There is a technology that has attracted a lot of attention on social media lately:

A person sitting in front of the computer talks to the artificial intelligence on his screen as if he were talking to a normal person.

“What is my schedule today?”

says.

AI responds:

“You have a meeting at 10 o'clock. I suggest you complete the file you need to prepare beforehand. We can spare 90 minutes for video editing in the afternoon.”

Then user:

“Then remind me 30 minutes before the meeting.”

says.

Artificial intelligence:

“OK.”

says.

Another command is coming:

“Open Chrome, bring up YouTube Studio, and look at the statistics of the last video.”

Artificial intelligence is really starting to operate on the computer.

Here are people's “Is JARVIS from Iron Man back?” asks.

But what actually happens is a little more interesting.

What actually is JARVIS?

Let's separate the name issue first.

J.A.R.V.I.S. is the fictional artificial intelligence assistant in the Iron Man universe.

In the real world, the name “JARVIS” is used in various real projects that allow people to talk to the computer, receive information, and make the computer do some work.

There is not a single official “JARVIS system” today.

On the contrary, there is more than one project under the name JARVIS.

Examples that actually exist now

Open source isair/Jarvis project is being developed as a local-first voice assistant that runs on a computer.

Current features of the project include:

• Voice speech

• Personal memory

• Web search

• Weather

• Time information

• Screenshot and OCR

• File access

• Personal diary/memory

• MCP tools

• Browser automation

• Smart home integrations

• Voice dictation

. Additionally, since it can work with local models, it also offers a structure that can be used without the need for a mandatory cloud AI subscription.

Another commercial JARVIS product runs on Windows, macOS and Linux; It offers features such as controlling desktop applications, real-time voice conversation, automating workflows and managing smart home devices.

So the “JARVIS” image you see on social media doesn't have to be completely fictional.

But there is likely more than a single model behind the system shown in the video.

So how does that system actually work?

In the simplest terms, there is approximately the following structure behind a JARVIS:

You're talking

↓

Voice recognition system

↓

Artificial intelligence brain

↓

Memory + tool selection + planning

↓

Computer / internet / calendar / files / applications

↓

Result

↓

Artificial intelligence responds by voice

So actually, what you are faced with is not a single artificial intelligence, but several technologies connected to each other.

1. Ear: Voice Recognition

The first part is to convert the user's voice into text.

For example:

“I have a dentist tomorrow at three o'clock, don't forget that.”

received from microphone.

System this:

“I have a dental appointment tomorrow at 15:00.”

.

Two different approaches can be used for this.

Cloud-based system

Audio is sent over the internet to a powerful speech model.

Advantage:

• Easy to install.

• The quality is high.

• Turkish speech recognition could be easier.

• Does not require a powerful computer.

Disadvantage:

• Internet required.

• Fees may apply depending on usage.

• Voice data may go to the service you selected.

Local system

Speech is processed on the computer.

For example, local speech recognition models such as Whisper can be used in Jarvis projects. The current configuration of the open source Jarvis project includes Whisper as one of the local voice recognition options.

Advantage:

Sound can be processed without leaving the computer.

Disadvantage:

• Installation is more complex.

• The performance of the computer is important.

• Larger models may require more RAM/VRAM.

2. Brain: Artificial Intelligence Model

After the audio is converted to text, the real decision-making part begins.

For example, the system receives the following sentence:

“Check my meetings tomorrow and come up with a study plan that suits me.”

Here, it is not enough for an ordinary chatbot to just answer.

The real JARVIS:

1. Calendar reading

2. Understanding meetings

3. Finding free time

4. Setting priorities

5. Plan creation

6. Adding transactions to the calendar if necessary

require.

At this point artificial intelligence models that can drive vehicles comes into play.

With OpenAI's current APIs, you can connect your own functions to the model, add tools such as web search and file search, and create real-time voice applications.

3. Hands: Tool Calling

I think this is the main point that distinguishes JARVIS from regular ChatGPT.

A chatbot:

“According to your calendar, you have two meetings today.”

may say.

JARVIS can actually open and check the calendar.

Because artificial intelligence tools is defined.

