226,"tunes":1087},{"text":1086},"You control provisioning, files, dependencies, network access and cleanup. The harness can therefore work against private infrastructure or custom software without requiring the entire agent runtime to move into your application.",{},{"id":424,"data":1089,"type":226,"tunes":1091},{"text":1090},"The cost is lifecycle responsibility. Your application must map sessions to compute, avoid duplicate provisioning, reconnect environments, coordinate shutdown and preserve any files that must outlive the environment.",{},{"id":429,"data":1093,"type":42,"tunes":1095},{"text":1094,"level":218},"When self-hosted execution is enough",{},{"id":434,"data":1097,"type":368,"tunes":1106},{"meta":1098,"items":1099,"style":367},{},[1100,1101,1102,1103,1104,1105],"The agent needs access to a private VPC or internal service.","The agent needs custom binaries, packages, drivers or system software.","The workload must run on hardware or cloud accounts you control.","Files must remain inside a controlled environment.","You need your own sandbox provider or isolation model.","You want platform-managed orchestration but infrastructure-controlled execution.",{},{"id":446,"data":1108,"type":42,"tunes":1110},{"text":1109,"level":219},"When you may need to own the harness too",{},{"id":451,"data":1112,"type":226,"tunes":1114},{"text":1113},"Owning the harness becomes justified when the harness itself is part of your product differentiation or constraint set. OpenAI's current runtime overview positions the Codex SDK for running the Codex harness in infrastructure you operate, while Responses is the lower-level option when you want to own the agent loop yourself.",{},{"id":456,"data":1116,"type":226,"tunes":1118},{"text":1117},"The key is to identify a requirement that genuinely lives in the harness plane, not the execution plane.",{},{"id":461,"data":1120,"type":297,"tunes":1150},{"content":1121,"stretched":43,"withHeadings":14},[1122,1126,1129,1131,1134,1137,1140,1143,1146],[1123,1124,1125],"Requirement","Execution-plane problem or harness-plane problem?","Likely direction",[1127,1000,1128],"Private database access","Managed harness + self-hosted environment may be sufficient",[1130,1000,980],"Custom Linux packages",[1132,1000,1133],"Custom GPU hardware","Managed harness + self-hosted environment where supported",[1135,997,1136],"Custom agent stopping logic","Self-operated harness \u002F custom loop",[1138,997,1139],"Cross-provider model routing at every step","Custom loop or harness you operate",[1141,997,1142],"Custom context-compaction algorithm","Self-operated harness if the managed runtime cannot expose it",[1144,997,1145],"Deterministic orchestration semantics required by product","Self-operated harness or tightly controlled custom loop",[1147,1148,1149],"Local-only product deployment with no managed harness dependency","Harness plane + execution plane","Self-operated runtime",{},{"id":497,"data":1152,"type":42,"tunes":1154},{"text":1153,"level":219},"The Control Escalation Test",{},{"id":502,"data":1156,"type":226,"tunes":1158},{"text":1157},"Use the least self-hosted architecture that satisfies the actual requirement. Escalate control one layer at a time.",{},{"id":507,"data":1160,"type":532,"tunes":1184},{"steps":1161,"title":1183,"orientation":531},[1162,1165,1168,1171,1174,1177,1180],{"label":1163,"description":1164},"1. Start with the application boundary","Keep domain truth, authorization and consequential business actions in your own product regardless of agent runtime.",{"label":1166,"description":1167},"2. Ask whether the agent needs local execution","If not, a managed harness without a dedicated environment may be enough.",{"label":1169,"description":1170},"3. Ask whether platform-hosted compute is acceptable","If yes, use a managed environment and avoid unnecessary