{"id":45742,"date":"2026-11-17T21:04:04","date_gmt":"2026-11-17T18:04:04","guid":{"rendered":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/"},"modified":"2026-11-17T21:04:37","modified_gmt":"2026-11-17T18:04:37","slug":"respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum","status":"publish","type":"post","link":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/","title":{"rendered":"Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm"},"content":{"rendered":"<h2>Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm<\/h2>\n<p>B\u00fcy\u00fck Dil Modellerinin (LLM) y\u00fckseli\u015fi, yapay zeka uygulamalar\u0131n\u0131n geli\u015fimine inan\u0131lmaz bir ivme kazand\u0131rd\u0131. Ancak bu modelleri verimli, g\u00fcvenli ve maliyet etkin bir \u015fekilde y\u00f6netmek, bir\u00e7ok geli\u015ftirici ve \u015firket i\u00e7in karma\u015f\u0131k bir meydan okumaya d\u00f6n\u00fc\u015ft\u00fc. Peki, LLM&#8217;lerinizle etkile\u015fimi basitle\u015ftirmek, maliyetleri d\u00fc\u015f\u00fcrmek ve performans\u0131 art\u0131rmak m\u00fcmk\u00fcn m\u00fc? \u0130\u015fte bu noktada, Rust ile yaz\u0131lm\u0131\u015f a\u00e7\u0131k kaynakl\u0131 bir LLM ge\u00e7idi olan Respan Gateway devreye giriyor. Bu makalede, Respan Gateway&#8217;in ne oldu\u011funu, neden \u00f6nemli oldu\u011funu ve LLM altyap\u0131n\u0131z\u0131 nas\u0131l d\u00f6n\u00fc\u015ft\u00fcrebilece\u011fini ad\u0131m ad\u0131m inceleyece\u011fiz.<\/p>\n<h2>B\u00fcy\u00fck Dil Modellerini Y\u00f6netmek Neden Zorlu Bir G\u00f6rev?<\/h2>\n<p>G\u00fcn\u00fcm\u00fcz\u00fcn h\u0131zla geli\u015fen teknoloji d\u00fcnyas\u0131nda, yapay zeka (AI) ve \u00f6zellikle B\u00fcy\u00fck Dil Modelleri (LLM&#8217;ler), i\u015fletmelerin ve geli\u015ftiricilerin g\u00fcndeminde \u00f6nemli bir yer tutuyor. ChatGPT, GPT-4, LLaMA gibi modeller, metin \u00fcretme, \u00f6zetleme, \u00e7eviri ve karma\u015f\u0131k sorular\u0131 yan\u0131tlama gibi bir\u00e7ok alanda \u00e7\u0131\u011f\u0131r a\u00e7an yetenekler sunuyor. Ancak bu modellerin g\u00fcc\u00fcyle birlikte, onlar\u0131 etkin bir \u015fekilde y\u00f6netmek de ciddi zorluklar\u0131 beraberinde getiriyor. Bir LLM ile do\u011frudan etkile\u015fim kurmak, \u00f6zellikle b\u00fcy\u00fck \u00f6l\u00e7ekli uygulamalarda, beraberinde bir\u00e7ok operasyonel ve maliyetle ilgili sorunu getirebilir. \u00d6rne\u011fin, birden fazla LLM sa\u011flay\u0131c\u0131s\u0131yla \u00e7al\u0131\u015fmak istedi\u011finizde, her birinin kendine \u00f6zg\u00fc API (Uygulama Programlama Aray\u00fcz\u00fc) yap\u0131s\u0131, kimlik do\u011frulama mekanizmalar\u0131 ve h\u0131z s\u0131n\u0131rlamalar\u0131yla u\u011fra\u015fmak zorunda kal\u0131rs\u0131n\u0131z. Bu durum, geli\u015ftirme s\u00fcrecini karma\u015f\u0131kla\u015ft\u0131r\u0131r ve hata yapma olas\u0131l\u0131\u011f\u0131n\u0131 art\u0131r\u0131r.<\/p>\n<p>Maliyetler de g\u00f6z ard\u0131 edilemez bir fakt\u00f6r. LLM API \u00e7a\u011fr\u0131lar\u0131 genellikle token (jeton) bazl\u0131 \u00fccretlendirilir ve yo\u011fun kullan\u0131mlarda faturalar h\u0131zla artabilir. Yanl\u0131\u015f yap\u0131land\u0131r\u0131lm\u0131\u015f bir uygulama, gereksiz yere y\u00fczlerce veya binlerce API \u00e7a\u011fr\u0131s\u0131 yaparak b\u00fct\u00e7eyi a\u015fabilir. Performans ise bir di\u011fer kritik konudur. LLM&#8217;lerin yan\u0131t s\u00fcreleri, a\u011f gecikmeleri ve modelin karma\u015f\u0131kl\u0131\u011f\u0131na ba\u011fl\u0131 olarak de\u011fi\u015febilir. Kullan\u0131c\u0131lar\u0131n ak\u0131c\u0131 bir deneyim ya\u015fayabilmesi i\u00e7in bu yan\u0131t s\u00fcrelerinin optimize edilmesi gerekir. Ayr\u0131ca, g\u00fcvenlik ve g\u00f6zlemlenebilirlik (observability) de hayati \u00f6neme sahiptir. Hassas verilerin LLM&#8217;lere g\u00f6nderilmesi durumunda, veri g\u00fcvenli\u011fini sa\u011flamak ve API \u00e7a\u011fr\u0131lar\u0131n\u0131 izlemek, olas\u0131 sorunlar\u0131 tespit etmek i\u00e7in g\u00fc\u00e7l\u00fc mekanizmalara ihtiya\u00e7 duyulur. T\u00fcm bu zorluklar, LLM&#8217;leri kullanan uygulamalar\u0131n geli\u015ftirilmesi ve bak\u0131m\u0131 s\u00fcrecinde ciddi engeller olu\u015fturabilir. \u0130\u015fte bu noktada, bir LLM ge\u00e7idi (gateway) gibi merkezi bir \u00e7\u00f6z\u00fcm, bu karma\u015f\u0131kl\u0131\u011f\u0131 y\u00f6netmek i\u00e7in vazge\u00e7ilmez bir ara\u00e7 haline gelir.<\/p>\n<h2>LLM Gateway Nedir ve Neden \u0130htiya\u00e7 Duyar\u0131z?<\/h2>\n<p>LLM Gateway (ge\u00e7idi), B\u00fcy\u00fck Dil Modelleri ile uygulaman\u0131z aras\u0131nda bir arac\u0131 katman g\u00f6revi g\u00f6ren bir sunucudur. T\u0131pk\u0131 bir API ge\u00e7idinin mikroservislerinizi y\u00f6netti\u011fi gibi, bir LLM ge\u00e7idi de LLM sa\u011flay\u0131c\u0131lar\u0131n\u0131zla olan t\u00fcm etkile\u015fimi tek bir noktadan y\u00f6netmenizi sa\u011flar. Bu, geli\u015ftiriciler i\u00e7in hayat\u0131 b\u00fcy\u00fck \u00f6l\u00e7\u00fcde kolayla\u015ft\u0131r\u0131r ve LLM altyap\u0131s\u0131n\u0131n daha verimli, g\u00fcvenli ve \u00f6l\u00e7eklenebilir olmas\u0131n\u0131 sa\u011flar. Peki, bir LLM ge\u00e7idine neden ihtiya\u00e7 duyar\u0131z? Temel olarak, LLM&#8217;lerin sundu\u011fu potansiyeli tam olarak kullan\u0131rken, beraberindeki operasyonel y\u00fck\u00fc hafifletmek i\u00e7in. \u0130\u015fte ba\u015fl\u0131ca nedenler:<\/p>\n<ul>\n<li><strong>Tekille\u015ftirilmi\u015f Eri\u015fim Noktas\u0131:<\/strong> Farkl\u0131 LLM sa\u011flay\u0131c\u0131lar\u0131n\u0131n (OpenAI, Google, Anthropic vb.) farkl\u0131 API&#8217;lerini y\u00f6netmek yerine, t\u00fcm istekleri tek bir LLM ge\u00e7idi \u00fczerinden y\u00f6nlendirebilirsiniz. Bu, kodunuzu basitle\u015ftirir ve yeni bir sa\u011flay\u0131c\u0131 eklemenin veya mevcut birini de\u011fi\u015ftirmenin maliyetini azalt\u0131r.<\/li>\n<li><strong>Maliyet Optimizasyonu:<\/strong> Ge\u00e7itler, \u00f6nbellekleme (caching) mekanizmalar\u0131 sunarak ayn\u0131 isteklere tekrar tekrar LLM API \u00e7a\u011fr\u0131s\u0131 yap\u0131lmas\u0131n\u0131 engeller. Bu, \u00f6zellikle s\u0131k tekrarlanan veya statik yan\u0131tlar i\u00e7in b\u00fcy\u00fck \u00f6l\u00e7\u00fcde maliyet tasarrufu sa\u011flar. Ayr\u0131ca, istekleri en uygun maliyetli modele veya sa\u011flay\u0131c\u0131ya y\u00f6nlendirme (routing) yetene\u011fi de sunabilirler.<\/li>\n<li><strong>Performans Art\u0131\u015f\u0131:<\/strong> \u00d6nbellekleme, LLM yan\u0131t s\u00fcrelerini \u00f6nemli \u00f6l\u00e7\u00fcde h\u0131zland\u0131rabilir. Ayr\u0131ca, ge\u00e7itler genellikle y\u00fcksek performansl\u0131 dillerle yaz\u0131ld\u0131\u011f\u0131 i\u00e7in, istekleri daha h\u0131zl\u0131 i\u015fleyebilir ve gecikmeyi (latency) azaltabilirler.<\/li>\n<li><strong>G\u00fcvenlik ve Kimlik Do\u011frulama:<\/strong> T\u00fcm LLM istekleri tek bir ge\u00e7it \u00fczerinden ge\u00e7ti\u011fi i\u00e7in, API anahtarlar\u0131n\u0131z\u0131 do\u011frudan uygulamalar\u0131n\u0131za g\u00f6mmek yerine ge\u00e7itte g\u00fcvenli bir \u015fekilde y\u00f6netebilirsiniz. Ge\u00e7it, kimlik do\u011frulama (authentication) ve yetkilendirme (authorization) katmanlar\u0131 ekleyerek g\u00fcvenlik duru\u015funuzu g\u00fc\u00e7lendirir.<\/li>\n<li><strong>Oran S\u0131n\u0131rlama (Rate Limiting):<\/strong> LLM sa\u011flay\u0131c\u0131lar\u0131n\u0131n API&#8217;lerinde genellikle belirli bir zaman diliminde yapabilece\u011finiz istek say\u0131s\u0131na ili\u015fkin s\u0131n\u0131rlamalar bulunur. Bir ge\u00e7it, bu s\u0131n\u0131rlamalar\u0131 uygulaman\u0131z\u0131n a\u015fmas\u0131n\u0131 engelleyerek hatalar\u0131 ve hizmet kesintilerini \u00f6nler.