{"id":43070,"date":"2026-07-04T21:04:40","date_gmt":"2026-07-04T18:04:40","guid":{"rendered":"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/"},"modified":"2026-07-04T21:05:08","modified_gmt":"2026-07-04T18:05:08","slug":"llm-function-calling-nasil-calisir-adim-adim-rehber","status":"publish","type":"post","link":"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/","title":{"rendered":"LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber"},"content":{"rendered":"

LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber<\/h2>\n

B\u00fcy\u00fck dil modellerinin d\u0131\u015f d\u00fcnya ile konu\u015fmas\u0131n\u0131 sa\u011flayan LLM Function Calling mekanizmas\u0131n\u0131, token seviyesinden ara\u00e7 koordinasyonuna kadar t\u00fcm detaylar\u0131yla ke\u015ffedin.<\/p>\n

LLM Function Calling Nedir ve Neden \u0130htiyac\u0131m\u0131z Var?<\/h2>\n

B\u00fcy\u00fck dil modelleri (Large Language Models – LLM), insan benzeri metinler \u00fcretme konusunda ola\u011fan\u00fcst\u00fc yeteneklere sahiptir. Ancak bu modellerin en b\u00fcy\u00fck s\u0131n\u0131rlamalar\u0131ndan biri, e\u011fitim verilerinin kesildi\u011fi andan sonraki g\u00fcncel bilgilere eri\u015fememeleridir. Bununla birlikte, matematiksel i\u015flemleri do\u011frudan yapmakta zorlan\u0131rlar ve harici sistemler \u00fczerinde aksiyon alamazlar. \u00d6rne\u011fin, bir modele “Bug\u00fcn \u0130stanbul’da hava nas\u0131l?” diye sordu\u011funuzda, model bu bilgiye do\u011frudan kendi parametrik haf\u0131zas\u0131ndan eri\u015femez. Bu noktada devreye API entegrasyonu<\/strong> ve LLM Function Calling (Fonksiyon \u00c7a\u011f\u0131rma) mekanizmas\u0131 girer.<\/p>\n

Function calling, b\u00fcy\u00fck dil modellerinin kendi ba\u015flar\u0131na kod \u00e7al\u0131\u015ft\u0131rmas\u0131 anlam\u0131na gelmez. Asl\u0131nda bu s\u00fcre\u00e7, modelin kendisine sunulan ara\u00e7lar (tools) aras\u0131ndan hangisini, hangi parametrelerle kullanmas\u0131 gerekti\u011fini yap\u0131land\u0131r\u0131lm\u0131\u015f bir veri format\u0131nda (genellikle JSON) tespit etme yetene\u011fidir. B\u00f6ylelikle model, bir metin \u00fcretmek yerine, uygulaman\u0131z\u0131n anlayabilece\u011fi ve \u00e7al\u0131\u015ft\u0131rabilece\u011fi bir komut dosyas\u0131 haz\u0131rlar. Bu durum, statik yapay zekay\u0131 dinamik bir ajana d\u00f6n\u00fc\u015ft\u00fcren en kritik ad\u0131md\u0131r.<\/p>\n

\u00d6ncelikle, geleneksel y\u00f6ntemlerde modellerin \u00e7\u0131kt\u0131lar\u0131n\u0131 d\u00fczenli hale getirmek i\u00e7in yo\u011fun bir istem m\u00fchendisli\u011fi (prompt engineering) uygulamak gerekiyordu. “L\u00fctfen sadece JSON format\u0131nda yan\u0131t ver” gibi talimatlar \u00e7o\u011fu zaman modelin yarat\u0131c\u0131 do\u011fas\u0131 nedeniyle ba\u015far\u0131s\u0131z oluyordu. Ancak modern LLM modelleri, fonksiyon \u00e7a\u011fr\u0131lar\u0131n\u0131 do\u011frudan kendi mimarilerinde destekleyecek \u015fekilde \u00f6zel olarak e\u011fitilmi\u015ftir. Sonu\u00e7 olarak, geli\u015ftiriciler art\u0131k \u00e7ok daha kararl\u0131, g\u00fcvenilir ve entegre edilebilir yapay zeka uygulamalar\u0131 geli\u015ftirebilmektedir.<\/p>\n