infrastructure ownership.",{"label":1172,"description":1173},"4. If not, self-host the execution plane","Connect your own environment for private network, files, packages or controlled compute.",{"label":1175,"description":1176},"5. Re-evaluate the remaining constraint","If the requirement is now satisfied, stop. Do not self-host the harness simply for architectural symmetry.",{"label":1178,"description":1179},"6. Escalate to harness ownership only for harness requirements","Own the Codex harness or custom agent loop when orchestration, context strategy, lifecycle or portability genuinely requires it.",{"label":1181,"description":1182},"7. Prove the extra control is worth the extra operations","Benchmark reliability, latency, cost, recovery, observability and engineering burden before committing.","Control Escalation Test",{},{"id":535,"data":1186,"type":42,"tunes":1188},{"text":1187,"level":219},"Operational burden grows nonlinearly when you own the harness",{},{"id":540,"data":1190,"type":226,"tunes":1192},{"text":1191},"A self-operated loop sounds simple in a demo: call model, inspect tool call, execute tool, append result, repeat. Production adds durable state, retries, duplicate events, cancellation, approval, context overflow, tool timeouts, process restarts, trace persistence, backpressure, concurrent work and recovery after partial side effects.",{},{"id":545,"data":1194,"type":226,"tunes":1196},{"text":1195},"Long-running-agent research from Anthropic repeatedly shows that harness design materially affects performance. Their work on long-running application development uses explicit planning, structured artifacts and evaluator agents because naïve loops tend to lose progress or terminate prematurely. The harness is therefore production logic, not plumbing.",{},{"id":550,"data":1198,"type":297,"tunes":1230},{"content":1199,"stretched":43,"withHeadings":14},[1200,1203,1206,1209,1212,1215,1218,1221,1224,1227],[1201,1202],"If you own the harness, you also need an answer for","Why it matters",[1204,1205],"Durable session state","Processes restart; long-running work must resume correctly",[1207,1208],"Context compaction","History eventually exceeds practical working context",[1210,1211],"Tool idempotency","Retries must not repeat irreversible side effects",[1213,1214],"Cancellation and interruption","Users and systems need to stop or redirect work",[1216,1217],"Recovery after partial execution","A tool may succeed even if the agent never receives the result",[1219,1220],"Concurrency","Multiple tasks, workers or agents can touch shared state",[1222,1223],"Observability","Final output is insufficient for debugging runtime failures",[1225,1226],"Versioning","Harness updates can change behaviour even when prompts remain constant",[1228,1229],"Evaluation","Runtime changes need regression testing across representative trajectories",{},{"id":585,"data":1232,"type":42,"tunes":1234},{"text":1233,"level":219},"Security boundary: self-hosting compute does not automatically make the agent private",{},{"id":590,"data":1236,"type":226,"tunes":1238},{"text":1237},"A self-hosted execution environment controls where commands run and where files live, but the managed harness and model interaction still cross the service boundary. Teams should therefore map data flows explicitly rather than use “self-hosted” as shorthand for a privacy property.",{},{"id":595,"data":1240,"type":226,"tunes":1242},{"text":1241},"OpenAI's self-hosted executor uses restricted environment credentials and outbound connections. That is useful isolation, but your application still needs its own rules for secrets, private-network exposure, user-to-environment isolation, file retention, tool authorization and data classification.",{},{"id":600,"data":1244,"type":234,"tunes":1247},{"body":1245,"title":1246,"variant":241},"\u003Cstrong>Infrastructure control, execution isolation and data-governance boundaries are related but not identical.