<\/li>\n<li><strong>G\u00f6zlemlenebilirlik (Observability):<\/strong> T\u00fcm LLM istekleri ve yan\u0131tlar\u0131 ge\u00e7it \u00fczerinden ge\u00e7ti\u011fi i\u00e7in, merkezi bir noktadan loglama (g\u00fcnl\u00fc\u011fe kaydetme), metrik toplama ve izleme (monitoring) yapabilirsiniz. Bu, sorunlar\u0131 te\u015fhis etmeyi, performans\u0131 analiz etmeyi ve kullan\u0131m e\u011filimlerini anlamay\u0131 kolayla\u015ft\u0131r\u0131r.<\/li>\n<li><strong>Y\u00fck Dengeleme ve Y\u00f6nlendirme:<\/strong> Birden fazla LLM sa\u011flay\u0131c\u0131s\u0131 veya model aras\u0131nda istekleri ak\u0131ll\u0131ca y\u00f6nlendirerek y\u00fck dengelemesi yapabilir ve her zaman en uygun veya en eri\u015filebilir kayna\u011f\u0131 kullanabilirsiniz.<\/li>\n<\/ul>\n<p>Respan Gateway&#8217;in Rust ile yaz\u0131lm\u0131\u015f olmas\u0131, bu faydalar\u0131 daha da peki\u015ftiriyor. Rust, y\u00fcksek performans, bellek g\u00fcvenli\u011fi ve e\u015fzamanl\u0131l\u0131k (concurrency) yetenekleriyle bilinen bir programlama dilidir. Bu \u00f6zellikler, Respan gibi bir ge\u00e7idin kritik g\u00f6revleri h\u0131zl\u0131 ve g\u00fcvenilir bir \u015fekilde yerine getirmesini sa\u011flar. A\u00e7\u0131k kaynak olmas\u0131 ise toplulu\u011fun katk\u0131lar\u0131yla s\u00fcrekli geli\u015fmesini ve \u015feffaf olmas\u0131n\u0131 garanti eder. Bu sayede, kurumlar kendi \u00f6zel ihtiya\u00e7lar\u0131na g\u00f6re Respan&#8217;\u0131 \u00f6zelle\u015ftirebilir ve denetleyebilirler.<\/p>\n<h2>Respan Gateway&#8217;i Kurulumu ve \u0130lk Ad\u0131mlar<\/h2>\n<p>Respan Gateway&#8217;i kullanmaya ba\u015flamak, Rust ekosistemine a\u015fina olanlar i\u00e7in olduk\u00e7a basit bir s\u00fcre\u00e7tir. E\u011fer Rust ile daha \u00f6nce \u00e7al\u0131\u015fmad\u0131ysan\u0131z bile endi\u015felenmeyin, ad\u0131mlar\u0131 takip ederek kolayca kurulum yapabilirsiniz. Respan&#8217;\u0131 kendi sisteminizde \u00e7al\u0131\u015ft\u0131rmak i\u00e7in \u00f6ncelikle Rust programlama dilinin geli\u015ftirme ortam\u0131na (toolchain) ihtiyac\u0131m\u0131z var.<\/p>\n<h3>Respan&#8217;\u0131 Nas\u0131l Derleyip \u00c7al\u0131\u015ft\u0131r\u0131rs\u0131n\u0131z?<\/h3>\n<p><strong>1. Rust Kurulumu:<\/strong><br \/>\nE\u011fer sisteminizde Rust y\u00fckl\u00fc de\u011filse, resmi Rustup arac\u0131yla kurulum yapman\u0131z gerekmektedir. Terminalinizi a\u00e7\u0131n ve a\u015fa\u011f\u0131daki komutu \u00e7al\u0131\u015ft\u0131r\u0131n:<\/p>\n<div class=\"code-container\">\n<pre><code>curl --proto '=https' --tlsv1.2 -sSf https:\/\/sh.rustup.rs | sh<\/code><\/pre>\n<\/div>\n<p>Kurulum tamamland\u0131ktan sonra, Rustup&#8217;\u0131n size \u00f6nerdi\u011fi komutu \u00e7al\u0131\u015ft\u0131rarak PATH de\u011fi\u015fkeninizi g\u00fcncelleyin:<\/p>\n<div class=\"code-container\">\n<pre><code>source $HOME\/.cargo\/env<\/code><\/pre>\n<\/div>\n<p>Bu komut, Rust derleyicisi (<code>rustc<\/code>) ve paket y\u00f6neticisi (<code>cargo<\/code>) gibi ara\u00e7lar\u0131 kullan\u0131ma haz\u0131r hale getirecektir.<\/p>\n<p><strong>2. Respan Gateway Kaynak Kodunu \u0130ndirme:<\/strong><br \/>\nRespan Gateway a\u00e7\u0131k kaynakl\u0131 bir proje oldu\u011fu i\u00e7in, kaynak kodunu GitHub deposundan klonlamam\u0131z gerekiyor. Bunun i\u00e7in Git&#8217;in sisteminizde y\u00fckl\u00fc olmas\u0131 gerekmektedir:<\/p>\n<div class=\"code-container\">\n<pre><code>git clone https:\/\/github.com\/respan-ai\/respan.git\ncd respan<\/code><\/pre>\n<\/div>\n<p>Bu komutlar, Respan projesini bilgisayar\u0131n\u0131za indirir ve proje dizinine ge\u00e7i\u015f yapar.<\/p>\n<p><strong>3. Respan&#8217;\u0131 Derleme:<\/strong><br \/>\nProje dizininde iken, Cargo kullanarak Respan&#8217;\u0131 derleyebilirsiniz. \u0130lk derleme biraz zaman alabilir, \u00e7\u00fcnk\u00fc Cargo gerekli t\u00fcm ba\u011f\u0131ml\u0131l\u0131klar\u0131 indirip derleyecektir:<\/p>\n<div class=\"code-container\">\n<pre><code>cargo build --release<\/code><\/pre>\n<\/div>\n<p><code>--release<\/code> bayra\u011f\u0131, daha optimize edilmi\u015f ve performansl\u0131 bir s\u00fcr\u00fcm derlememizi sa\u011flar. \u00dcretim ortamlar\u0131 i\u00e7in bu bayra\u011f\u0131 kullanmak \u00f6nemlidir. Derleme i\u015flemi ba\u015far\u0131yla tamamland\u0131\u011f\u0131nda, \u00e7al\u0131\u015ft\u0131r\u0131labilir dosya <code>target\/release\/respan<\/code> konumunda bulunacakt\u0131r.<\/p>\n<p><strong>4. Respan&#8217;\u0131 \u00c7al\u0131\u015ft\u0131rma:<\/strong><br \/>\nDerlenmi\u015f Respan uygulamas\u0131n\u0131 \u00e7al\u0131\u015ft\u0131rmak i\u00e7in a\u015fa\u011f\u0131daki komutu kullanabilirsiniz:<\/p>\n<div class=\"code-container\">\n<pre><code>.\/target\/release\/respan<\/code><\/pre>\n<\/div>\n<p>Varsay\u0131lan olarak, Respan belirli bir portta dinlemeye ba\u015flayacak ve yap\u0131land\u0131rma dosyas\u0131n\u0131 arayacakt\u0131r. Ancak, hen\u00fcz bir yap\u0131land\u0131rma dosyam\u0131z olmad\u0131\u011f\u0131 i\u00e7in, bu ad\u0131mda temel bir hata veya varsay\u0131lan bir \u00e7al\u0131\u015fma modu g\u00f6rebilirsiniz. Bu bizi bir sonraki ad\u0131ma, yani yap\u0131land\u0131rmaya getiriyor.<\/p>\n<h3>Temel Yap\u0131land\u0131rma: \u0130lk LLM Entegrasyonu Nas\u0131l Yap\u0131l\u0131r?<\/h3>\n<p>Respan Gateway, yap\u0131land\u0131rma i\u00e7in genellikle bir <code>config.toml<\/code> dosyas\u0131 kullan\u0131r. Bu dosya, Respan&#8217;\u0131n hangi portta dinleyece\u011fi, hangi LLM sa\u011flay\u0131c\u0131lar\u0131na istekleri y\u00f6nlendirece\u011fi, \u00f6nbellekleme ayarlar\u0131 ve di\u011fer t\u00fcm davran\u0131\u015flar\u0131 tan\u0131mlar. \u0130\u015fte basit bir <code>config.toml<\/code> \u00f6rne\u011fi:<\/p>\n<div class=\"code-container\">\n<pre><code># config.toml dosyas\u0131\nlisten_address = \"0.0.0.0:8000\"\n\n[llm_providers.openai]\napi_key = \"sk-YOUR_OPENAI_API_KEY\" # Ger\u00e7ek API anahtar\u0131n\u0131z\u0131 buraya yaz\u0131n\nbase_url = \"https:\/\/api.openai.com\/v1\"\n\n[[routes]]\npath = \"\/v1\/chat\/completions\"\nprovider = \"openai\"\nmodel_mapping = { \"gpt-4-turbo\": \"gpt-4-turbo-preview\", \"gpt-3.5-turbo\": \"gpt-3.5-turbo\" }\n\n[cache]\nenabled = true\nttl_seconds = 300 # 5 dakika \u00f6nbellekleme s\u00fcresi\nmax_size_mb = 100 # \u00d6nbellek boyutu s\u0131n\u0131r\u0131<\/code><\/pre>\n<\/div>\n<p>Bu yap\u0131land\u0131rma dosyas\u0131n\u0131 projenizin ana dizinine (<code>respan<\/code> klas\u00f6r\u00fcne) kaydedin ve <code>config.toml<\/code> olarak adland\u0131r\u0131n. Daha sonra Respan&#8217;\u0131 tekrar \u00e7al\u0131\u015ft\u0131r\u0131n:<\/p>\n<div class=\"code-container\">\n<pre><code>.\/target\/release\/respan<\/code><\/pre>\n<\/div>\n<p>Art\u0131k Respan, 8000 portunda gelen istekleri dinleyecek ve <code>\/v1\/chat\/completions<\/code> yoluna gelen istekleri OpenAI sa\u011flay\u0131c\u0131s\u0131na y\u00f6nlendirecektir. <code>model_mapping<\/code> k\u0131sm\u0131, sizin uygulaman\u0131z\u0131n kulland\u0131\u011f\u0131 model ad\u0131n\u0131, sa\u011flay\u0131c\u0131n\u0131n bekledi\u011fi model ad\u0131na d\u00f6n\u00fc\u015ft\u00fcrmenizi sa\u011flar. \u00d6rne\u011fin, uygulaman\u0131z <code>gpt-4-turbo<\/code> \u00e7a\u011f\u0131rsa bile, Respan bunu OpenAI&#8217;nin g\u00fcncel ad\u0131 olan <code>gpt-4-turbo-preview<\/code> olarak de\u011fi\u015ftirebilir.<\/p>\n<p>Bu temel kurulumla birlikte, art\u0131k uygulaman\u0131zdan do\u011frudan OpenAI API&#8217;sine istek yapmak yerine, Respan Gateway&#8217;e istek g\u00f6nderebilirsiniz. \u00d6rne\u011fin, Python&#8217;da bir istek yapmak isterseniz:<\/p>\n<div class=\"code-container\">\n<pre><code>import requests\nimport json\n\nheaders = {\n    \"Content-Type\": \"application\/json\",\n    # Respan'a \u00f6zel bir kimlik do\u011frulama anahtar\u0131 ekleyebilirsiniz,\n    # ancak \u015fimdilik bu \u00f6rnekte gerekli de\u011fil.