AgentX · Get StartedBaşla
Does any jobHer işi yapar

AgentX

Say it, and it's done. Söyleyin, yapsın.

An autonomous AI agent that runs on your own computer. Whatever the task, it understands, plans, builds the tool it needs, and delivers end to end — writing code, making real phone calls, sending WhatsApp and email, filling out web forms, and producing real, openable files.

Kendi bilgisayarınızda çalışan, özerk bir yapay zeka ajanı. Hangi görev olursa olsun anlar, planlar, gereken aracı üretir ve işi uçtan uca teslim eder — kod yazar, gerçek telefon görüşmesi yapar, WhatsApp/e-posta gönderir, web formlarını doldurur, gerçek (açılabilir) dosyalar üretir.

Windows desktop appWindows masaüstü uygulaması Python 3.9+ 5 LLM providers5 LLM sağlayıcısı Bilingual TR · ENİki dilli TR · EN
38
BUILT-IN TOOLSYERLEŞİK ARAÇ
15
SKILLSYETENEK (SKILL)
5
PROVIDERSSAĞLAYICI
2
CONSOLESKONSOL
Live demoCanlı gösterim

Watch it in actionÇalışırken izleyin

A real conversation in the actual Agent Console — full screen, no cuts, no mockup.

Gerçek Agent Konsolu'nda, gerçek bir sohbet — tam ekran, kurgusuz, mockup değil.

CapabilitiesYetenekler

What it can doNeler yapabilir

Thirteen areas of real-world capability, all reachable from a single conversation.

Tek bir sohbetten erişilebilen, gerçek dünyaya dokunan on üç yetenek alanı.

Chat & Code
Sohbet & Kod
Choose from 5 providers (Anthropic, OpenAI, DeepSeek, Gemini, Claude Pro/Max) — reads and writes files, runs shell commands.
5 sağlayıcı arasından seçim (Anthropic, OpenAI, DeepSeek, Gemini, Claude Pro/Max) — dosya okur/yazar, kabuk komutu çalıştırır.
Documents
Belgeler
Extracts real data from .pdf/.docx/.xlsx and produces real, openable .pdf/.docx/.xlsx — also reads source code. Scanned or photographed documents fall back to OCR automatically.
.pdf/.docx/.xlsx'ten gerçek veri çıkarır, gerçek (açılabilir) .pdf/.docx/.xlsx üretir — kaynak kodu da okur. Taranmış/fotoğraflanmış belgelerde otomatik OCR'a düşer.
Phone
Telefon
Makes real phone calls with Twilio, talks, and listens for a spoken or keypad reply.
Twilio ile gerçek arama yapar, konuşur, sesli/tuşlu cevabı dinler.
WhatsApp
Sends and reads messages over WhatsApp Web or through the phone; auto-downloads and transcribes incoming voice messages, and can reply in its own voice — sent as a file attachment.
WhatsApp Web (kablosuz) ya da telefon üzerinden mesaj gönderir/okur; gelen sesli mesajları indirip dinleyebilir, kendi sesiyle (TTS) yanıt gönderebilir — dosya eki olarak.
Email
E-posta
Sends from any provider over SMTP/IMAP, reads the inbox.
Herhangi bir sağlayıcıdan (SMTP/IMAP) gönderir, gelen kutusunu okur.
Web
Web
Opens a page, reads its DOM, fills forms, clicks — by human-readable label, not CSS selector.
Bir sayfayı açar, DOM'unu okur, form doldurur, tıklar — CSS seçici değil, insan-okunabilir etiketle.
Desktop
Masaüstü
Finds other programs on the computer by name (Calculator, Settings, most Win32/WPF/UWP apps) and clicks/types into them.
Bilgisayardaki diğer programları (Hesap Makinesi, Ayarlar, çoğu Win32/WPF/UWP uygulaması) isimle bulup tıklar/yazar.
Search
Arama
Searches the web and returns title, URL, and summary.
İnternette arama yapıp başlık/URL/özet döner.
Scheduling
Zamanlama
Runs a task in the future — once, or daily/weekly — genuinely triggered by Windows Task Scheduler.
Bir görevi ileride (bir kez ya da günlük/haftalık) çalıştırır — Görev Zamanlayıcı ile gerçekten tetiklenir.
Self tool-writing
Kendi kendine araç yazma
When existing tools can't cover a task, it writes its own Python tool into custom_tools/ and activates it in the same chat — without ever touching core/.
Elindeki araçlar bir görevi karşılamıyorsa custom_tools/'a kendi Python aracını yazıp aynı sohbette etkinleştirir — core/'a hiç dokunmadan.
Android
Controls a connected phone — SMS, calls, screen mirroring, app control.
Bağlı telefonu kontrol eder — SMS, arama, ekran yansıtma, uygulama kontrolü.
Voice
Ses
Speaks (TTS), listens (STT), and can hold a full voice conversation.
Konuşur (TTS), dinler (STT), sesli sohbet edebilir.
Visual
Görsel
Generates images from text, or writes SVG directly for architecture/flow diagrams — shown as a thumbnail, click to enlarge and download.
Metinden görsel üretir ya da mimari/akış diyagramı için doğrudan SVG yazar — konsolda küçük resim olarak görünür, tıklayınca büyür ve indirilebilir.
Why AgentXNeden AgentX