\u003C\u002Fstrong> Decide them separately.","Important distinction",{},{"id":606,"data":1249,"type":42,"tunes":1251},{"text":1250,"level":219},"Latency and cost: control can move bottlenecks rather than remove them",{},{"id":611,"data":1253,"type":226,"tunes":1255},{"text":1254},"Self-hosting may reduce some data-path or environment-startup costs, but it can also add provisioning time, WebSocket lifecycle, cold starts, sandbox cleanup, observability infrastructure and engineering overhead. A managed environment may cost more per unit of compute while being cheaper to operate at low or irregular volume.",{},{"id":616,"data":1257,"type":226,"tunes":1259},{"text":1258},"The correct comparison is total system cost: model and tool usage, environment time, infrastructure, engineering effort, on-call burden, failure recovery and the cost of slower iteration.",{},{"id":621,"data":1261,"type":42,"tunes":1263},{"text":1262,"level":219},"A production decision matrix",{},{"id":626,"data":1265,"type":297,"tunes":1306},{"content":1266,"stretched":43,"withHeadings":14},[1267,1270,1275,1278,1280,1284,1288,1291,1295,1298,1301],[1268,975,980,1269],"Constraint","Self-operated harness \u002F loop",[1271,1272,1273,1274],"Fastest path to production","Strong","Moderate","Weakest",[1276,1277,1272,1272],"Private-network execution","Weak \u002F depends on connectivity design",[1279,1273,1272,1272],"Custom packages \u002F system software",[1281,1282,1282,1283],"Harness-level control","Low","Highest",[1285,1286,1287,1283],"Operational burden","Lowest","Medium",[1289,1286,1287,1290],"Portability","Potentially highest if intentionally designed",[1292,1293,1293,1294],"Context strategy control","Platform-managed","Application-controlled",[1296,1282,1297,1297],"Execution infrastructure control","High",[1299,1283,1283,1300],"Ability to benefit from managed harness updates","You own adoption",[1302,1303,1304,1305],"Best fit","Teams differentiating at product\u002Ftool layer","Teams needing private\u002Fcustom compute without owning orchestration","Teams whose runtime semantics are themselves a requirement",{},{"id":677,"data":1308,"type":42,"tunes":1310},{"text":1309,"level":219},"Hybrid is not a compromise — it is often the clean architecture",{},{"id":682,"data":1312,"type":226,"tunes":1314},{"text":1313},"A managed harness with self-hosted execution is not “half self-hosted.” It is an intentional separation of concerns. The platform owns long-horizon agent-runtime complexity, while your infrastructure owns execution, private connectivity and files.",{},{"id":687,"data":1316,"type":226,"tunes":1318},{"text":1317},"That boundary resembles other cloud architectures: managed control plane, customer-controlled data or execution plane. The important design work is defining the contract between them — session identity, environment identity, credentials, files, tool permissions, lifecycle events and cleanup.",{},{"id":692,"data":1320,"type":698,"tunes":1325},{"url":1321,"title":1322,"excerpt":1323,"ctaLabel":1324},"https:\u002F\u002Fstajic.de\u002Fblog\u002Fopenai-agents-api-vs-agents-sdk-vs-responses-api-what-should-you-build-on-in-2026","OpenAI Agents API vs Agents SDK vs Responses API: What Should You Build On in 2026?","A runtime-ownership comparison of OpenAI's current agent surfaces and where each control boundary belongs.","Read the runtime comparison",{},{"id":701,"data":1327,"type":42,"tunes":1329},{"text":1328,"level":219},"What would change this answer?",{},{"id":706,"data":1331,"type":226,"tunes":1333},{"text":1332},"The recommendation changes if managed harnesses expose substantially more runtime control, if self-hosted harnesses gain simpler durable-session and recovery primitives, or if regulation requires