\n}\n\ndata = {\n    \"model\": \"gpt-3.5-turbo\",\n    \"messages\": [\n        {\"role\": \"system\", \"content\": \"You are a helpful assistant.\"},\n        {\"role\": \"user\", \"content\": \"Merhaba, nas\u0131ls\u0131n?\"}\n    ]\n}\n\nresponse = requests.post(\"http:\/\/localhost:8000\/v1\/chat\/completions\", headers=headers, data=json.dumps(data))\n\nprint(response.json())<\/code><\/pre>\n<\/div>\n<p>Bu \u00f6rnekte, iste\u011fi do\u011frudan <code>api.openai.com<\/code> yerine <code>localhost:8000<\/code> adresine g\u00f6nderiyoruz. Respan bu iste\u011fi alacak, yap\u0131land\u0131rmaya g\u00f6re i\u015fleyecek (\u00f6rne\u011fin \u00f6nbelle\u011fe bakacak, yoksa OpenAI&#8217;ye y\u00f6nlendirecek) ve yan\u0131t\u0131 bize geri d\u00f6nd\u00fcrecektir. Bu basit ad\u0131mlarla Respan Gateway&#8217;in g\u00fcc\u00fcn\u00fc ke\u015ffetmeye ba\u015flayabilirsiniz.<\/p>\n<h2>Respan ile Geli\u015fmi\u015f LLM Y\u00f6netimi Senaryolar\u0131<\/h2>\n<p>Respan Gateway&#8217;in temel kurulumunu tamamlad\u0131ktan sonra, \u015fimdi daha geli\u015fmi\u015f y\u00f6netim senaryolar\u0131na g\u00f6z atal\u0131m. Bir LLM ge\u00e7idinin ger\u00e7ek de\u011feri, sadece istekleri y\u00f6nlendirmekten \u00f6te, karma\u015f\u0131k i\u015f ak\u0131\u015flar\u0131n\u0131 basitle\u015ftirmesi ve operasyonel verimlili\u011fi art\u0131rmas\u0131yla ortaya \u00e7\u0131kar. Respan, bu alanda Rust&#8217;\u0131n getirdi\u011fi performans ve g\u00fcvenlikle birlikte g\u00fc\u00e7l\u00fc yetenekler sunar.<\/p>\n<h3>Farkl\u0131 LLM Sa\u011flay\u0131c\u0131lar\u0131 Aras\u0131nda Ak\u0131ll\u0131 Y\u00f6nlendirme (Routing) Nas\u0131l Yap\u0131l\u0131r?<\/h3>\n<p>Uygulaman\u0131z\u0131n farkl\u0131 ihtiya\u00e7lar\u0131 i\u00e7in birden fazla LLM sa\u011flay\u0131c\u0131s\u0131 veya modeli kullanmas\u0131 yayg\u0131n bir senaryodur. \u00d6rne\u011fin, maliyet etkinli\u011fi i\u00e7in d\u00fc\u015f\u00fck kaliteli istekleri daha ucuz bir modele, y\u00fcksek kaliteli veya kritik istekleri ise daha pahal\u0131 ama yetenekli bir modele y\u00f6nlendirmek isteyebilirsiniz. Respan, yap\u0131land\u0131rma \u00fczerinden esnek y\u00f6nlendirme kurallar\u0131 tan\u0131mlaman\u0131za olanak tan\u0131r. A\u015fa\u011f\u0131daki \u00f6rnek, farkl\u0131 yollara (path) veya hatta iste\u011fin i\u00e7eri\u011fine (\u00f6rne\u011fin, kullan\u0131lan modele) g\u00f6re y\u00f6nlendirme yapmay\u0131 g\u00f6sterir:<\/p>\n<div class=\"code-container\">\n<pre><code># config.toml\n# ... (\u00f6nceki provider tan\u0131mlamalar\u0131) ...\n\n[llm_providers.google]\napi_key = \"YOUR_GOOGLE_API_KEY\"\nbase_url = \"https:\/\/generativelanguage.googleapis.com\/v1beta\" # \u00d6rnek bir Google API URL'si\n\n[[routes]]\npath = \"\/v1\/chat\/completions\"\nprovider = \"openai\"\nmodel_mapping = { \"gpt-4-turbo\": \"gpt-4-turbo-preview\", \"gpt-3.5-turbo\": \"gpt-3.5-turbo\" }\n\n[[routes]]\npath = \"\/v1\/google\/generate\" # Google i\u00e7in \u00f6zel bir yol\nprovider = \"google\"\nmodel_mapping = { \"gemini-pro\": \"gemini-pro\" }\n\n[[routes]]\npath = \"\/v1\/smart\/completions\" # Ak\u0131ll\u0131 y\u00f6nlendirme \u00f6rne\u011fi\n# Bu rotada, iste\u011fin i\u00e7eri\u011fine g\u00f6re hangi sa\u011flay\u0131c\u0131ya gidece\u011fine karar verebiliriz.\n# Respan'\u0131n geli\u015fmi\u015f y\u00f6nlendirme \u00f6zellikleri bu t\u00fcr mant\u0131\u011f\u0131 destekleyebilir.\n# \u00d6rne\u011fin, request body'deki 'model' alan\u0131na g\u00f6re dinamik y\u00f6nlendirme.\n# Bu k\u0131s\u0131m, Respan'\u0131n daha detayl\u0131 yap\u0131land\u0131rma dok\u00fcmantasyonuna g\u00f6re \u015fekillenecektir.\n<\/code><\/pre>\n<\/div>\n<p>Bu yap\u0131land\u0131rma ile <code>\/v1\/chat\/completions<\/code> adresine gelen istekler OpenAI&#8217;ye giderken, <code>\/v1\/google\/generate<\/code> adresine gelen istekler Google&#8217;a y\u00f6nlendirilebilir. Uygulaman\u0131z\u0131n hangi LLM&#8217;i kullanaca\u011f\u0131na karar vermesi, Respan&#8217;\u0131n arkas\u0131ndaki karma\u015f\u0131kl\u0131\u011f\u0131 y\u00f6netmesini sa\u011flar.<\/p>\n<h3>Maliyet ve Performans Optimizasyonu \u0130\u00e7in \u00d6nbellekleme (Caching) Nas\u0131l Kullan\u0131l\u0131r?<\/h3>\n<p>\u00d6nbellekleme, LLM maliyetlerini d\u00fc\u015f\u00fcrmenin ve yan\u0131t s\u00fcrelerini k\u0131saltman\u0131n en etkili yollar\u0131ndan biridir. Ayn\u0131 veya benzer istekler i\u00e7in LLM API&#8217;sine tekrar tekrar \u00e7a\u011fr\u0131 yapmak yerine, daha \u00f6nce al\u0131nm\u0131\u015f yan\u0131tlar\u0131 \u00f6nbellekten sunabiliriz. Respan, yap\u0131land\u0131rma dosyas\u0131nda basit ayarlarla \u00f6nbelleklemeyi etkinle\u015ftirmenize olanak tan\u0131r:<\/p>\n<div class=\"code-container\">\n<pre><code># config.toml\n# ... (di\u011fer ayarlar) ...\n\n[cache]\nenabled = true\nttl_seconds = 3600 # \u00d6nbellekteki \u00f6\u011felerin 1 saat (3600 saniye) boyunca ge\u00e7erli olmas\u0131\nmax_size_mb = 500 # \u00d6nbelle\u011fin maksimum boyutu 500 MB\n# cache_type = \"in_memory\" # veya \"redis\", \"memcached\" gibi farkl\u0131 \u00f6nbellek t\u00fcrleri\n<\/code><\/pre>\n<\/div>\n<p>Bu ayarlar, Respan&#8217;\u0131n gelen LLM isteklerini ve yan\u0131tlar\u0131n\u0131 belirli bir s\u00fcre ve boyut limitine g\u00f6re \u00f6nbelle\u011fe almas\u0131n\u0131 sa\u011flar. Bir istek geldi\u011finde, Respan \u00f6nce \u00f6nbelle\u011fe bakacak, e\u011fer ge\u00e7erli bir yan\u0131t bulursa do\u011frudan onu d\u00f6nd\u00fcrecektir. Bu, hem LLM sa\u011flay\u0131c\u0131s\u0131na yap\u0131lan API \u00e7a\u011fr\u0131s\u0131 say\u0131s\u0131n\u0131 azaltarak maliyetten tasarruf etmenizi sa\u011flar hem de yan\u0131t s\u00fcrelerini d\u00fc\u015f\u00fcrerek kullan\u0131c\u0131 deneyimini iyile\u015ftirir.<\/p>\n<h3>A\u015f\u0131r\u0131 Y\u00fcklenmeyi \u00d6nlemek \u0130\u00e7in Oran S\u0131n\u0131rlama (Rate Limiting) Nas\u0131l Uygulan\u0131r?<\/h3>\n<p>LLM sa\u011flay\u0131c\u0131lar\u0131n\u0131n \u00e7o\u011fu, API kullan\u0131mlar\u0131n\u0131 belirli oranlarla s\u0131n\u0131rlar (\u00f6rne\u011fin, dakikada X istek). Uygulaman\u0131z bu limitleri a\u015farsa, hizmet kesintileri veya hatal\u0131 yan\u0131tlarla kar\u015f\u0131la\u015fabilirsiniz. Respan, uygulaman\u0131z\u0131n bu limitleri a\u015fmas\u0131n\u0131 engelleyen bir oran s\u0131n\u0131rlama katman\u0131 eklemenize olanak tan\u0131r:<\/p>\n<div class=\"code-container\">\n<pre><code># config.toml\n# ... (di\u011fer ayarlar) ...\n\n[rate_limit]\nenabled = true\n# Per-minute limits\nrequests_per_minute = 600 # Dakikada maksimum 600 istek\ntokens_per_minute = 1_000_000 # Dakikada maksimum 1 milyon token\n# Per-hour limits\nrequests_per_hour = 30_000 # Saatte maksimum 30 bin istek\n# ... daha fazla detayl\u0131 yap\u0131land\u0131rma olabilir (\u00f6rne\u011fin, kullan\u0131c\u0131 baz\u0131nda limitler)\n<\/code><\/pre>\n<\/div>\n<p>Bu yap\u0131land\u0131rma ile Respan, gelen istekleri belirledi\u011finiz limitlere g\u00f6re kontrol eder. Bir limit a\u015f\u0131ld\u0131\u011f\u0131nda, Respan iste\u011fi LLM sa\u011flay\u0131c\u0131s\u0131na g\u00f6ndermek yerine, do\u011frudan bir hata yan\u0131t\u0131 (\u00f6rne\u011fin, HTTP 429 Too Many Requests) d\u00f6nd\u00fcr\u00fcr. Bu, hem LLM sa\u011flay\u0131c\u0131s\u0131n\u0131n API limitlerine uyum sa\u011flaman\u0131za yard\u0131mc\u0131 olur hem de uygulaman\u0131z\u0131n beklenmedik hatalarla kar\u015f\u0131la\u015fmas\u0131n\u0131 engeller.<\/p>\n<h3>G\u00fcvenlik \u0130\u00e7in Kimlik Do\u011frulama ve Yetkilendirme (Authentication &#038; Authorization) Nas\u0131l Sa\u011flan\u0131r?