Not a cloud sandbox — your machineBulut kum havuzu değil — kendi bilgisayarınız

Most "autonomous agent" products run in an isolated container on someone else's cloud, reaching only their own platform's APIs. AgentX makes the opposite design choice.

Çoğu "otonom ajan" ürünü, kendi bulut altyapısında izole bir konteynerde çalışır ve yalnızca kendi platformunun API'lerine erişir. AgentX tam tersi bir tasarım tercihi yapar.

DimensionBoyut AgentX Typical cloud agentTipik bulut ajanı
Runtime environmentÇalışma ortamı Your own computer — a real file systemKullanıcının kendi bilgisayarı — gerçek dosya sistemi Isolated cloud VM / containerİzole bulut sanal makine / konteyner
WhatsApp Your real WhatsApp account, via WhatsApp Web automationGerçek WhatsApp hesabınız, WhatsApp Web otomasyonuyla Usually absent, or official Business API onlyGenelde yok, ya da yalnızca resmi Business API
Phone & AndroidTelefon & Android Real calls (Twilio) + a connected Android phone via adbGerçek arama (Twilio) + bağlı bir Android telefonu adb ile kontrol Not availableYok
Desktop appsMasaüstü uygulamaları Finds and controls other Windows programs by nameDiğer Windows programlarını isimle bulup kontrol eder Not available — sandbox onlyYok — yalnızca kendi kum havuzu
Self-extensionKendi kendine genişleme Writes a new tool mid-task and uses it in the same chatGörev sırasında yeni bir araç yazıp aynı sohbette kullanabilir Fixed, predefined tool listSabit, önceden tanımlı araç listesi
Provider independenceSağlayıcı bağımsızlığı 5 different LLM providers, same tools5 farklı LLM sağlayıcısı, aynı araçlarla Locked to a single modelTek bir modele kilitli
Source codeKaynak kod Fully open, readable, editable from its own consoleTamamen açık, okunabilir, kendi konsolundan düzenlenebilir Closed boxKapalı kutu
Persistent personality/memoryKalıcı kişilik/bellek Human-editable .md files, survives restartsİnsan tarafından düzenlenebilen .md dosyaları, restart'a dayanır Usually single-session or opaque memoryGenelde tek oturumluk ya da kapalı bellek
This has a cost. Because AgentX runs on a real computer rather than an isolated sandbox, any tool with a real-world effect — sending a message, making a call, changing a file — asks for confirmation first. Autonomy is not the same as having no guardrails. Bunun bir bedeli var. AgentX izole bir kum havuzunda değil gerçek bir bilgisayarda çalıştığı için, gerçek dünyaya etkisi olan her araçtan (mesaj gönderme, arama yapma, dosya değiştirme) önce onay istenir. Otonomi, kontrolsüzlük demek değildir.
Multi-provider
Çok sağlayıcılı
Switch between 5 different LLM providers, same tool set every time.
5 farklı LLM sağlayıcısı arasında, aynı araç setiyle geçiş yapılabilir.
Persistent memory
Kalıcı bellek
Personality, rules, and memory live in editable .md files under library/ — survives model switches and restarts.
Kişilik, kurallar ve bellek library/ altındaki .md dosyalarında yaşar — model/restart'tan etkilenmez.
Transparent
Şeffaf
Every tool call is logged to logs.txt with a timestamp; the source is fully open and editable.
Her araç çağrısı zaman damgasıyla logs.txt'e yazılır; kaynak kod tamamen açık ve düzenlenebilir.
ArchitectureMimari