the entire agent loop and model interaction to remain inside infrastructure you operate.",{},{"id":711,"data":1335,"type":226,"tunes":1337},{"text":1336},"It also changes with model capability. Anthropic explicitly notes that harness assumptions can become stale as models improve. A control mechanism that is essential today may become unnecessary later, while a new model capability can create a new governance requirement.",{},{"id":716,"data":1339,"type":42,"tunes":1341},{"text":1340,"level":219},"Limitations",{},{"id":721,"data":1343,"type":226,"tunes":1345},{"text":1344},"This article separates architecture responsibilities; it does not claim that one hosting model is universally more secure, cheaper or more reliable. Those outcomes depend on implementation, workload, compliance requirements, team skill and provider behaviour.",{},{"id":726,"data":1347,"type":226,"tunes":1349},{"text":1348},"The OpenAI Agents API is still in public beta, and managed-agent products from different vendors expose different boundaries. The two-plane model is intended to help compare those architectures without assuming that every vendor uses identical terms.",{},{"id":731,"data":1351,"type":42,"tunes":1353},{"text":1352,"level":219},"Conclusion",{},{"id":736,"data":1355,"type":226,"tunes":1357},{"text":1356},"The useful question is not “Should we self-host the agent?” It is: Which plane do we actually need to control?",{},{"id":741,"data":1359,"type":226,"tunes":1361},{"text":1360},"If the requirement is private compute, custom packages, local files or internal-network access, self-host the execution plane and keep the harness managed. If the requirement is orchestration semantics, context strategy, provider control or runtime lifecycle itself, then harness ownership may be justified. Escalate control only as far as the requirement demands.",{},{"id":746,"data":1363,"type":42,"tunes":1365},{"text":1364,"level":219},"FAQ",{},{"id":751,"data":1367,"type":751,"tunes":1385},{"items":1368,"title":1384},[1369,1372,1375,1378,1381],{"id":755,"answer":1370,"question":1371},"Not completely. OpenAI still runs the managed Codex harness, while your infrastructure runs the execution environment used for commands, files and local tools.","Is an OpenAI Agents API self-hosted environment a self-hosted agent?",{"id":759,"answer":1373,"question":1374},"It is often enough when your requirements concern private-network access, custom packages, controlled files, specific hardware or infrastructure policy rather than control over the agent loop itself.","When is a self-hosted environment enough?",{"id":763,"answer":1376,"question":1377},"Consider harness ownership when you need custom orchestration semantics, custom context management, provider routing, local-only runtime behaviour, or another requirement that lives in the agent loop rather than the execution environment.","When should I run the harness myself?",{"id":767,"answer":1379,"question":1380},"No. It changes which components you control. Security depends on data flow, isolation, credentials, tool permissions, networking, logging and lifecycle design across all components.","Does self-hosting automatically improve security?",{"id":771,"answer":1382,"question":1383},"You become responsible for durable state, context management, retries, cancellation, recovery, observability, concurrency, runtime upgrades and evaluation of harness changes.","What is the main operational cost of owning the agent loop?","Managed harnesses and self-hosted agent runtimes",{},{"id":777,"data":1387,"type":42,"tunes":1389},{"text":1388,"level":219},"Glossary",{},{"id":782,"data":1391,"type":782,"tunes":1409},{"title":1392,"entries":1393},"Key architecture terms",[1394,1396,1398,1401,1404,1407],{"term":997,"anchor":788,"definition":1395},"The agent-runtime layer responsible for loop