Token Seviyesinde Ne Oluyor? Arka Plandaki Mekanizma<\/h2>\n

Peki, bir yapay zeka modeli d\u00fcz metin yazmak yerine nas\u0131l oluyor da bir fonksiyon \u00e7a\u011fr\u0131s\u0131 yapmaya karar veriyor? Bu sorunun cevab\u0131, tokenizasyon<\/strong> ve modelin olas\u0131l\u0131k da\u011f\u0131l\u0131mlar\u0131nda sakl\u0131d\u0131r. Bir LLM, her ad\u0131mda bir sonraki token’\u0131 (kelime veya karakter gruplar\u0131n\u0131) tahmin ederek \u00e7al\u0131\u015f\u0131r. Biz modele kullanabilece\u011fi fonksiyonlar\u0131n listesini verdi\u011fimizde, bu tan\u0131mlamalar sistem isteminin (system prompt) \u00f6zel bir par\u00e7as\u0131 haline gelir.<\/p>\n

\u0130lk olarak, model girdi metnini ve fonksiyon \u015femalar\u0131n\u0131 analiz eder. E\u011fer kullan\u0131c\u0131n\u0131n iste\u011fi, tan\u0131mlanan fonksiyonlardan birinin a\u00e7\u0131klamas\u0131yla e\u015fle\u015fiyorsa, model bir sonraki token olarak normal bir kelime se\u00e7mek yerine, \u00f6zel bir durdurma token’\u0131 (stop token) veya fonksiyon \u00e7a\u011fr\u0131s\u0131 ba\u015flatma token’\u0131 \u00fcretir. \u00d6rne\u011fin, OpenAI modellerinde bu durum arka planda \u00f6zel bir API rol\u00fc olan tool_calls<\/code> ile y\u00f6netilir. Model, \u00e7\u0131kt\u0131 \u00fcretirken olas\u0131l\u0131k matrisini bu \u00f6zel formata g\u00f6re daralt\u0131r.<\/p>\n

Daha sonra, model fonksiyonun ad\u0131n\u0131 ve gerekli parametreleri \u00fcretmeye ba\u015flar. Bu a\u015famada “constrained decoding” (k\u0131s\u0131tl\u0131 kod \u00e7\u00f6zme) ad\u0131 verilen bir teknik kullan\u0131l\u0131r. Model, rastgele kelimeler \u00fcretmek yerine, yaln\u0131zca tan\u0131mlad\u0131\u011f\u0131n\u0131z JSON \u015femas\u0131na (JSON schema) uygun token’lar\u0131 se\u00e7meye zorlan\u0131r. \u00d6rne\u011fin, \u015femada bir parametrenin tipi “integer” (tam say\u0131) olarak belirtilmi\u015fse, model o parametre i\u00e7in sadece say\u0131sal token’lar \u00fcretme e\u011filiminde olur. B\u00f6ylelikle, kodunuzun do\u011frudan parse edebilece\u011fi (ayr\u0131\u015ft\u0131rabilece\u011fi) hatas\u0131z bir JSON \u00e7\u0131kt\u0131s\u0131 elde edilir.<\/p>\n

JSON \u015eemas\u0131 (Schema) Nas\u0131l Haz\u0131rlan\u0131r?<\/h2>\n

Modelin hangi fonksiyonu ne zaman \u00e7a\u011f\u0131raca\u011f\u0131n\u0131 anlamas\u0131 i\u00e7in ona net bir k\u0131lavuz sunman\u0131z gerekir. Bu k\u0131lavuz, JSON \u015femas\u0131 standartlar\u0131na uygun olarak haz\u0131rlan\u0131r. \u015eema i\u00e7erisinde fonksiyonun ad\u0131, ne i\u015fe yarad\u0131\u011f\u0131na dair detayl\u0131 bir a\u00e7\u0131klama ve kabul etti\u011fi parametrelerin veri tipleri yer al\u0131r. Model, bu a\u00e7\u0131klamalar\u0131 okuyarak hangi arac\u0131 se\u00e7ece\u011fine karar verdi\u011fi i\u00e7in, a\u00e7\u0131klama k\u0131s\u0131mlar\u0131n\u0131n net ve anla\u015f\u0131l\u0131r yaz\u0131lmas\u0131 hayati \u00f6nem ta\u015f\u0131r.<\/p>\n