<\/h3>\n<p>Respan Gateway, LLM API anahtarlar\u0131n\u0131z\u0131 do\u011frudan uygulaman\u0131za g\u00f6mmek yerine, merkezi bir noktada y\u00f6netmenizi sa\u011flar. Ayr\u0131ca, kendi kimlik do\u011frulama mekanizmalar\u0131n\u0131z\u0131 ekleyerek uygulaman\u0131z\u0131n LLM&#8217;lere eri\u015fimini daha g\u00fcvenli hale getirebilirsiniz. \u00d6rne\u011fin, Respan&#8217;a gelen isteklerin belirli bir API anahtar\u0131 veya JWT (JSON Web Token) i\u00e7ermesini zorunlu k\u0131labilirsiniz:<\/p>\n<div class=\"code-container\">\n<pre><code># config.toml\n# ... (di\u011fer ayarlar) ...\n\n[auth]\nenabled = true\n# api_key_header = \"X-Respan-API-Key\"\n# valid_api_keys = [\"my-secret-key-123\", \"another-secret-key-456\"]\n# jwt_public_key_path = \"\/etc\/respan\/jwt_public.pem\" # JWT do\u011frulama i\u00e7in\n<\/code><\/pre>\n<\/div>\n<p>Bu \u00f6zellikler sayesinde, sadece yetkili uygulamalar\u0131n veya kullan\u0131c\u0131lar\u0131n LLM&#8217;lere Respan \u00fczerinden eri\u015febilmesini sa\u011flayabilirsiniz. Bu, g\u00fcvenlik duru\u015funuzu \u00f6nemli \u00f6l\u00e7\u00fcde g\u00fc\u00e7lendirir ve hassas API anahtarlar\u0131n\u0131n if\u015fa olma riskini azalt\u0131r.<\/p>\n<h3>G\u00f6zlemlenebilirlik (Observability) ile LLM Kullan\u0131m\u0131n\u0131 \u0130zleme<\/h3>\n<p>Bir LLM ge\u00e7idi, t\u00fcm LLM trafi\u011finin merkezi bir noktadan ge\u00e7mesi nedeniyle m\u00fckemmel bir g\u00f6zlemlenebilirlik noktas\u0131 sunar. Respan, loglama ve metrik toplama yetenekleri arac\u0131l\u0131\u011f\u0131yla LLM kullan\u0131m\u0131n\u0131z\u0131 derinlemesine anlaman\u0131za yard\u0131mc\u0131 olabilir. \u00d6rne\u011fin, her iste\u011fin yan\u0131t s\u00fcresi, kullan\u0131lan token say\u0131s\u0131, \u00f6nbellek isabet oranlar\u0131 gibi metrikleri toplayarak, LLM altyap\u0131n\u0131z\u0131n performans\u0131n\u0131 ve maliyetini optimize edebilirsiniz. Respan&#8217;\u0131n loglar\u0131, olas\u0131 sorunlar\u0131 h\u0131zl\u0131ca tespit etmenize ve gidermenize yard\u0131mc\u0131 olur.<\/p>\n<p><strong>Vaka Analizi: Bir E-ticaret Sitesinin M\u00fc\u015fteri Hizmetleri Chatbotu<\/strong><br \/>\nBir e-ticaret \u015firketi, m\u00fc\u015fteri hizmetleri i\u00e7in bir yapay zeka destekli chatbot kullan\u0131yor. Bu chatbot, m\u00fc\u015fterilerin sorular\u0131n\u0131 yan\u0131tlamak, sipari\u015f durumlar\u0131n\u0131 kontrol etmek ve \u00fcr\u00fcn \u00f6nerileri sunmak i\u00e7in birden fazla LLM sa\u011flay\u0131c\u0131s\u0131n\u0131 kullan\u0131yor. \u00d6rne\u011fin, basit sorular i\u00e7in daha ucuz bir model (GPT-3.5 Turbo), karma\u015f\u0131k veya hassas m\u00fc\u015fteri sorunlar\u0131 i\u00e7in ise daha geli\u015fmi\u015f ve pahal\u0131 bir model (GPT-4) kullan\u0131l\u0131yor. Respan Gateway burada merkezi bir rol oynuyor:<\/p>\n<ul>\n<li><strong>Ak\u0131ll\u0131 Y\u00f6nlendirme:<\/strong> Chatbot, gelen m\u00fc\u015fteri sorusunun karma\u015f\u0131kl\u0131\u011f\u0131n\u0131 analiz eder ve iste\u011fi Respan&#8217;a g\u00f6nderir. Respan, \u00f6nceden tan\u0131mlanm\u0131\u015f kurallara g\u00f6re (\u00f6rne\u011fin, soru anahtar kelimeler i\u00e7eriyorsa veya belirli bir uzunlu\u011fu a\u015f\u0131yorsa), iste\u011fi uygun LLM modeline y\u00f6nlendirir.<\/li>\n<li><strong>\u00d6nbellekleme:<\/strong> S\u0131k\u00e7a sorulan sorular\u0131n (\u00f6rne\u011fin, &#8220;Kargom nerede?&#8221;, &#8220;\u0130ade politikas\u0131 nedir?&#8221;) yan\u0131tlar\u0131 Respan taraf\u0131ndan \u00f6nbelle\u011fe al\u0131n\u0131r. Bu sayede, ayn\u0131 soru tekrar soruldu\u011funda, LLM&#8217;ye yeni bir \u00e7a\u011fr\u0131 yap\u0131lmaz, yan\u0131t an\u0131nda \u00f6nbellekten d\u00f6nd\u00fcr\u00fcl\u00fcr. Bu, hem yan\u0131t s\u00fcresini milisaniyelere d\u00fc\u015f\u00fcr\u00fcr hem de API maliyetlerinden %30&#8217;a varan tasarruf sa\u011flar.<\/li>\n<li><strong>Oran S\u0131n\u0131rlama:<\/strong> Tatil d\u00f6nemlerinde veya kampanya zamanlar\u0131nda m\u00fc\u015fteri trafi\u011fi zirve yapt\u0131\u011f\u0131nda, chatbotun LLM API limitlerini a\u015fma riski vard\u0131r. Respan, bu limitleri izler ve olas\u0131 a\u015f\u0131m durumlar\u0131nda, iste\u011fi ge\u00e7ici olarak kuyru\u011fa al\u0131r veya daha d\u00fc\u015f\u00fck \u00f6ncelikli istekleri reddeder, b\u00f6ylece kritik m\u00fc\u015fteri hizmetleri ak\u0131\u015f\u0131n\u0131n kesintisiz devam etmesini sa\u011flar.<\/li>\n<li><strong>G\u00fcvenlik ve \u0130zleme:<\/strong> M\u00fc\u015fteri verileri hassas oldu\u011fu i\u00e7in, Respan t\u00fcm LLM isteklerini izler ve loglar. Herhangi bir anormal kullan\u0131m deseni veya g\u00fcvenlik a\u00e7\u0131\u011f\u0131 denemesi durumunda uyar\u0131lar tetiklenir. Ayr\u0131ca, t\u00fcm API anahtarlar\u0131 Respan&#8217;da g\u00fcvenli bir \u015fekilde saklan\u0131r ve chatbot uygulamas\u0131n\u0131n do\u011frudan eri\u015fimi engellenir.<\/li>\n<\/ul>\n<p>Bu senaryo, Respan Gateway&#8217;in bir LLM altyap\u0131s\u0131n\u0131 nas\u0131l daha esnek, maliyet etkin ve g\u00fcvenli hale getirebilece\u011fini a\u00e7\u0131k\u00e7a g\u00f6stermektedir. Geli\u015ftiriciler, LLM&#8217;lerin karma\u015f\u0131k y\u00f6netimi yerine, do\u011frudan i\u015f mant\u0131\u011f\u0131na odaklanabilirler.<\/p>\n<h2>Respan Gateway&#8217;i Maksimum Verimlilikle Kullanmak<\/h2>\n<p>Respan Gateway, temel \u00f6zelliklerinin \u00f6tesinde, deneyimli kullan\u0131c\u0131lar ve \u00f6zel ihtiya\u00e7lar\u0131 olan kurulu\u015flar i\u00e7in daha derinlemesine optimizasyon ve \u00f6zelle\u015ftirme imkanlar\u0131 sunar. Rust&#8217;\u0131n mod\u00fcler yap\u0131s\u0131 ve performans\u0131 sayesinde, Respan&#8217;\u0131 kendi i\u015f ak\u0131\u015flar\u0131n\u0131za m\u00fckemmel \u015fekilde entegre edebilirsiniz.<\/p>\n<h3>\u00d6zel Eklentiler (Plugins) veya Middleware Geli\u015ftirme<\/h3>\n<p>Respan Gateway, muhtemelen bir eklenti (plugin) veya middleware (ara yaz\u0131l\u0131m) mimarisine sahip olacakt\u0131r. Bu, geli\u015ftiricilerin Respan&#8217;\u0131n \u00e7ekirdek kodunu de\u011fi\u015ftirmeden, kendi \u00f6zel i\u015f mant\u0131klar\u0131n\u0131 veya entegrasyonlar\u0131n\u0131 eklemelerine olanak tan\u0131r. \u00d6rne\u011fin:<\/p>\n<ul>\n<li><strong>\u00d6zel \u0130stek D\u00f6n\u00fc\u015f\u00fcmleri:<\/strong> Gelen bir LLM iste\u011fini, LLM sa\u011flay\u0131c\u0131s\u0131na g\u00f6ndermeden \u00f6nce belirli bir formata d\u00f6n\u00fc\u015ft\u00fcrmeniz gerekebilir. Bir middleware yazarak bu d\u00f6n\u00fc\u015f\u00fcm\u00fc otomatik olarak yapabilirsiniz.<\/li>\n<li><strong>Dinamik Y\u00f6nlendirme Mant\u0131\u011f\u0131:<\/strong> Statik yap\u0131land\u0131rma dosyalar\u0131n\u0131n \u00f6tesinde, iste\u011fin i\u00e7eri\u011fine, kullan\u0131c\u0131n\u0131n rol\u00fcne veya harici bir veritaban\u0131ndan al\u0131nan bilgilere g\u00f6re dinamik olarak LLM sa\u011flay\u0131c\u0131s\u0131 se\u00e7mek isteyebilirsiniz. \u00d6zel bir eklenti, bu karma\u015f\u0131k y\u00f6nlendirme mant\u0131\u011f\u0131n\u0131 uygulayabilir.<\/li>\n<li><strong>Maliyet Tahmini ve B\u00fct\u00e7e Kontrol\u00fc:<\/strong> Her LLM iste\u011fi i\u00e7in beklenen maliyeti tahmin eden ve belirli bir b\u00fct\u00e7eyi a\u015fma durumunda uyar\u0131 veren veya istekleri engelleyen bir eklenti geli\u015ftirebilirsiniz.<\/li>\n<li><strong>Veri Maskeleme\/Anonimle\u015ftirme:<\/strong> Hassas m\u00fc\u015fteri verilerini LLM&#8217;lere g\u00f6ndermeden \u00f6nce otomatik olarak maskeleyen veya anonimle\u015ftiren bir middleware, veri gizlili\u011fi (data privacy) endi\u015felerini giderebilir.