execution, orchestration, context management, session continuity and recovery.",{"term":1000,"anchor":792,"definition":1397},"The environment in which commands run, code executes and files, packages and local resources are accessed.",{"term":1399,"anchor":796,"definition":1400},"Managed harness","An agent harness whose runtime, session management and orchestration are operated by a platform provider.",{"term":1402,"anchor":800,"definition":1403},"Self-hosted environment","Compute and files operated by the application owner while a separate agent harness may remain managed elsewhere.",{"term":1405,"anchor":804,"definition":1406},"Self-operated harness","An agent runtime whose loop, hosting, context strategy and lifecycle are operated by the application team.",{"term":1183,"anchor":808,"definition":1408},"A decision method that increases infrastructure and runtime ownership only when a requirement cannot be satisfied at a lower-control layer.",{},{"id":812,"data":1411,"type":42,"tunes":1413},{"text":1412,"level":219},"Primary sources and further reading",{},{"id":817,"data":1415,"type":824,"tunes":1420},{"link":819,"meta":1416},{"image":1417,"title":1418,"description":1419},{"url":312},"OpenAI — Agents API Architecture","Current separation between the hosted harness, execution environment and application server.",{},{"id":827,"data":1422,"type":824,"tunes":1427},{"link":829,"meta":1423},{"image":1424,"title":1425,"description":1426},{"url":312},"OpenAI — Self-hosted sandboxes","How customer-operated execution environments connect to the managed harness and which lifecycle responsibilities remain with the application.",{},{"id":836,"data":1429,"type":824,"tunes":1434},{"link":838,"meta":1430},{"image":1431,"title":1432,"description":1433},{"url":312},"OpenAI — Sandbox lifecycle","Provisioning, reconnection, duplicate-environment prevention and cleanup responsibilities for self-hosted compute.",{},{"id":845,"data":1436,"type":824,"tunes":1441},{"link":847,"meta":1437},{"image":1438,"title":1439,"description":1440},{"url":312},"OpenAI — Agent runtime options","Current comparison of Agents API, Codex SDK and Responses API by managed versus application-operated responsibilities.",{},{"id":854,"data":1443,"type":824,"tunes":1448},{"link":856,"meta":1444},{"image":1445,"title":1446,"description":1447},{"url":312},"OpenAI — Codex as a platform","Open-source Codex harness and integration layers for applications that want deeper runtime control.",{},{"id":863,"data":1450,"type":824,"tunes":1455},{"link":865,"meta":1451},{"image":1452,"title":1453,"description":1454},{"url":312},"Anthropic — Scaling Managed Agents: Decoupling the brain from the hands","Discussion of managed-agent architecture and why harness assumptions need to evolve with model capability.",{},{"id":872,"data":1457,"type":824,"tunes":1462},{"link":874,"meta":1458},{"image":1459,"title":1460,"description":1461},{"url":312},"Anthropic — Effective harnesses for long-running agents","Engineering lessons showing that long-running agent performance depends materially on harness design and persistent artifacts.",{},"2.31.6","“Self-hosted agent” can mean very different architectures. This guide separates the managed harness, self-hosted execution environment, and fully self-operated agent loop—and shows which control boundary teams actually