Two consoles, one agentİki konsol, bir ajan

One console to use AgentX, one to build it — and a clear boundary between the tools it ships with and the tools it writes for itself.

Biri AgentX'i kullanmak, diğeri onu geliştirmek için iki ayrı konsol — ve yerleşik araçlarla kendi yazdığı araçlar arasında net bir sınır.

AgentX system architecture diagram AgentX sistem mimarisi diyagramı
01
Agent Console — core/, port 3001
Agent Konsolu — core/, 3001 portu
The console you actually talk to. Backed by any of 5 providers, with the full 38-tool set and every installed skill and custom tool.
Gerçekten sohbet ettiğiniz konsol. 5 sağlayıcıdan biriyle çalışır, 38 aracın tamamına ve yüklü her skill/özel araca erişir.
02
Developer Console — developer/, port 3000
Geliştirici Konsolu — developer/, 3000 portu
Claude-only, via Claude Pro/Max OAuth — used to program AgentX itself. Unlike the agent, it can edit core/, the agent's own executable code.
Yalnızca Claude, Claude Pro/Max OAuth ile — AgentX'i programlamak için kullanılır. Ajanın aksine core/'u, ajanın kendi çalıştırılabilir kodunu değiştirebilir.
03
Tools → Skills → Custom Tools
Araçlar → Skiller → Özel Araçlar
38 built-in tools are given full JSON schemas. Each skill teaches how/when to use them. When nothing fits, the agent writes a brand-new tool into custom_tools/ — fully isolated from core/.
38 yerleşik araç modele tam JSON şema olarak verilir. Her skill bunların nasıl/ne zaman kullanılacağını öğretir. Hiçbiri yetmezse ajan custom_tools/'a yepyeni bir araç yazar — core/'dan tamamen izole.
04
Security boundaries
Güvenlik sınırları
write_file is scoped to library/, rag/, skills/, tools/, temp/ — core/ and .env can never be modified this way. Real-world, irreversible actions ask for confirmation before they run.
write_file yazma erişimi yalnızca library/, rag/, skills/, tools/, temp/ ile sınırlıdır — core/ ve .env bu yolla asla değiştirilemez. Gerçek dünyaya etkisi olan, geri alınamaz eylemler çalışmadan önce onay ister.
05
Doctor & Scheduler
Doctor & Zamanlayıcı
A Windows Task Scheduler entry checks every ~5 minutes: restarts a crashed console, and can propose a fix for a repeating error in logs.txt (backed up before applying). A second entry fires scheduled tasks on time.
Bir Görev Zamanlayıcı kaydı ~5 dakikada bir kontrol eder: çöken bir konsolu yeniden başlatır, logs.txt'te tekrarlayan bir hataya düzeltme önerebilir (uygulamadan önce yedekler). İkinci bir kayıt da zamanlanmış görevleri saatinde tetikler.
ProvidersSağlayıcılar

Bring your own modelKendi modelinizi getirin

providers.py talks to whichever provider is configured, calling each one according to its real, measured capabilities.

providers.py yapılandırılan sağlayıcıyla konuşur, her birini kendi gerçek/ölçülmüş yeteneğine göre çağırır.