<\/li>\n<\/ul>\n<p>Rust&#8217;\u0131n g\u00fc\u00e7l\u00fc tip sistemi ve performans odakl\u0131 do\u011fas\u0131, bu t\u00fcr eklentilerin hem g\u00fcvenli hem de h\u0131zl\u0131 olmas\u0131n\u0131 sa\u011flar. Geli\u015ftiriciler, Respan&#8217;\u0131n sundu\u011fu API&#8217;leri kullanarak kendi \u00f6zel eklentilerini olu\u015fturabilir ve LLM ge\u00e7idinin yeteneklerini kendi \u00f6zel gereksinimlerine g\u00f6re geni\u015fletebilirler.<\/p>\n<div class=\"code-container\">\n<pre><code>\/\/ \u00d6rnek bir Rust middleware yap\u0131s\u0131 (konseptsel)\n\/\/ Respan'\u0131n ger\u00e7ek API'si farkl\u0131l\u0131k g\u00f6sterebilir.\n\npub trait RespanMiddleware {\n    fn process_request(&self, request: &mut LLMRequest) -> Result<(), MiddlewareError>;\n    fn process_response(&self, response: &mut LLMResponse) -> Result<(), MiddlewareError>;\n}\n\npub struct DataMaskingMiddleware;\n\nimpl RespanMiddleware for DataMaskingMiddleware {\n    fn process_request(&self, request: &mut LLMRequest) -> Result<(), MiddlewareError> {\n        \/\/ \u0130stek i\u00e7eri\u011findeki hassas verileri maskele\n        request.body = mask_sensitive_data(request.body);\n        Ok(())\n    }\n\n    fn process_response(&self, response: &mut LLMResponse) -> Result<(), MiddlewareError> {\n        \/\/ Yan\u0131t i\u00e7eri\u011finde de maskeleme gerekebilir\n        response.body = mask_sensitive_data(response.body);\n        Ok(())\n    }\n}\n<\/code><\/pre>\n<\/div>\n<p>Bu \u00f6rnek, bir veri maskeleme middleware&#8217;inin nas\u0131l g\u00f6r\u00fcnebilece\u011fine dair basit bir fikir vermektedir. Ger\u00e7ek uygulama, Respan&#8217;\u0131n API&#8217;sine ve entegrasyon noktalar\u0131na ba\u011fl\u0131 olacakt\u0131r.<\/p>\n<h3>Y\u00fcksek Kullan\u0131labilirlik (High Availability) ve \u00d6l\u00e7eklenebilirlik (Scalability) Stratejileri<\/h3>\n<p>\u00dcretim ortamlar\u0131nda, Respan Gateway&#8217;in y\u00fcksek kullan\u0131labilirli\u011fe sahip olmas\u0131 ve artan y\u00fcke g\u00f6re \u00f6l\u00e7eklenebilmesi kritik \u00f6neme sahiptir. Respan&#8217;\u0131 tek bir sunucuda \u00e7al\u0131\u015ft\u0131rmak yerine, birden fazla sunucuda k\u00fcme (cluster) halinde da\u011f\u0131tabilirsiniz. Bu, bir sunucunun ar\u0131zalanmas\u0131 durumunda bile hizmetin kesintisiz devam etmesini sa\u011flar.<\/p>\n<ul>\n<li><strong>Y\u00fck Dengeleyici (Load Balancer):<\/strong> Respan \u00f6rneklerinin \u00f6n\u00fcne bir y\u00fck dengeleyici (\u00f6rne\u011fin Nginx, HAProxy veya bulut sa\u011flay\u0131c\u0131lar\u0131n\u0131n y\u00fck dengeleyicileri) yerle\u015ftirerek gelen trafi\u011fi birden fazla Respan \u00f6rne\u011fine da\u011f\u0131tabilirsiniz.<\/li>\n<li><strong>Payla\u015f\u0131ml\u0131 \u00d6nbellek:<\/strong> E\u011fer \u00f6nbellekleme kullan\u0131yorsan\u0131z, \u00f6nbellek verilerinin t\u00fcm Respan \u00f6rnekleri aras\u0131nda payla\u015f\u0131lmas\u0131 gerekir. Redis veya Memcached gibi da\u011f\u0131t\u0131k \u00f6nbellek sistemleri bu ihtiyac\u0131 kar\u015f\u0131layabilir. Respan&#8217;\u0131n yap\u0131land\u0131rmas\u0131nda harici \u00f6nbellek sistemlerini kullanma se\u00e7ene\u011fi bulunmal\u0131d\u0131r.<\/li>\n<li><strong>Konteynerle\u015ftirme ve Orkestrasyon:<\/strong> Docker ve Kubernetes gibi ara\u00e7lar, Respan \u00f6rneklerini kolayca da\u011f\u0131tman\u0131za, y\u00f6netmenize ve otomatik olarak \u00f6l\u00e7eklendirmenize olanak tan\u0131r. Kubernetes, trafik artt\u0131\u011f\u0131nda otomatik olarak yeni Respan podlar\u0131 (\u00f6rnekleri) olu\u015fturabilir ve y\u00fck azald\u0131\u011f\u0131nda onlar\u0131 kapatabilir.<\/li>\n<\/ul>\n<h3>Performans Ayarlamalar\u0131 (Tuning) ve \u0130zleme<\/h3>\n<p>Rust ile yaz\u0131lm\u0131\u015f bir uygulama olarak Respan, zaten y\u00fcksek performansl\u0131d\u0131r. Ancak, belirli i\u015f y\u00fckleri i\u00e7in daha fazla optimizasyon yapmak m\u00fcmk\u00fcnd\u00fcr:<\/p>\n<ul>\n<li><strong>Kaynak Tahsisi:<\/strong> Respan&#8217;\u0131n \u00e7al\u0131\u015ft\u0131\u011f\u0131 sunucularda yeterli CPU ve bellek kaynaklar\u0131n\u0131n oldu\u011fundan emin olun. \u00d6zellikle yo\u011fun trafik alt\u0131nda, kaynak yetersizli\u011fi darbo\u011fazlara yol a\u00e7abilir.<\/li>\n<li><strong>Ba\u011flant\u0131 Havuzlar\u0131:<\/strong> LLM sa\u011flay\u0131c\u0131lar\u0131na yap\u0131lan ba\u011flant\u0131lar\u0131n yeniden kullan\u0131lmas\u0131 (connection pooling), yeni ba\u011flant\u0131 kurma maliyetini azaltarak performans\u0131 art\u0131rabilir. Respan, bu t\u00fcr optimizasyonlar\u0131 dahili olarak yapacakt\u0131r.<\/li>\n<li><strong>Metrik Toplama ve Uyar\u0131lar:<\/strong> Prometheus ve Grafana gibi ara\u00e7larla Respan&#8217;dan metrikler toplayarak (istek say\u0131s\u0131, yan\u0131t s\u00fcreleri, hata oranlar\u0131, \u00f6nbellek isabet oranlar\u0131 vb.) performans darbo\u011fazlar\u0131n\u0131 tespit edebilir ve proaktif olarak m\u00fcdahale edebilirsiniz. Belirli e\u015fik de\u011ferleri a\u015f\u0131ld\u0131\u011f\u0131nda uyar\u0131lar (alerts) ayarlamak, sorunlar\u0131 b\u00fcy\u00fcmeden yakalaman\u0131za yard\u0131mc\u0131 olur.<\/li>\n<\/ul>\n<h3>CI\/CD Entegrasyonu<\/h3>\n<p>Respan Gateway&#8217;i yaz\u0131l\u0131m geli\u015ftirme ya\u015fam d\u00f6ng\u00fcn\u00fcze (SDLC) entegre etmek, s\u00fcrekli entegrasyon (CI) ve s\u00fcrekli da\u011f\u0131t\u0131m (CD) s\u00fcre\u00e7leriyle kolayla\u015f\u0131r. Yeni \u00f6zellikler eklendi\u011finde veya hata d\u00fczeltmeleri yap\u0131ld\u0131\u011f\u0131nda, otomatik testler \u00e7al\u0131\u015ft\u0131r\u0131labilir ve yeni s\u00fcr\u00fcm g\u00fcvenli bir \u015fekilde \u00fcretim ortam\u0131na da\u011f\u0131t\u0131labilir. Bu, Respan altyap\u0131n\u0131z\u0131n g\u00fcncel, g\u00fcvenli ve hatas\u0131z kalmas\u0131n\u0131 sa\u011flar.<\/p>\n<ul>\n<li><strong>Otomatik Testler:<\/strong> Yeni bir kod de\u011fi\u015fikli\u011fi yap\u0131ld\u0131\u011f\u0131nda, bir CI sistemi (\u00f6rne\u011fin Jenkins, GitHub Actions, GitLab CI) otomatik olarak birim testlerini, entegrasyon testlerini ve performans testlerini \u00e7al\u0131\u015ft\u0131r\u0131r.<\/li>\n<li><strong>Konteyner Olu\u015fturma:<\/strong> Ba\u015far\u0131l\u0131 testlerden sonra, Respan&#8217;\u0131n yeni s\u00fcr\u00fcm\u00fc bir Docker g\u00f6r\u00fcnt\u00fcs\u00fc olarak olu\u015fturulur ve bir konteyner kay\u0131t defterine (\u00f6rne\u011fin Docker Hub, Google Container Registry) g\u00f6nderilir.<\/li>\n<li><strong>Otomatik Da\u011f\u0131t\u0131m:<\/strong> CD ara\u00e7lar\u0131 (\u00f6rne\u011fin Argo CD, Spinnaker) bu yeni Docker g\u00f6r\u00fcnt\u00fcs\u00fcn\u00fc alarak Kubernetes k\u00fcmenize veya di\u011fer da\u011f\u0131t\u0131m hedeflerinize otomatik olarak da\u011f\u0131tabilir. Bu, insan hatas\u0131n\u0131 azalt\u0131r ve da\u011f\u0131t\u0131m s\u00fcre\u00e7lerini h\u0131zland\u0131r\u0131r.<\/li>\n<\/ul>\n<p>Bu ileri d\u00fczey stratejilerle, Respan Gateway&#8217;i sadece bir arac\u0131 katman olmaktan \u00e7\u0131kar\u0131p, LLM altyap\u0131n\u0131z\u0131n omurgas\u0131 haline getirebilir, i\u015fletmenizin yapay zeka kapasitesini g\u00fcvenilir ve verimli bir \u015fekilde \u00f6l\u00e7eklendirebilirsiniz.<\/p>\n<h2>Ger\u00e7ek D\u00fcnya Senaryolar\u0131nda Respan Gateway Kullan\u0131m\u0131<\/h2>\n<p>Respan Gateway&#8217;in teorik faydalar\u0131n\u0131 ve teknik \u00f6zelliklerini ele ald\u0131k. \u015eimdi, ger\u00e7ek d\u00fcnyadan iki farkl\u0131 senaryoda Respan&#8217;\u0131n nas\u0131l bir fark yaratabilece\u011fine dair somut \u00f6rneklere bakal\u0131m. Bu vaka analizleri, Respan&#8217;\u0131n sadece bir &#8220;ge\u00e7it&#8221; olmaktan \u00f6te, stratejik bir ara\u00e7 olarak nas\u0131l konumland\u0131r\u0131labilece\u011fini g\u00f6sterecektir.