need.",{"lang":7,"title":208,"content":210,"contentJson":1466,"excerpt":881},{"time":212,"blocks":1467,"version":880},[1468,1471,1474,1477,1480,1483,1486,1489,1492,1495,1498,1506,1509,1523,1526,1529,1532,1535,1540,1543,1546,1549,1559,1562,1565,1568,1571,1574,1579,1582,1585,1588,1601,1604,1607,1618,1621,1624,1627,1641,1644,1647,1650,1653,1656,1659,1662,1665,1680,1683,1686,1689,1692,1695,1698,1701,1704,1707,1710,1713,1716,1719,1722,1731,1734,1744,1747,1752,1757,1762,1767,1772,1777],{"id":215,"data":1469,"type":220,"tunes":1470},{"title":217,"maxLevel":218,"minLevel":219},{},{"id":223,"data":1472,"type":226,"tunes":1473},{"text":225},{},{"id":229,"data":1475,"type":234,"tunes":1476},{"body":231,"title":232,"variant":233},{},{"id":237,"data":1478,"type":234,"tunes":1479},{"body":239,"title":240,"variant":241},{},{"id":244,"data":1481,"type":234,"tunes":1482},{"body":246,"title":247,"variant":244},{},{"id":250,"data":1484,"type":42,"tunes":1485},{"text":252,"level":219},{},{"id":255,"data":1487,"type":226,"tunes":1488},{"text":257},{},{"id":260,"data":1490,"type":226,"tunes":1491},{"text":262},{},{"id":265,"data":1493,"type":226,"tunes":1494},{"text":267},{},{"id":270,"data":1496,"type":42,"tunes":1497},{"text":272,"level":219},{},{"id":275,"data":1499,"type":297,"tunes":1505},{"content":1500,"stretched":43,"withHeadings":14},[1501,1502,1503,1504],[279,280,281,282],[284,285,286,287],[289,285,290,291],[293,294,295,296],{},{"id":300,"data":1507,"type":42,"tunes":1508},{"text":302,"level":219},{},{"id":305,"data":1510,"type":332,"tunes":1522},{"rows":1511,"title":321,"layout":297,"columns":1518},[1512,1514,1516],{"id":309,"label":310,"values":1513},[312,312,312],{"id":314,"label":315,"values":1515},[312,312,312],{"id":318,"label":319,"values":1517},[312,312,312],[1519,1520,1521],{"id":324,"label":325},{"id":327,"label":328},{"id":330,"label":331},{},{"id":335,"data":1524,"type":234,"tunes":1525},{"body":337,"title":338,"variant":339},{},{"id":342,"data":1527,"type":42,"tunes":1528},{"text":344,"level":219},{},{"id":347,"data":1530,"type":226,"tunes":1531},{"text":349},{},{"id":352,"data":1533,"type":226,"tunes":1534},{"text":354},{},{"id":357,"data":1536,"type":368,"tunes":1539},{"meta":1537,"items":1538,"style":367},{},[361,362,363,364,365,366],{},{"id":371,"data":1541,"type":42,"tunes":1542},{"text":373,"level":219},{},{"id":376,"data":1544,"type":226,"tunes":1545},{"text":378},{},{"id":381,"data":1547,"type":226,"tunes":1548},{"text":383},{},{"id":386,"data":1550,"type":297,"tunes":1558},{"content":1551,"stretched":43,"withHeadings":14},[1552,1553,1554,1555,1556,1557],[390,391],[393,394],[396,397],[399,400],[402,403],[405,406],{},{"id":409,"data":1560,"type":42,"tunes":1561},{"text":411,"level":219},{},{"id":414,"data":1563,"type":226,"tunes":1564},{"text":416},{},{"id":419,"data":1566,"type":226,"tunes":1567},{"text":421},{},{"id":424,"data":1569,"type":226,"tunes":1570},{"text":426},{},{"id":429,"data":1572,"type":42,"tunes":1573},{"text":431,"level":218},{},{"id":434,"data":1575,"type":368,"tunes":1578},{"meta":1576,"items":1577,"style":367},{},[438,439,440,441,442,443],{},{"id":446,"data":1580,"type":42,"tunes":1581},{"text":448,"level":219},{},{"id":451,"data":1583,"type":226,"tunes":1584},{"text":453},{},{"id":456,"data":1586,"type":226,"tunes":1587},{"text":458},{},{"id":461,"data":1589,"type":297,"tunes":1600},{"content":1590,"stretched":43,"withHeadings":14},[1591,1592,1593,1594,1595,1596,1597,1598,1599],[465,466,467],[469,470,471],[473,470,474],[476,470,477],[479,480,481],[483,480,484],[486,480,487],[489,480,490],[492,493,494],{},{"id":497,"data":1602,"type":42,"tunes":1603},{"text":499,"level":219},{},{"id":502,"data":1605,"type":226,"tunes":1606},{"text":504},{},{"id":507,"data":1608,"type":532,"tunes":1617},{"steps":1609,"title":499,"orientation":531},[1610,1611,1612,1613,1614,1615,1616],{"label":511,"description":512},{"label":514,"description":515},{"label":517,"description":518},{"label":520,"description":521},{"label":523,"description":524},{"label":526,"description":527},{"label":529,"description":530},{},{"id":535,"data":1619,"type":42,"tunes":1620},{"text":537,"level":219},{},{"id":540,"data":1622,"type":226,"tunes":1623},{"text":542},{},{"id":545,"data":1625,"type":226,"tunes":1626},{"text":547},{},{"id":550,"data":1628,"type":297,"tunes":1640},{"content":1629,"stretched":43,"withHeadings":14},[1630,1631,1632,1633,1634,1635,1636,1637,1638,1639],[554,555],[557,558],[560,561],[563,564],[566,567],[569,570],[572,573],[575,576],[578,579],[581,582],{},{"id":585,"data":1642,"type":42,"tunes":1643},{"text":587,"level":219},{},{"id":590,"data":1645,"type":226,"tunes":1646},{"text":592},{},{"id":595,"data":1648,"type":226,"tunes":1649},{"text":597},{},{"id":600,"data":1651,"type":234,"tunes":1652},{"body":602,"title":603,"variant":241},{},{"id":606,"data":1654,"type":42,"tunes":1655},{"text":608,"level":219},{},{"id":611,"data":1657,"type":226,"tunes":1658},{"text":613},{},{"id":616,"data":1660,"type":226,"tunes":1661},{"text":618},{},{"id":621,"data":1663,"type":42,"tunes":1664},{"text":623,"level":219},{},{"id":626,"data":1666,"type":297,"tunes":1679},{"content":1667,"stretched":43,"withHeadings":14},[1668,1669,1670,1671,1672,1673,1674,1675,1676,1677,1678],[630,631,632,633],[635,636,637,638],[640,641,642,642],[644,645,642,642],[647,648,648,649],[651,652,653,654],[656,657,658,659],[661,662,662,663],[665,648,666,666],[668,649,649,669],[671,672,673,674],{},{"id":677,"data":1681,"type":42,"tunes":1682},{"text":679,"level":219},{},{"id":682,"data":1684,"type":226,"tunes":1685},{"text":684},{},{"id":687,"data":1687,"type":226,"tunes":1688},{"text":689},{},{"id":692,"data":1690,"type":698,"tunes":1691},{"url":694,"title":695,"excerpt":696,"ctaLabel":697},{},{"id":701,"data":1693,"type":42,"tunes":1694},{"text":703,"level":219},{},{"id":706,"data":1696,"type":226,"tunes":1697},{"text":708},{},{"id":711,"data":1699,"type":226,"tunes":1700},{"text":713},{},{"id":716,"data":1702,"type":42,"tunes":1703},{"text":718,"level":219},{},{"id":721,"data":1705,"type":226,"tunes":1706},{"text":723},{},{"id":726,"data":1708,"type":226,"tunes":1709},{"text":728},{},{"id":731,"data":1711,"type":42,"tunes":1712},{"text":733,"level":219},{},{"id":736,"data":1714,"type":226,"tunes":1715},{"text":738},{},{"id":741,"data":1717,"type":226,"tunes":1718},{"text":743},{},{"id":746,"data":1720,"type":42,"tunes":1721},{"text":748,"level":219},{},{"id":751,"data":1723,"type":751,"tunes":1730},{"items":1724,"title":774},[1725,1726,1727,1728,1729],{"id":755,"answer":756,"question":757},{"id":759,"answer":760,"question":761},{"id":763,"answer":764,"question":765},{"id":767,"answer":768,"question":769},{"id":771,"answer":772,"question":773},{},{"id":777,"data":1732,"type":42,"tunes":1733},{"text":779,"level":219},{},{"id":782,"data":1735,"type":782,"tunes":1743},{"title":784,"entries":1736},[1737,1738,1739,1740,1741,1742],{"term":787,"anchor":788,"definition":789},{"term":791,"anchor":792,"definition":793},{"term":795,"anchor":796,"definition":797},{"term":799,"anchor":800,"definition":801},{"term":803,"anchor":804,"definition":805},{"term":807,"anchor":808,"definition":809},{},{"id":812,"data":1745,"type":42,"tunes":1746},{"text":814,"level":219},{},{"id":817,"data":1748,"type":824,"tunes":1751},{"link":819,"meta":1749},{"image":1750,"title":822,"description":823},{"url":312},{},{"id":827,"data":1753,"type":824,"tunes":1756},{"link":829,"meta":1754},{"image":1755,"title":832,"description":833},{"url":312},{},{"id":836,"data":1758,"type":824,"tunes":1761},{"link":838,"meta":1759},{"image":1760,"title":841,"description":842},{"url":312},{},{"id":845,"data":1763,"type":824,"tunes":1766},{"link":847,"meta":1764},{"image":1765,"title":850,"description":851},{"url":312},{},{"id":854,"data":1768,"type":824,"tunes":1771},{"link":856,"meta":1769},{"image":1770,"title":859,"description":860},{"url":312},{},{"id":863,"data":1773,"type":824,"tunes":1776},{"link":865,"meta":1774},{"image":1775,"title":868,"description":869},{"url":312},{},{"id":872,"data":1778,"type":824,"tunes":1781},{"link":874,"meta":1779},{"image":1780,"title":877,"description":878},{"url":312},{},"Post erfolgreich