For example, you can add the following functions to the system:

get_calendar()
create_calendar_event()
get_weather()
open_application()
close_application()
search_web()
read_file()
write_file()
send_email()
take_screenshot()
move_mouse()
click_screen()
run_terminal_command()

Artificial intelligence understands the user's request and selects the appropriate tool.

For example:

“Add a meeting for tomorrow at 14.00.”

AI:

create_calendar_event()

tool.

4. MCP: The System That Extends JARVIS' Hands and Feet

One of the technologies that has become very important in such AI agents recently is MCP — Model Context Protocol.

Thanks to MCP, different vehicles can be connected to an artificial intelligence system.

For example:

JARVIS

↓

MCP

↓

Google Calendar

↓

GitHub

↓

File system

↓

Web browser

↓

Home Assistant

can be established.

The current open source JARVIS project can also connect to MCP servers and include browser automation, smart home, GitHub, database and different tools into the system.

Therefore JARVIS:

“system that multiplies hands and external connections”

.

5. Memory: How Does JARVIS Remember You?

The part that really makes the system feel “like JARVIS” is the memory.

For example today:

“I usually prepare YouTube videos in the evening.”

you said.

One week later:

“When should I make a video today?”

, the system can remember it.

For this purpose, a separate memory system is established instead of sending all old conversations to artificial intelligence over and over again.

For example:

SQLite

•  

vector database

•  

recessed models

available.

Alternatively, on simpler systems, Markdown files or a local database can be used.

In the current open source Jarvis project, the speech memory is stored on the computer and the local embedding system can be used for meaning-based search.

6. How Does JARVIS Control the Computer?

This is the part that makes videos on social media viral.

Artificial intelligence:

“Open Chrome.”

, Chrome can actually open it.

To do this, a command can be sent to the operating system.

For example:

Windows

PowerShell / Python / Windows API

macOS

AppleScript / Shortcuts / Python / macOS API

Linux

Shell / Python / system tools

available.

In more advanced systems, artificial intelligence can also see the screenshot and direct mouse and keyboard movements.

There are two different approaches here.

Method 1 — Predefined tools

For example:

open_chrome()

open_spotify()

create_folder()

.

This method is safer.

Method 2 — Seeing and using the computer

Artificial intelligence:

• Sees the screen

• Detects buttons

• Moves the mouse

• Clicks

• Writes text

• Scrolls through pages

This is more powerful, but also more error prone.

Systems like Anthropic's approach to computing enable AI to see the screen and perform operations that humans do on a computer, such as cursor movement, clicking, and typing; However, this technology is still an area that can make mistakes and needs to be limited carefully.

What exactly can JARVIS do on your computer?

A well-designed system can, for example:

Morning

“Good morning, Jarvis.”

Jarvis:

“Good morning. You have a meeting at 10.00 today. You are free at 13.00. The weather is 18 degrees. Your priority task is to complete the video script.”

Later

“Warn me 30 minutes before the meeting.”

Jarvis:

Adds a reminder to the calendar.

During work

“Open YouTube Studio.”

Opens Chrome.

Goes to YouTube Studio.

More advanced usage

“Check the performance of the last video, compare it with the previous three videos, and give me a brief summary.”

Here is the system:

1. Can open the browser.

2. Can go to YouTube Studio.

3. Can read data.

4. Can compare previous data.

5. Can have the AI model analyzed.

6. Can explain the result out loud.

This is actually the "JARVIS" effect you see on social media. from chain vehicle use is coming.

Can You Make Your Own JARVIS in Türkiye?

Yes.

It's even easier than you think.

But there are three different ways here.

PATH 1 — Using Ready-made JARVIS

If you do not want to write code, the easiest way is to use a ready-made JARVIS application.

For example jarvisdesktop.com currently:

• Windows

• macOS

• Linux

.

Current prices on their website:

Regular

Monthly: $14

Regular annual

Annual: $132

Pro

Monthly: $16

and it is stated on the site that additional fees may occur depending on the tax/card transaction.

The approximate USD/TRY exchange rate on October 5, 2026 was 49.15 TL.