ProviderSağlayıcı AuthenticationKimlik doğrulama NotesNotlar
Anthropic API keyAPI anahtarı Full tool-calling loop, adaptive thinkingTam araç çağrım döngüsü, adaptif düşünme
Claude Pro/Max OAuth — no API spendOAuth — API kredisi harcamaz Full tool-calling loop via Claude Agent SDK, adaptive thinking, effort levels (low–max)Claude Agent SDK ile tam araç çağrım döngüsü, adaptif düşünme, effort seviyeleri (low–max)
OpenAI API keyAPI anahtarı Full tool-calling loopTam araç çağrım döngüsü
DeepSeek API keyAPI anahtarı Full tool-calling loop; thinking only surfaces on deepseek-reasonerTam araç çağrım döngüsü; düşünme yalnızca deepseek-reasoner'da yüzeye çıkar
Gemini API keyAPI anahtarı Plain chat only — no tool-callingYalnızca düz sohbet — araç çağrımı yok
Mustafa Tokat
About the developerGeliştirici hakkında

Mustafa Tokat

Senior AI and full-stack engineer with 15+ years of software delivery and a decade of applied AI work — from one of the first Turkish voice assistants in 2015 to production-grade generative AI and agentic systems today. He works end to end: problem discovery, data and workflow design, model orchestration, deployment, security and client adoption.

15+ yıllık yazılım geliştirme ve bir on yıllık uygulamalı yapay zeka deneyimine sahip kıdemli bir AI ve full-stack mühendisi — 2015'te geliştirdiği ilk Türkçe sesli asistanlardan bugünün üretim kalitesinde üretken yapay zeka ve ajan sistemlerine uzanan bir yolculuk. Sorunu keşfetmekten veri/iş akışı tasarımına, model orkestrasyonundan dağıtıma, güvenlikten müşteri benimsemesine kadar uçtan uca çalışıyor.

Connect on LinkedInLinkedIn'de bağlantı kur
Get in touchİletişime geç

Request a live demoCanlı demo talep et

Leave your details and I'll personally walk you through AgentX.

Bilgilerini bırak, AgentX'i sana kişisel olarak göstereyim.

Get StartedBaşlangıç

Running in under a minuteBir dakikadan kısa sürede çalışır

No terminal required for everyday use — the desktop shortcut and tray icon handle everything.

Günlük kullanımda terminale gerek yok — masaüstü kısayolu ve tepsi simgesi her şeyi halleder.

1
Double-click the AgentX desktop shortcut
Masaüstündeki AgentX kısayoluna çift tıklayın
Created automatically the first time start.bat runs.
İlk start.bat çalıştırmasında otomatik oluşturulur.
2
Approve the one-time Administrator prompt
Bir kerelik yönetici izni verin
A single UAC window — needed for the setup checks that follow.
Tek bir UAC penceresi — ardından gelen kurulum kontrolleri için gerekir.
3
Setup checks and installs what's missing
Kurulum eksikleri kontrol edip kurar
setup/check.py verifies Python packages and the Claude Pro/Max connection, installing anything missing automatically.
setup/check.py Python paketlerini ve Claude Pro/Max bağlantısını kontrol edip eksikleri otomatik kurar.
4
Use the tray icon for everything else
Gerisi için tepsi simgesini kullanın
Right-click it for the Agent Console, Developer Console, Restart, Stop, Logs, and Language (TR/EN) — no terminal needed.
Sağ tıklayarak Agent Konsolu, Geliştirici Konsolu, Yeniden Başlat, Durdur, Günlükler ve Dil (TR/EN) seçeneklerine ulaşılır — terminale gerek kalmaz.
# to run it by hand (e.g. while developing):elle çalıştırmak isterseniz (ör. geliştirme sırasında):
start.bat
Windows 10/11 Python 3.9+ API keys optional — per featureAPI anahtarları isteğe bağlı — özellik bazlı