<\/p>\n<h3>Vaka Analizi 1: Startup &#8220;Ak\u0131ll\u0131 Asistan\u0131m&#8221;\u0131n Maliyetleri D\u00fc\u015f\u00fcrme ve Performans\u0131 Art\u0131rma Hikayesi<\/h3>\n<p>Ak\u0131ll\u0131 Asistan\u0131m, k\u00fc\u00e7\u00fck ve orta \u00f6l\u00e7ekli i\u015fletmelere (KOB\u0130&#8217;ler) y\u00f6nelik yapay zeka destekli bir sanal asistan hizmeti sunan yeni bir startup&#8217;t\u0131r. M\u00fc\u015fterilerinin e-postalar\u0131n\u0131 yan\u0131tlamak, randevular\u0131 planlamak ve temel veri analizleri yapmak i\u00e7in LLM&#8217;leri yo\u011fun bir \u015fekilde kullanmaktad\u0131r. Ba\u015flang\u0131\u00e7ta, do\u011frudan OpenAI API&#8217;sine ba\u011flanarak i\u015flerini y\u00fcr\u00fct\u00fcyorlard\u0131. Ancak, h\u0131zla artan kullan\u0131c\u0131 taban\u0131yla birlikte iki b\u00fcy\u00fck sorunla kar\u015f\u0131la\u015ft\u0131lar:<\/p>\n<ol>\n<li><strong>Y\u00fcksek API Maliyetleri:<\/strong> Her istek i\u00e7in yap\u0131lan API \u00e7a\u011fr\u0131lar\u0131, \u00f6zellikle s\u0131k tekrarlanan veya benzer sorular i\u00e7in bile maliyetleri h\u0131zla art\u0131r\u0131yordu. B\u00fct\u00e7eleri zorlanmaya ba\u015flam\u0131\u015ft\u0131.<\/li>\n<li><strong>De\u011fi\u015fken Yan\u0131t S\u00fcreleri:<\/strong> OpenAI API&#8217;sinin yo\u011fun oldu\u011fu zamanlarda veya a\u011f gecikmelerinde, asistan\u0131n yan\u0131t s\u00fcreleri uzuyor, bu da m\u00fc\u015fteri deneyimini olumsuz etkiliyordu.<\/li>\n<\/ol>\n<p>Ak\u0131ll\u0131 Asistan\u0131m ekibi, bu sorunlar\u0131 \u00e7\u00f6zmek i\u00e7in Respan Gateway&#8217;i ara\u015ft\u0131rmaya ba\u015flad\u0131 ve h\u0131zl\u0131ca bir pilot proje ba\u015flatt\u0131. Respan&#8217;\u0131 kendi sunucular\u0131na kurdular ve a\u015fa\u011f\u0131daki yap\u0131land\u0131rmalar\u0131 uygulad\u0131lar:<\/p>\n<ul>\n<li><strong>\u00d6nbellekleme:<\/strong> S\u0131k\u00e7a sorulan sorular ve bunlar\u0131n LLM yan\u0131tlar\u0131 i\u00e7in Respan&#8217;\u0131n dahili \u00f6nbelle\u011fini etkinle\u015ftirdiler. \u00d6zellikle KOB\u0130&#8217;lerin destek taleplerinde belirli bir oranda tekrar eden sorular oldu\u011funu fark ettiler.<\/li>\n<li><strong>Oran S\u0131n\u0131rlama:<\/strong> OpenAI&#8217;nin API limitlerine uyum sa\u011flamak ve hizmet kesintilerini \u00f6nlemek i\u00e7in oran s\u0131n\u0131rlama kurallar\u0131 belirlediler.<\/li>\n<li><strong>Basit Y\u00f6nlendirme:<\/strong> Ba\u015flang\u0131\u00e7ta sadece OpenAI kullansalar da, gelecekte farkl\u0131 LLM sa\u011flay\u0131c\u0131lar\u0131n\u0131 da de\u011ferlendirme potansiyeli i\u00e7in Respan&#8217;\u0131n y\u00f6nlendirme yeteneklerini aktif hale getirdiler.<\/li>\n<\/ul>\n<p><strong>Sonu\u00e7lar:<\/strong> Respan Gateway&#8217;i devreye ald\u0131ktan sonra Ak\u0131ll\u0131 Asistan\u0131m, \u00e7arp\u0131c\u0131 sonu\u00e7lar elde etti:<\/p>\n<ul>\n<li><strong>Maliyet Tasarrufu:<\/strong> \u00d6nbellekleme sayesinde, LLM API \u00e7a\u011fr\u0131lar\u0131n\u0131n %40&#8217;\u0131n\u0131 \u00f6nbellekten kar\u015f\u0131lad\u0131lar. Bu, ayl\u0131k LLM faturalar\u0131nda %35&#8217;lik bir d\u00fc\u015f\u00fc\u015f anlam\u0131na geliyordu.<\/li>\n<li><strong>Performans Art\u0131\u015f\u0131:<\/strong> \u00d6nbellekten d\u00f6nen yan\u0131tlar anl\u0131k oldu\u011fu i\u00e7in, ortalama yan\u0131t s\u00fcresi 1.5 saniyeden 300 milisaniyeye d\u00fc\u015ft\u00fc. Bu, m\u00fc\u015fteri memnuniyetini \u00f6nemli \u00f6l\u00e7\u00fcde art\u0131rd\u0131.<\/li>\n<li><strong>Operasyonel Kolayl\u0131k:<\/strong> Respan, t\u00fcm LLM trafi\u011fini tek bir noktadan izlemelerini ve y\u00f6netmelerini sa\u011flad\u0131\u011f\u0131 i\u00e7in, sorun giderme ve optimizasyon s\u00fcre\u00e7leri \u00e7ok daha kolayla\u015ft\u0131.<\/li>\n<\/ul>\n<p>Bu ba\u015far\u0131 hikayesi, Respan&#8217;\u0131n k\u00fc\u00e7\u00fck ve orta \u00f6l\u00e7ekli i\u015fletmeler i\u00e7in bile ne kadar de\u011ferli bir ara\u00e7 olabilece\u011fini g\u00f6steriyor.<\/p>\n<h3>Vaka Analizi 2: \u00c7ok Modelli Bir Yapay Zeka Uygulamas\u0131nda Respan&#8217;\u0131n Rol\u00fc<\/h3>\n<p>B\u00fcy\u00fck bir kurumsal yaz\u0131l\u0131m \u015firketi olan &#8220;Global \u00c7\u00f6z\u00fcmler&#8221;, m\u00fc\u015fterilerine farkl\u0131 yapay zeka yetenekleri sunan karma\u015f\u0131k bir platform geli\u015ftiriyordu. Bu platform, metin \u00f6zetleme i\u00e7in bir LLM (\u00f6rne\u011fin, GPT-4), kod \u00fcretimi i\u00e7in ba\u015fka bir LLM (\u00f6rne\u011fin, Google Gemini Code), ve \u00e7ok dilli \u00e7eviri i\u00e7in \u00fc\u00e7\u00fcnc\u00fc bir LLM (\u00f6rne\u011fin, Anthropic Claude) kullan\u0131yordu. Uygulama, her bir LLM sa\u011flay\u0131c\u0131s\u0131n\u0131n API&#8217;sine do\u011frudan ba\u011flanmak yerine, Respan Gateway&#8217;i merkezi bir LLM y\u00f6netim katman\u0131 olarak konumland\u0131rd\u0131.<\/p>\n<p><strong>Respan&#8217;\u0131n Rol\u00fc:<\/strong><\/p>\n<ul>\n<li><strong>Dinamik Y\u00f6nlendirme:<\/strong> Global \u00c7\u00f6z\u00fcmler, Respan&#8217;\u0131, gelen iste\u011fin t\u00fcr\u00fcne (\u00f6zetleme, kod \u00fcretimi, \u00e7eviri) g\u00f6re en uygun LLM sa\u011flay\u0131c\u0131s\u0131na otomatik olarak y\u00f6nlendirecek \u015fekilde yap\u0131land\u0131rd\u0131. Bu, uygulaman\u0131n kodunu sadele\u015ftirdi ve yeni bir LLM ekleme veya mevcut birini de\u011fi\u015ftirme esnekli\u011fi sa\u011flad\u0131.<\/li>\n<li><strong>Model Haritalama:<\/strong> Uygulama, LLM&#8217;leri kendi dahili adland\u0131rma \u015femas\u0131yla \u00e7a\u011f\u0131r\u0131rken, Respan bu adlar\u0131 ilgili sa\u011flay\u0131c\u0131n\u0131n bekledi\u011fi ger\u00e7ek model adlar\u0131na d\u00f6n\u00fc\u015ft\u00fcrd\u00fc. \u00d6rne\u011fin, <code>summarize-model<\/code> \u00e7a\u011fr\u0131s\u0131 GPT-4&#8217;e, <code>code-gen-model<\/code> \u00e7a\u011fr\u0131s\u0131 Gemini Code&#8217;a y\u00f6nlendirildi.<\/li>\n<li><strong>Geli\u015fmi\u015f \u00d6nbellekleme:<\/strong> \u00d6zellikle s\u0131k kullan\u0131lan metin \u00f6zetleri ve kod par\u00e7ac\u0131klar\u0131 i\u00e7in da\u011f\u0131t\u0131k bir Redis \u00f6nbelle\u011fi ile Respan&#8217;\u0131 entegre ettiler. Bu, b\u00fcy\u00fck \u00f6l\u00e7ekte maliyet tasarrufu ve performans art\u0131\u015f\u0131 sa\u011flad\u0131.<\/li>\n<li><strong>Merkezi G\u00fcvenlik:<\/strong> T\u00fcm LLM API anahtarlar\u0131 Respan&#8217;da g\u00fcvenli bir \u015fekilde sakland\u0131 ve platformun mikroservisleri, Respan&#8217;a kendi i\u00e7 kimlik do\u011frulama mekanizmalar\u0131yla eri\u015fti. Bu, g\u00fcvenlik a\u00e7\u0131\u011f\u0131 riskini minimize etti.<\/li>\n<li><strong>Kapsaml\u0131 G\u00f6zlemlenebilirlik:<\/strong> Respan \u00fczerinden ge\u00e7en t\u00fcm LLM \u00e7a\u011fr\u0131lar\u0131, Prometheus ve Grafana ile izlendi. Bu, hangi LLM&#8217;in ne kadar kullan\u0131ld\u0131\u011f\u0131n\u0131, performans metriklerini ve maliyet da\u011f\u0131l\u0131m\u0131n\u0131 ger\u00e7ek zamanl\u0131 olarak g\u00f6rmelerini sa\u011flad\u0131. B\u00f6ylece, kaynak tahsisini ve b\u00fct\u00e7e planlamas\u0131n\u0131 daha bilin\u00e7li yapabildiler.<\/li>\n<\/ul>\n<p><strong>Sonu\u00e7lar:<\/strong> Global \u00c7\u00f6z\u00fcmler, Respan Gateway sayesinde, \u00e7ok modelli yapay zeka platformunu \u00e7ok daha y\u00f6netilebilir, g\u00fcvenli ve maliyet etkin bir \u015fekilde \u00f6l\u00e7eklendirebildi. Geli\u015ftirme ekipleri, farkl\u0131 LLM API&#8217;lerinin karma\u015f\u0131kl\u0131\u011f\u0131yla u\u011fra\u015fmak yerine, do\u011frudan i\u015f de\u011feri yaratan \u00f6zelliklere odaklanabildiler. Respan, bu karma\u015f\u0131k altyap\u0131n\u0131n &#8220;tek kap\u0131s\u0131&#8221; haline gelerek, t\u00fcm LLM etkile\u015fimini basitle\u015ftirdi.