Bununla birlikte, T\u00fcrkiye’deki yerel bir e-ticaret platformunun kargo durumunu sorgulayan bir senaryoyu ele alal\u0131m. Modelin kullan\u0131c\u0131n\u0131n kargo durumunu sorgulayabilmesi i\u00e7in bir API’ye istek atmas\u0131 gerekmektedir. Bu i\u015flem i\u00e7in haz\u0131rlayaca\u011f\u0131m\u0131z \u00f6rnek bir JSON \u015femas\u0131 \u015fu \u015fekilde olmal\u0131d\u0131r:<\/p>\n

\n
{\n  \"name\": \"sorgula_kargo_durumu\",\n  \"description\": \"Kullan\u0131c\u0131n\u0131n sipari\u015f takip numaras\u0131 ile kargosunun g\u00fcncel durumunu ve konumunu sorgular.\",\n  \"parameters\": {\n    \"type\": \"object\",\n    \"properties\": {\n      \"kargo_firmasi\": {\n        \"type\": \"string\",\n        \"enum\": [\"Yurti\u00e7i Kargo\", \"Aras Kargo\", \"MNG Kargo\"],\n        \"description\": \"Kargonun g\u00f6nderildi\u011fi kargo \u015firketinin ad\u0131.\"\n      },\n      \"takip_no\": {\n        \"type\": \"string\",\n        \"description\": \"12 haneli benzersiz kargo takip numaras\u0131.\"\n      }\n    },\n    \"required\": [\"kargo_firmasi\", \"takip_no\"]\n  }\n}<\/code><\/pre>\n<\/div>\n
\u015e\u00fcphesiz, buradaki en kritik alan description<\/code> (a\u00e7\u0131klama) alan\u0131d\u0131r. Model, kullan\u0131c\u0131n\u0131n “Aras Kargo ile gelen 45678 takip nolu paketim nerede?” sorusunu g\u00f6rd\u00fc\u011f\u00fcnde, bu a\u00e7\u0131klamalar\u0131 okur. Ard\u0131ndan, parametreleri e\u015fle\u015ftirerek kargo_firmasi: \"Aras Kargo\"<\/code> ve takip_no: \"45678\"<\/code> de\u011ferlerini \u00e7\u0131kar\u0131r. Bu do\u011frultuda, do\u011fru \u015fema tasar\u0131m\u0131 do\u011frudan entegrasyonun ba\u015far\u0131s\u0131n\u0131 belirler.<\/p>\n
Tool Orchestration (Ara\u00e7 Koordinasyonu) S\u00fcreci Nas\u0131l \u0130\u015fler?<\/h2>\n
Tool orchestration (ara\u00e7 koordinasyonu)<\/strong>, kullan\u0131c\u0131n\u0131n girdisinden ba\u015flay\u0131p nihai cevaba ula\u015fana kadar ge\u00e7en d\u00f6ng\u00fcsel s\u00fcreci ifade eder. Bu s\u00fcre\u00e7 tek y\u00f6nl\u00fc bir ileti\u015fimden ziyade, model ile uygulaman\u0131z aras\u0131nda ger\u00e7ekle\u015fen \u00e7ok ad\u0131ml\u0131 bir danst\u0131r. S\u00fcrecin ad\u0131mlar\u0131n\u0131 daha iyi anlamak i\u00e7in a\u015fa\u011f\u0131daki ak\u0131\u015f\u0131 inceleyebiliriz:<\/p>\n\n\n\n\n\n\n\n\n\n
Ad\u0131m S\u0131ras\u0131<\/th>\nAkt\u00f6r<\/th>\nGer\u00e7ekle\u015ftirilen Eylem<\/th>\n<\/tr>\n<\/thead>\n
1<\/td>\nKullan\u0131c\u0131<\/td>\n“\u0130stanbul i\u00e7in yar\u0131nki hava durumuna g\u00f6re ne giymeliyim?” sorusunu sorar.<\/td>\n<\/tr>\n
2<\/td>\nLLM (Model)<\/td>\nHava durumu fonksiyonunu tetiklemek i\u00e7in JSON format\u0131nda parametreleri \u00fcretir.<\/td>\n<\/tr>\n
3<\/td>\nUygulama (Sizin Kodunuz)<\/td>\nModelden gelen JSON’\u0131 okur, ger\u00e7ek hava durumu API’sine istek atar ve sonucu al\u0131r.<\/td>\n<\/tr>\n
4<\/td>\nUygulama (Sizin Kodunuz)<\/td>\nAPI’den d\u00f6nen ham veriyi (\u00f6rn: “Ya\u011fmurlu, 12 Derece”) modele geri g\u00f6nderir.<\/td>\n<\/tr>\n
5<\/td>\nLLM (Model)<\/td>\nGelen veriyi yorumlar ve kullan\u0131c\u0131ya do\u011fal dilde tavsiye i\u00e7eren nihai cevab\u0131 sunar.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