abgerufen",{"items":1784,"source":1826,"manualIds":1827,"manualMatchedIds":1828},[1785,1792,1799,1805,1812,1819],{"id":1786,"slug":1787,"title":1788,"excerpt":1789,"featuredImage":1790,"publishedAt":1791},"364","tipps-fuer-die-verbesserung-der-seo-suchmaschinenoptimierung","Ovladavanje SEO radnim tokom: Ključne strategije optimizacije za organski rast","Strukturiran SEO tok posla je ključan za održiv organski rast. Naučite deset osnovnih strategija, od istraživanja ključnih reči i tehničke optimizacije do kvaliteta sadržaja i analize performansi.","\u002Fuploads\u002F2026\u002F03\u002Ftipps-fuer-die-verbesserung-der-seo-suchmaschinenoptimierung-1774866098131-hwkzrg.webp","2024-01-26T06:35:00.000Z",{"id":1793,"slug":1794,"title":1795,"excerpt":1796,"featuredImage":1797,"publishedAt":1798},"469","rag-failed-but-which-layer-actually-failed-a-diagnostic-method","RAG nije uspeo — ali koji sloj je zapravo zakazao? Dijagnostička metoda","Kada je RAG odgovor pogrešan, kriviti pretragu ili model je previše neodređeno. Ova dijagnostička metoda izoluje pokrivenost izvora, konstrukciju upita, pretragu, rangiranje, sastavljanje konteksta, generisanje, pripisivanje dokaza i svežinu—tako da se stvarni kvar može reprodukovati i ispraviti.","\u002Fuploads\u002F2026\u002F09\u002Frag-failed-but-which-layer-actually-failed-a-diagnostic-method-1790350847177-pior4c.webp","2026-09-24T19:39:00.000Z",{"id":1800,"slug":901,"title":1801,"excerpt":1802,"featuredImage":1803,"publishedAt":1804},"434","Sveobuhvatan vodič za Evaluation Harness: Ovladavanje evaluacijom performansi LLM-ova","Ovaj vodič pruža detaljan pregled Evaluation Harness-a, ključnog okvira za rigoroznu procenu sposobnosti velikih jezičkih modela (LLM) u korporativnim LLMOps procesima. Naučite podešavanje, najbolje prakse i napredne tehnike kako biste osigurali pouzdano benčmarkovanje i optimizaciju modela.","\u002Fuploads\u002F2026\u002F04\u002Fevaluation-harness-1775466944495-4s0xv2.webp","2026-03-01T17:50:00.000Z",{"id":1806,"slug":1807,"title":1808,"excerpt":1809,"featuredImage":1810,"publishedAt":1811},"362","suchmaschinen-webcrawler-suchsoftware-und-suchergebnissen","Optimizacija za pretraživače: Pouzdan radni proces za visoko rangiranje","Detaljna analiza optimizacije za pretraživače (SEO), njenih tehničkih osnova, uloge veb krolera i strateških koraka za postizanje organskih vrhunskih pozicija.","\u002Fuploads\u002F2026\u002F03\u002Fsuchmaschinen-webcrawler-suchsoftware-und-suchergebnissen-1774865001795-5wv0cw.webp","2023-04-12T13:03:00.000Z",{"id":1813,"slug":1814,"title":1815,"excerpt":1816,"featuredImage":1817,"publishedAt":1818},"459","ollama-is-not-the-product-building-production-ready-open-llm-applications","Ollama nije proizvod: Izgradnja aplikacija spremnih za produkciju sa otvorenim LLM-ovima","Pokretanje lokalnog modela pomoću Ollama-e je jednostavno. Izgradnja Open-LLM aplikacije spremne za produkciju je teža: zahteva RAG, kontrolu pristupa, apstrakciju provajdera, evaluaciju, logovanje, disciplinu puštanja u rad i kontrolisani aplikativni sloj oko modela.","\u002Fuploads\u002F2026\u002F06\u002Follama-is-not-the-product-building-production-ready-open-llm-applications-1782679361640-h0usqf.webp","2026-06-28T16:39:00.000Z",{"id":1820,"slug":1821,"title":1822,"excerpt":1823,"featuredImage":1824,"publishedAt":1825},"455","zbt-z8102ax-dual-sim-failover-test","ZBT Z8102AX Dual-SIM failover: Šta radi, šta nedostaje i šta zahteva bolji firmver","ZBT Z8102AX je dual-SIM 5G OpenWrt ruter, ali sam dual-SIM hardver nije isto što i inteligentni failover. Ruter prepoznaje SIM karticu i uspešno se povezuje, ali automatsko prebacivanje, oporavak modema, odluke zasnovane na signalu i čista failover logika i dalje zahtevaju dublje testiranje.","\u002Fuploads\u002F2026\u002F06\u002Fopenwrt-router-review-dual-sim-03-1781620592829-7t77j7.webp","2026-06-16T10:40:00.000Z","fallback",[],[]]