<\/p>\n<p>Bu iki vaka analizi, Respan Gateway&#8217;in hem k\u00fc\u00e7\u00fck bir startup&#8217;\u0131n operasyonel verimlili\u011fini art\u0131rmada hem de b\u00fcy\u00fck bir \u015firketin karma\u015f\u0131k \u00e7ok modelli yapay zeka stratejisini y\u00f6netmede nas\u0131l kritik bir rol oynayabilece\u011fini a\u00e7\u0131k\u00e7a g\u00f6stermektedir. Respan, LLM&#8217;leri kullanan her t\u00fcrl\u00fc uygulama i\u00e7in esneklik, maliyet kontrol\u00fc ve g\u00fcvenlik sunan g\u00fc\u00e7l\u00fc bir ara\u00e7t\u0131r.<\/p>\n<h2>Respan Gateway ile LLM Altyap\u0131n\u0131z\u0131 Gelece\u011fe Ta\u015f\u0131y\u0131n<\/h2>\n<p>B\u00fcy\u00fck Dil Modelleri, i\u015f d\u00fcnyas\u0131nda ve teknolojide devrim yaratmaya devam ederken, bu g\u00fc\u00e7l\u00fc ara\u00e7lar\u0131 y\u00f6netme ihtiyac\u0131 da ayn\u0131 oranda art\u0131yor. Respan Gateway, Rust&#8217;\u0131n getirdi\u011fi performans, g\u00fcvenlik ve g\u00fcvenilirlik avantajlar\u0131yla birlikte, a\u00e7\u0131k kaynak bir \u00e7\u00f6z\u00fcm olarak bu karma\u015f\u0131k y\u00f6netim s\u00fcrecini basitle\u015ftirmeyi hedefliyor. Bu makalede, Respan&#8217;\u0131n temel kurulumundan, ak\u0131ll\u0131 y\u00f6nlendirme, \u00f6nbellekleme, oran s\u0131n\u0131rlama gibi geli\u015fmi\u015f \u00f6zelliklerine ve ger\u00e7ek d\u00fcnya kullan\u0131m senaryolar\u0131na kadar bir\u00e7ok y\u00f6n\u00fcn\u00fc ele ald\u0131k.<\/p>\n<p>Respan Gateway, LLM API&#8217;leri ile uygulaman\u0131z aras\u0131nda tek bir arac\u0131 katman olu\u015fturarak, geli\u015ftirme s\u00fcrecini h\u0131zland\u0131r\u0131r, operasyonel maliyetleri d\u00fc\u015f\u00fcr\u00fcr ve LLM altyap\u0131n\u0131z\u0131n genel performans\u0131n\u0131 ve g\u00fcvenli\u011fini art\u0131r\u0131r. \u0130ster maliyetleri optimize etmek isteyen bir startup olun, ister birden fazla LLM sa\u011flay\u0131c\u0131s\u0131n\u0131 y\u00f6neten b\u00fcy\u00fck bir kurulu\u015f, Respan size esneklik ve kontrol sunar. Rust&#8217;\u0131n h\u0131z ve bellek g\u00fcvenli\u011fi gibi do\u011fal avantajlar\u0131, Respan&#8217;\u0131n yo\u011fun y\u00fck alt\u0131nda bile istikrarl\u0131 ve h\u0131zl\u0131 \u00e7al\u0131\u015fmas\u0131n\u0131 garanti eder. A\u00e7\u0131k kaynak olmas\u0131 ise toplulu\u011fun katk\u0131lar\u0131yla s\u00fcrekli geli\u015fimini ve \u015feffafl\u0131\u011f\u0131n\u0131 sa\u011flar.<\/p>\n<p>LLM&#8217;lerin kullan\u0131m\u0131 daha da yayg\u0131nla\u015ft\u0131k\u00e7a, bu modelleri verimli bir \u015fekilde y\u00f6netebilen \u00e7\u00f6z\u00fcmlere olan ihtiya\u00e7 artacakt\u0131r. Respan Gateway, bu ihtiyac\u0131 kar\u015f\u0131lamak i\u00e7in g\u00fc\u00e7l\u00fc bir aday olarak \u00f6ne \u00e7\u0131k\u0131yor ve LLM altyap\u0131n\u0131z\u0131 gelece\u011fe haz\u0131r hale getirmenize yard\u0131mc\u0131 olabilir. Kendi LLM uygulamalar\u0131n\u0131zda kar\u015f\u0131la\u015ft\u0131\u011f\u0131n\u0131z zorluklar\u0131 d\u00fc\u015f\u00fcn\u00fcn ve Respan&#8217;\u0131n bu sorunlara nas\u0131l \u00e7\u00f6z\u00fcm getirebilece\u011fini de\u011ferlendirin. Bu a\u00e7\u0131k kaynakl\u0131 ara\u00e7, yapay zeka yolculu\u011funuzda size \u00f6nemli bir avantaj sa\u011flayabilir.<\/p>\n<h2>S\u0131k\u00e7a Sorulan Sorular<\/h2>\n<h3>Respan Gateway nedir ve ne i\u015fe yarar?<\/h3>\n<p>Respan Gateway, Rust ile yaz\u0131lm\u0131\u015f a\u00e7\u0131k kaynakl\u0131 bir B\u00fcy\u00fck Dil Modeli (LLM) ge\u00e7ididir. Uygulaman\u0131z ile farkl\u0131 LLM sa\u011flay\u0131c\u0131lar\u0131n\u0131n API&#8217;leri aras\u0131nda bir arac\u0131 katman g\u00f6revi g\u00f6r\u00fcr. Temel i\u015flevleri aras\u0131nda istek y\u00f6nlendirme, \u00f6nbellekleme, oran s\u0131n\u0131rlama, kimlik do\u011frulama ve g\u00f6zlemlenebilirlik bulunur. Bu sayede LLM y\u00f6netimini basitle\u015ftirir, maliyetleri d\u00fc\u015f\u00fcr\u00fcr ve performans\u0131 art\u0131r\u0131r.<\/p>\n<h3>Neden Respan Gateway kullanmal\u0131y\u0131m?<\/h3>\n<p>Respan Gateway kullanman\u0131z i\u00e7in bir\u00e7ok neden vard\u0131r: maliyet optimizasyonu (\u00f6nbellekleme ile), performans art\u0131\u015f\u0131 (d\u00fc\u015f\u00fck gecikme ve \u00f6nbellek), g\u00fcvenlik (API anahtarlar\u0131n\u0131n merkezi y\u00f6netimi), esneklik (farkl\u0131 LLM sa\u011flay\u0131c\u0131lar\u0131 aras\u0131nda kolay ge\u00e7i\u015f), oran s\u0131n\u0131rlama (API limitlerini a\u015fmay\u0131 \u00f6nleme) ve g\u00f6zlemlenebilirlik (LLM kullan\u0131m\u0131n\u0131 izleme ve analiz etme).<\/p>\n<h3>Respan Gateway&#8217;i kurmak zor mu?<\/h3>\n<p>Rust geli\u015ftirme ortam\u0131na a\u015fina olanlar i\u00e7in kurulum olduk\u00e7a basittir. Rustup ile Rust&#8217;\u0131 kurduktan sonra, Respan&#8217;\u0131n kaynak kodunu klonlay\u0131p Cargo ile derlemeniz yeterlidir. Ard\u0131ndan, <code>config.toml<\/code> dosyas\u0131 ile temel yap\u0131land\u0131rmalar\u0131 yaparak hemen kullanmaya ba\u015flayabilirsiniz.<\/p>\n<h3>Respan hangi LLM sa\u011flay\u0131c\u0131lar\u0131n\u0131 destekliyor?<\/h3>\n<p>Respan, yap\u0131land\u0131rma yoluyla farkl\u0131 LLM sa\u011flay\u0131c\u0131lar\u0131n\u0131 destekleyecek \u015fekilde tasarlanm\u0131\u015ft\u0131r. Genellikle OpenAI, Google Gemini, Anthropic Claude gibi pop\u00fcler LLM API&#8217;leri ile entegrasyon i\u00e7in yap\u0131land\u0131rma se\u00e7enekleri sunar. Esnek y\u00f6nlendirme ve model haritalama yetenekleri sayesinde yeni sa\u011flay\u0131c\u0131lar\u0131 kolayca ekleyebilirsiniz.<\/p>\n<h3>Respan Gateway a\u00e7\u0131k kaynak m\u0131 ve nas\u0131l katk\u0131da bulunabilirim?<\/h3>\n<p>Evet, Respan Gateway tamamen a\u00e7\u0131k kaynakl\u0131 bir projedir. Genellikle GitHub \u00fczerinde bar\u0131nd\u0131r\u0131l\u0131r. Projeye katk\u0131da bulunmak i\u00e7in kod geli\u015ftirebilir, hata raporlar\u0131 sunabilir, dok\u00fcmantasyonu iyile\u015ftirebilir veya yeni \u00f6zellik \u00f6nerileri getirebilirsiniz. Projenin GitHub deposunu ziyaret ederek katk\u0131da bulunma y\u00f6nergelerini inceleyebilirsiniz.<\/p>\n<p>#RespanGateway #LLMGateway #Rust #YapayZeka #A\u00e7\u0131kKaynak #B\u00fcy\u00fckDilModelleri #APIY\u00f6netimi #Teknoloji #WebGeli\u015ftirme<\/p>\n<div class=\"github-example-link\"><strong>\u00d6rnek kod:<\/strong> <a href=\"https:\/\/github.com\/fatihsoysalcom\/respan-gateway-llm-client-mock\" target=\"_blank\" rel=\"noopener noreferrer\">github.com\/fatihsoysalcom\/respan-gateway-llm-client-mock<\/a><\/div>\n","protected":false},"excerpt":{"rendered":"B\u00fcy\u00fck Dil Modellerinin (LLM) y\u00fckseli\u015fi, yapay zeka uygulamalar\u0131n\u0131n geli\u015fimine inan\u0131lmaz bir ivme kazand\u0131rd\u0131.","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"csco_page_header_type":"","csco_page_load_nextpost":"","csco_page_subscribe_form":"","csco_page_contact_form":"","footnotes":""},"categories":[1],"tags":[],"class_list":{"0":"post-45742","1":"post","2":"type-post","3":"status-publish","4":"format-standard","6":"category-genel","7":"cs-entry","8":"cs-video-wrap"},"yoast_head":"<!-- This site is optimized with the Yoast SEO Premium plugin v20.5 (Yoast SEO v25.3.1) - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm - Kodlar\u0131n Gizemli D\u00fcnyas\u0131<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/\" \/>\n<meta property=\"og:locale\" content=\"tr_TR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm\" \/>\n<meta property=\"og:description\" content=\"B\u00fcy\u00fck Dil Modellerinin (LLM) y\u00fckseli\u015fi, yapay zeka uygulamalar\u0131n\u0131n geli\u015fimine inan\u0131lmaz bir ivme kazand\u0131rd\u0131.