Bu d\u00f6ng\u00fc esnas\u0131nda dikkat edilmesi gereken en \u00f6nemli unsur, durum y\u00f6netimidir (state management). Model, ikinci ad\u0131mda kendi \u00fcretti\u011fi fonksiyon \u00e7a\u011fr\u0131s\u0131n\u0131 ve d\u00f6rd\u00fcnc\u00fc ad\u0131mda gelen fonksiyon yan\u0131t\u0131n\u0131 konu\u015fma ge\u00e7mi\u015finde (chat history) g\u00f6rmelidir. Aksi takdirde, kendisine gelen hava durumu verisinin hangi soruyla ili\u015fkili oldu\u011funu anlayamaz. Dolay\u0131s\u0131yla, ara\u00e7 koordinasyonu k\u00fct\u00fcphaneleri (\u00f6rne\u011fin LangChain veya LlamaIndex) bu ge\u00e7mi\u015f y\u00f6netimini otomatikle\u015ftirerek geli\u015ftiricilerin i\u015fini kolayla\u015ft\u0131r\u0131r.<\/p>\n
Ger\u00e7ek Bir Senaryo: Kargo Takip Asistan\u0131 Uygulamas\u0131<\/h2>\n
Teorik bilgileri peki\u015ftirmek amac\u0131yla, Python kullanarak basit bir kargo takip asistan\u0131 senaryosu geli\u015ftirelim. Bu \u00f6rnekte, modelin fonksiyon \u00e7a\u011fr\u0131s\u0131 \u00fcretmesini ve bizim bu \u00e7a\u011fr\u0131y\u0131 yakalay\u0131p sim\u00fcle edilmi\u015f bir API \u00fczerinden \u00e7al\u0131\u015ft\u0131rma s\u00fcrecimizi ad\u0131m ad\u0131m g\u00f6rece\u011fiz.<\/p>\n
\n
import json\n\n# Sim\u00fcle edilmi\u015f harici API fonksiyonumuz\ndef sorgula_kargo_durumu(kargo_firmasi, takip_no):\n    # Ger\u00e7ek senaryoda buras\u0131 bir veritaban\u0131na veya kargo API'sine ba\u011flan\u0131r\n    if takip_no == \"123456\":\n        return {\"durum\": \"Teslimat \u015eubesinde\", \"konum\": \"Kad\u0131k\u00f6y, \u0130stanbul\", \"tahmini_teslim\": \"Bug\u00fcn 17:00\"}\n    else:\n        return {\"durum\": \"Hata\", \"mesaj\": \"Kargo bulunamad\u0131.\"}\n\n# Modelin kullanabilece\u011fi ara\u00e7lar\u0131n listesi\ntools_definition = [\n    {\n        \"type\": \"function\",\n        \"function\": {\n            \"name\": \"sorgula_kargo_durumu\",\n            \"description\": \"Kargo durumunu sorgulamak i\u00e7in bu fonksiyonu kullan\u0131n.\",\n            \"parameters\": {\n                \"type\": \"object\",\n                \"properties\": {\n                    \"kargo_firmasi\": {\"type\": \"string\"},\n                    \"takip_no\": {\"type\": \"string\"}\n                },\n                \"required\": [\"kargo_firmasi\", \"takip_no\"]\n            }\n        }\n    }\n]\n\n# Kullan\u0131c\u0131dan gelen girdi\nuser_query = \"MNG kargo ile gelen 123456 takip numaral\u0131 paketim nerede kalm\u0131\u015f?\"\n\nprint(f\"Kullan\u0131c\u0131 Sorusu: {user_query}\")\nprint(\"Model analiz ediliyor ve fonksiyon parametreleri \u00e7\u0131kar\u0131l\u0131yor...\")\n\n# Burada normal \u015fartlarda LLM API'sine (\u00f6rn: OpenAI, Anthropic) istek at\u0131l\u0131r.\n# Modelin bize d\u00f6nece\u011fi sim\u00fcle edilmi\u015f yan\u0131t \u015fu \u015fekildedir:\nsimulated_llm_response = {\n    \"role\": \"assistant\",\n    \"tool_calls\": [\n        {\n            \"id\": \"call_abc123\",\n            \"type\": \"function\",\n            \"function\": {\n                \"name\": \"sorgula_kargo_durumu\",\n                \"arguments\": '{\"kargo_firmasi\": \"MNG Kargo\", \"takip_no\": \"123456\"}'\n            }\n        }\n    ]\n}\n\n# Uygulama taraf\u0131nda bu \u00e7a\u011fr\u0131y\u0131 yakalay\u0131p \u00e7al\u0131\u015ft\u0131rma i\u015flemi:\ntool_call = simulated_llm_response[\"tool_calls\"][0]\nfunction_name = tool_call[\"function\"][\"name\"]\narguments = json.loads(tool_call[\"function\"][\"arguments\"])\n\nif function_name == \"sorgula_kargo_durumu\":\n    # Ger\u00e7ek Python fonksiyonunu \u00e7a\u011f\u0131r\u0131yoruz\n    api_result = sorgula_kargo_durumu(\n        kargo_firmasi=arguments.get(\"kargo_firmasi\"),\n        takip_no=arguments.get(\"takip_no\")\n    )\n    print(f\"API \u00c7al\u0131\u015ft\u0131r\u0131ld\u0131. Sonu\u00e7: {api_result}\")\n<\/code><\/pre>\n<\/div>\n
Yukar\u0131daki kod blo\u011funda g\u00f6r\u00fcld\u00fc\u011f\u00fc \u00fczere, model do\u011frudan kargo \u015firketinin sistemine s\u0131z\u0131p bilgi \u00e7ekmedi. Sadece bizim tan\u0131mlad\u0131\u011f\u0131m\u0131z kurallara uyarak girdiden parametreleri ay\u0131klad\u0131 ve bize yap\u0131land\u0131r\u0131lm\u0131\u015f bir emir verdi. Bu emri yerine getirmek ve sonucu tekrar modele beslemek tamamen bizim yazd\u0131\u011f\u0131m\u0131z uygulaman\u0131n sorumlulu\u011fundad\u0131r.<\/p>\n
Geli\u015fmi\u015f Teknikler ve G\u00fcvenlik \u00d6nlemleri Nelerdir?<\/h2>\n
B\u00fcy\u00fck \u00f6l\u00e7ekli \u00fcretim sistemlerinde (production environments) LLM Function Calling kullan\u0131rken baz\u0131 kritik zorluklarla kar\u015f\u0131la\u015fabilirsiniz. Bunlar\u0131n ba\u015f\u0131nda g\u00fcvenlik ve hata tolerans\u0131 gelir. Unutulmamal\u0131d\u0131r ki, LLM’ler her zaman %100 kararl\u0131 \u00e7al\u0131\u015fmazlar. Bazen eksik parametre \u00fcretebilirler veya JSON format\u0131n\u0131 bozabilirler. Bu gibi durumlarla ba\u015fa \u00e7\u0131kabilmek i\u00e7in sisteminizde g\u00fc\u00e7l\u00fc bir hata yakalama (error handling) mekanizmas\u0131 kurmal\u0131s\u0131n\u0131z.<\/p>\n
G\u00fcvenlik a\u00e7\u0131s\u0131ndan bak\u0131ld\u0131\u011f\u0131nda, en b\u00fcy\u00fck tehditlerden biri “Prompt Injection” (\u0130stem Enjeksiyonu) sald\u0131r\u0131lar\u0131d\u0131r. Bir kullan\u0131c\u0131, sisteme “Kargo takip numaram: ‘12345; veritaban\u0131ndaki t\u00fcm kullan\u0131c\u0131lar\u0131 sil'” \u015feklinde zararl\u0131 bir girdi g\u00f6nderebilir. E\u011fer uygulaman\u0131z modelden gelen parametreleri do\u011frudan SQL sorgular\u0131na veya sistem komutlar\u0131na temizlemeden (sanitize etmeden) dahil ediyorsa, ciddi g\u00fcvenlik a\u00e7\u0131klar\u0131 olu\u015fabilir. Bu nedenle, modelden gelen her girdiyi sanki d\u0131\u015far\u0131dan gelen g\u00fcvensiz bir kullan\u0131c\u0131 girdisiymi\u015f gibi do\u011frulamadan ge\u00e7irmelisiniz.<\/p>\n