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/\" \/>\n<meta property=\"og:site_name\" content=\"Kodlar\u0131n Gizemli D\u00fcnyas\u0131\" \/>\n<meta property=\"article:published_time\" content=\"2026-11-17T18:04:04+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-11-17T18:04:37+00:00\" \/>\n<meta name=\"author\" content=\"Fatih Soysal\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Yazan:\" \/>\n\t<meta name=\"twitter:data1\" content=\"Fatih Soysal\" \/>\n\t<meta name=\"twitter:label2\" content=\"Tahmini okuma s\u00fcresi\" \/>\n\t<meta name=\"twitter:data2\" content=\"29 dakika\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/\"},\"author\":{\"name\":\"Fatih Soysal\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1\"},\"headline\":\"Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm\",\"datePublished\":\"2026-11-17T18:04:04+00:00\",\"dateModified\":\"2026-11-17T18:04:37+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/\"},\"wordCount\":5392,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1\"},\"inLanguage\":\"tr\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/#respond\"]}],\"copyrightYear\":\"2026\",\"copyrightHolder\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#organization\"}},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/\",\"url\":\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/\",\"name\":\"Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm - Kodlar\u0131n Gizemli D\u00fcnyas\u0131\",\"isPartOf\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#website\"},\"datePublished\":\"2026-11-17T18:04:04+00:00\",\"dateModified\":\"2026-11-17T18:04:37+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/#breadcrumb\"},\"inLanguage\":\"tr\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Anasayfa\",\"item\":\"https:\/\/fatihsoysal.com\/blog\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#website\",\"url\":\"https:\/\/fatihsoysal.com\/blog\/\",\"name\":\"Fatihsoysal.com\",\"description\":\"Blog - Yaz\u0131l\u0131m D\u00fcnyas\u0131 Tecr\u00fcbelerim\",\"publisher\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/fatihsoysal.com\/blog\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"tr\"},{\"@type\":[\"Person\",\"Organization\"],\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1\",\"name\":\"Fatih Soysal\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"tr\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/fatihsoysal.com\/blog\/wp-content\/uploads\/2024\/04\/cropped-replicate-prediction-3kgg1hgjn5rgp0cf0p5tr0jw7w-1.png\",\"contentUrl\":\"https:\/\/fatihsoysal.com\/blog\/wp-content\/uploads\/2024\/04\/cropped-replicate-prediction-3kgg1hgjn5rgp0cf0p5tr0jw7w-1.png\",\"width\":512,\"height\":512,\"caption\":\"Fatih Soysal\"},\"logo\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/image\/\"},\"description\":\"Kullan\u0131m ve kodlama m\u00fckemmeliyetini odak alan uygulamalar olu\u015fturma deneyimine sahip, profesyonel olarak 15+ y\u0131l \u00fczeri deneyime sahip bir yaz\u0131l\u0131m m\u00fchendisi.\",\"url\":\"https:\/\/fatihsoysal.com\/blog\/author\/fatihsoysal\/\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm - Kodlar\u0131n Gizemli D\u00fcnyas\u0131","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/","og_locale":"tr_TR","og_type":"article","og_title":"Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm","og_description":"B\u00fcy\u00fck Dil Modellerinin (LLM) y\u00fckseli\u015fi, yapay zeka uygulamalar\u0131n\u0131n geli\u015fimine inan\u0131lmaz bir ivme kazand\u0131rd\u0131.","og_url":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/","og_site_name":"Kodlar\u0131n Gizemli D\u00fcnyas\u0131","article_published_time":"2026-11-17T18:04:04+00:00","article_modified_time":"2026-11-17T18:04:37+00:00","author":"Fatih Soysal","twitter_card":"summary_large_image","twitter_misc":{"Yazan:":"Fatih Soysal","Tahmini okuma s\u00fcresi":"29 dakika"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/#article","isPartOf":{"@id":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/"},"author":{"name":"Fatih Soysal","@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1"},"headline":"Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm","datePublished":"2026-11-17T18:04:04+00:00","dateModified":"2026-11-17T18:04:37+00:00","mainEntityOfPage":{"@id":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/"},"wordCount":5392,"commentCount":0,"publisher":{"@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1"},"inLanguage":"tr","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/#respond"]}],"copyrightYear":"2026","copyrightHolder":{"@id":"https:\/\/fatihsoysal.com\/blog\/#organization"}},{"@type":"WebPage","@id":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/","url":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/","name":"Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm - Kodlar\u0131n Gizemli D\u00fcnyas\u0131","isPartOf":{"@id":"https:\/\/fatihsoysal.com\/blog\/#website"},"datePublished":"2026-11-17T18:04:04+00:00","dateModified":"2026-11-17T18:04:37+00:00","breadcrumb":{"@id":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/#breadcrumb"},"inLanguage":"tr","potentialAction":[{"@type":"ReadAction","target":["https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/fatihsoysal.com\/blog\/respan-gateway-ile-llm-yonetiminde-yeni-bir-donem-rust-gucuyle-acik-kaynak-cozum\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Anasayfa","item":"https:\/\/fatihsoysal.com\/blog\/"},{"@type":"ListItem","position":2,"name":"Respan Gateway ile LLM Y\u00f6netiminde Yeni Bir D\u00f6nem: Rust G\u00fcc\u00fcyle A\u00e7\u0131k Kaynak \u00c7\u00f6z\u00fcm"}]},{"@type":"WebSite","@id":"https:\/\/fatihsoysal.com\/blog\/#website","url":"https:\/\/fatihsoysal.com\/blog\/","name":"Fatihsoysal.com","description":"Blog - Yaz\u0131l\u0131m D\u00fcnyas\u0131 Tecr\u00fcbelerim","publisher":{"@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/fatihsoysal.com\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"tr"},{"@type":["Person","Organization"],"@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1","name":"Fatih Soysal","image":{"@type":"ImageObject","inLanguage":"tr","@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/image\/","url":"https:\/\/fatihsoysal.com\/blog\/wp-content\/uploads\/2024\/04\/cropped-replicate-prediction-3kgg1hgjn5rgp0cf0p5tr0jw7w-1.png","contentUrl":"https:\/\/fatihsoysal.com\/blog\/wp-content\/uploads\/2024\/04\/cropped-replicate-prediction-3kgg1hgjn5rgp0cf0p5tr0jw7w-1.png","width":512,"height":512,"caption":"Fatih Soysal"},"logo":{"@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/image\/"},"description":"Kullan\u0131m ve kodlama m\u00fckemmeliyetini odak alan uygulamalar olu\u015fturma deneyimine sahip, profesyonel olarak 15+ y\u0131l \u00fczeri deneyime sahip bir yaz\u0131l\u0131m m\u00fchendisi.","url":"https:\/\/fatihsoysal.com\/blog\/author\/fatihsoysal\/"}]}},"yoast_meta":{"yoast_wpseo_title":"","yoast_wpseo_metadesc":"","yoast_wpseo_canonical":""},"amp_enabled":true,"_links":{"self":[{"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/posts\/45742","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/comments?post=45742"}],"version-history":[{"count":1,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/posts\/45742\/revisions"}],"predecessor-version":[{"id":45743,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/posts\/45742\/revisions\/45743"}],"wp:attachment":[{"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/media?parent=45742"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/categories?post=45742"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/tags?post=45742"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}