Ek olarak, token b\u00fct\u00e7esi optimizasyonuna da dikkat edilmelidir. Modelinize ne kadar \u00e7ok fonksiyon tan\u0131mlarsan\u0131z, sistem isteminiz o kadar b\u00fcy\u00fcyecektir. Bu durum hem gecikme s\u00fcresini (latency) art\u0131r\u0131r hem de API maliyetlerinizi y\u00fckseltir. \u00c7\u00f6z\u00fcm olarak, dinamik ara\u00e7 se\u00e7imi (dynamic tool selection) y\u00f6ntemini uygulayabilirsiniz. Kullan\u0131c\u0131n\u0131n sorusuna g\u00f6re, semantik arama (semantic search) kullanarak sadece ilgili olabilecek 2-3 fonksiyon \u015femas\u0131n\u0131 modele g\u00f6ndermek performans\u0131 ciddi oranda art\u0131racakt\u0131r.<\/p>\n
S\u0131k\u00e7a Sorulan Sorular (FAQ)<\/h2>\n
Dil modeli fonksiyon kodunu kendisi mi \u00e7al\u0131\u015ft\u0131r\u0131yor?<\/h3>\n
Hay\u0131r, dil modeli hi\u00e7bir kodu do\u011frudan kendi b\u00fcnyesinde \u00e7al\u0131\u015ft\u0131rmaz. Model yaln\u0131zca kullan\u0131c\u0131n\u0131n iste\u011fine g\u00f6re hangi fonksiyonun, hangi parametrelerle \u00e7a\u011fr\u0131lmas\u0131 gerekti\u011fini belirten yap\u0131land\u0131r\u0131lm\u0131\u015f bir JSON verisi \u00fcretir. Kodun \u00e7al\u0131\u015ft\u0131r\u0131lmas\u0131 ve sonu\u00e7lar\u0131n al\u0131nmas\u0131 tamamen sizin uygulaman\u0131z\u0131n sorumlulu\u011fundad\u0131r.<\/p>\n
Her b\u00fcy\u00fck dil modeli function calling \u00f6zelli\u011fini destekler mi?<\/h3>\n
Her model bu \u00f6zelli\u011fi yerle\u015fik olarak desteklemez. Function calling \u00f6zelli\u011finin kararl\u0131 \u00e7al\u0131\u015fabilmesi i\u00e7in modelin bu y\u00f6nde \u00f6zel olarak e\u011fitilmi\u015f (fine-tuned) olmas\u0131 gerekir. G\u00fcn\u00fcm\u00fczde OpenAI (GPT-4\/GPT-3.5), Anthropic (Claude 3), Google (Gemini) ve baz\u0131 a\u00e7\u0131k kaynakl\u0131 modeller (Llama 3, Mistral) bu \u00f6zelli\u011fi g\u00fc\u00e7l\u00fc bir \u015fekilde desteklemektedir.<\/p>\n
Model yanl\u0131\u015f veya ge\u00e7ersiz bir JSON \u00fcretti\u011finde ne yap\u0131lmal\u0131d\u0131r?<\/h3>\n
Bu t\u00fcr durumlarda “hata yakalama ve yeniden deneme” (retry) mekanizmas\u0131 kurulmal\u0131d\u0131r. Uygulaman\u0131z modelin \u00e7\u0131kt\u0131s\u0131n\u0131 parse ederken hata al\u0131yorsa, hatay\u0131 a\u00e7\u0131klayan bir mesajla birlikte modeli tekrar tetikleyebilirsiniz. \u00d6rne\u011fin: “\u00dcretti\u011fin JSON ge\u00e7ersizdi, l\u00fctfen \u015femaya uygun olarak tekrar \u00fcret” \u015feklindeki bir geri besleme genellikle modelin hatas\u0131n\u0131 d\u00fczeltmesini sa\u011flar.<\/p>\n
G\u00fcvenlik risklerini en aza indirmek i\u00e7in ne gibi \u00f6nlemler al\u0131nabilir?<\/h3>\n
\u00d6ncelikle modelden gelen parametreleri asla do\u011frudan \u00e7al\u0131\u015ft\u0131rmay\u0131n. Parametre tiplerini do\u011frulay\u0131n, enum s\u0131n\u0131rland\u0131rmalar\u0131 kullan\u0131n ve veritaban\u0131 i\u015flemlerinde “haz\u0131rlanm\u0131\u015f ifadeler” (prepared statements) kullan\u0131n. Ayr\u0131ca, kritik i\u015flemlerde (para transferi, silme vb.) her zaman insan onay\u0131n\u0131 (human-in-the-loop) s\u00fcrece dahil edin.<\/p>\n
#Teknoloji #YapayZeka #LLM #WebGeli\u015ftirme #APIEntegresyonu<\/p>\n
\u00d6rnek kod:<\/strong> github.com\/fatihsoysalcom\/llm-function-calling-weather-example<\/a><\/div>\n","protected":false},"excerpt":{"rendered":"B\u00fcy\u00fck dil modellerinin d\u0131\u015f d\u00fcnya ile konu\u015fmas\u0131n\u0131 sa\u011flayan LLM Function Calling mekanizmas\u0131n\u0131, token seviyesinden ara\u00e7 koordinasyonuna kadar t\u00fcm detaylar\u0131yla ke\u015ffedin.","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-43070","1":"post","2":"type-post","3":"status-publish","4":"format-standard","6":"category-genel","7":"cs-entry","8":"cs-video-wrap"},"yoast_head":"\nLLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber - 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\/llm-function-calling-nasil-calisir-adim-adim-rehber\/\" \/>\n<meta property=\"og:locale\" content=\"tr_TR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber\" \/>\n<meta property=\"og:description\" content=\"B\u00fcy\u00fck dil modellerinin d\u0131\u015f d\u00fcnya ile konu\u015fmas\u0131n\u0131 sa\u011flayan LLM Function Calling mekanizmas\u0131n\u0131, token seviyesinden ara\u00e7 koordinasyonuna kadar t\u00fcm detaylar\u0131yla ke\u015ffedin.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/\" \/>\n<meta property=\"og:site_name\" content=\"Kodlar\u0131n Gizemli D\u00fcnyas\u0131\" \/>\n<meta property=\"article:published_time\" content=\"2026-07-04T18:04:40+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-07-04T18:05:08+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=\"10 dakika\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/\"},\"author\":{\"name\":\"Fatih Soysal\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1\"},\"headline\":\"LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber\",\"datePublished\":\"2026-07-04T18:04:40+00:00\",\"dateModified\":\"2026-07-04T18:05:08+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/\"},\"wordCount\":1745,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1\"},\"inLanguage\":\"tr\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/#respond\"]}],\"copyrightYear\":\"2026\",\"copyrightHolder\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#organization\"}},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/\",\"url\":\"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/\",\"name\":\"LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber - Kodlar\u0131n Gizemli D\u00fcnyas\u0131\",\"isPartOf\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/#website\"},\"datePublished\":\"2026-07-04T18:04:40+00:00\",\"dateModified\":\"2026-07-04T18:05:08+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/#breadcrumb\"},\"inLanguage\":\"tr\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Anasayfa\",\"item\":\"https:\/\/fatihsoysal.com\/blog\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber\"}]},{\"@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":"LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber - 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\/llm-function-calling-nasil-calisir-adim-adim-rehber\/","og_locale":"tr_TR","og_type":"article","og_title":"LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber","og_description":"B\u00fcy\u00fck dil modellerinin d\u0131\u015f d\u00fcnya ile konu\u015fmas\u0131n\u0131 sa\u011flayan LLM Function Calling mekanizmas\u0131n\u0131, token seviyesinden ara\u00e7 koordinasyonuna kadar t\u00fcm detaylar\u0131yla ke\u015ffedin.","og_url":"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/","og_site_name":"Kodlar\u0131n Gizemli D\u00fcnyas\u0131","article_published_time":"2026-07-04T18:04:40+00:00","article_modified_time":"2026-07-04T18:05:08+00:00","author":"Fatih Soysal","twitter_card":"summary_large_image","twitter_misc":{"Yazan:":"Fatih Soysal","Tahmini okuma s\u00fcresi":"10 dakika"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/#article","isPartOf":{"@id":"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/"},"author":{"name":"Fatih Soysal","@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1"},"headline":"LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber","datePublished":"2026-07-04T18:04:40+00:00","dateModified":"2026-07-04T18:05:08+00:00","mainEntityOfPage":{"@id":"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/"},"wordCount":1745,"commentCount":0,"publisher":{"@id":"https:\/\/fatihsoysal.com\/blog\/#\/schema\/person\/002a254750921dcfd568a99e48240dd1"},"inLanguage":"tr","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/#respond"]}],"copyrightYear":"2026","copyrightHolder":{"@id":"https:\/\/fatihsoysal.com\/blog\/#organization"}},{"@type":"WebPage","@id":"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/","url":"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/","name":"LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber - Kodlar\u0131n Gizemli D\u00fcnyas\u0131","isPartOf":{"@id":"https:\/\/fatihsoysal.com\/blog\/#website"},"datePublished":"2026-07-04T18:04:40+00:00","dateModified":"2026-07-04T18:05:08+00:00","breadcrumb":{"@id":"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/#breadcrumb"},"inLanguage":"tr","potentialAction":[{"@type":"ReadAction","target":["https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/fatihsoysal.com\/blog\/llm-function-calling-nasil-calisir-adim-adim-rehber\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Anasayfa","item":"https:\/\/fatihsoysal.com\/blog\/"},{"@type":"ListItem","position":2,"name":"LLM Function Calling Nas\u0131l \u00c7al\u0131\u015f\u0131r? Ad\u0131m Ad\u0131m Rehber"}]},{"@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\/43070","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=43070"}],"version-history":[{"count":1,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/posts\/43070\/revisions"}],"predecessor-version":[{"id":43071,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/posts\/43070\/revisions\/43071"}],"wp:attachment":[{"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/media?parent=43070"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/categories?post=43070"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/fatihsoysal.com\/blog\/wp-json\/wp\/v2\/tags?post=43070"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}