So roughly:

Plan	Dollar	Approx. TL
Regular monthly	$14	688 TL
Regular annual	$132	6.488 TL/year
Pro monthly	$16	786 TL

These approximately exchange rate; The final amount to be withdrawn from the card in Türkiye may differ due to the bank's exchange rate, taxes and fees added during payment.

Installation time

Ready product:

10–30 minutes

PATH 2 — Free Open Source JARVIS

There is a much more interesting option if you can play around with the code.

isair/Jarvis free for personal use and uses native AI as the default approach.

The project offers packages for Windows, macOS and Linux.

Setup logic:

Jarvis

↓

Local speech recognition

↓

Don't Be

↓

Local AI model

↓

MCP tools

↓

Browser/file/web/smart home etc.

.

According to the project's own documentation, OpenAI compatible local model servers such as Ollama, LM Studio, llama.cpp, vLLM and similar can also be used.

Cost

Software side for personal use:

0 TL

could be.

The main cost is your computer.

Is internet required?

Cloud connection not required for basic AI conversation in purely local use.

However:

• Web search

• Online services

• Calendar

• Internet data

• External APIs

.

The open source Jarvis project also notes that local speech can work without a cloud AI account, but the web and connected tools can use a network connection.

Installation time

Installing the ready project to the computer:

1–3 hours

level.

But configuring each feature properly:

1 day or more

may take.

The hardware of the computer and the model used affect this time.

PATH 3 — Making Your Own JARVIS From Scratch

This is the most powerful option.

But this is the one that requires the most effort.

For a truly personal JARVIS, approximately the following architecture can be constructed:

 🎙️ MICROPHONE
                       │
                       ▼
               SPEECH RECOGNITION
                 Whisper etc.
                       │
                       ▼
              🧠 AI ORCHESTRATOR
                       │
        ┌──────────────┼──────────────┐
        ▼ ▼ ▼
     MEMORY PLANNING TOOLS
        │ │ │
     SQLite Tasks MCP/API
        │ │ │
        └──────────────┼──────────────┘
                       │
          ┌────────────┼────────────┐
          ▼ ▼ ▼
      COMPUTER CALENDAR INTERNET
          │ │ │
          ▼ ▼ ▼
      APPLICATIONS REMINDER WEB
                       │
                       ▼
                 🔊 VOICE ANSWER

With what technologies can you make such a system?

Programming language

One of the most logical options:

Python

Because:

• There are many artificial intelligence libraries.

• API integrations are easy.

• Windows/macOS/Linux support is good.

• Audio processing tools available.

• Automation libraries are extensive.

For interface:

Tauri

or

Electron

available.

For web-based interface:

React / Next.js

available.

Which model can be used on the artificial intelligence side?

If you're going to use the cloud, check out OpenAI's current GPT-Realtime-2 model available.

According to OpenAI's current model sheet GPT-Realtime-2:

• Voice input

• Audio output

• Visual input

• Function call

• Vehicle operation

supports. Current listed price in text entry $4 / million tokens in text output $24 / million tokens; $32 / million audio token in audio output $64 / million audio token.

This model is very suitable for directly creating real-time voice JARVIS experience.

How much could JARVIS' monthly AI cost?

The biggest misunderstanding here is:

“I used AI, then there is a fixed monthly fee.”

No.

In API-based systems, the cost varies depending on the amount of usage.

Let's do a very rough calculation using the listed voice prices of GPT-Realtime-2 for example.

User:

30 seconds talking

AI:

30 seconds responding

be.

According to OpenAI's stated voice token rates:

1 minute ≈ $0.048

.

This account:

• User talks for 30 seconds

• AI talking for 30 seconds

• No additional text/tool costs

is a simplified example.

OpenAI also specifically states that the actual Realtime cost varies depending on session duration, conversation history, input/output tokens and tools used.

For example, if you talk for 30 minutes a day

Approx:

900 minutes/month

becomes

With just this sample audio cost:

about $43/month

remove.

Approximate exchange rate today:

approximately 2.100 TL/month

level.

60 minutes a day

Approx:

$86/month

ie:

approximately 4.200 TL/month

.

But do not consider these as final invoices.

Actual figure:

• Talk time

• AI talking time

• Context held by the model

• Web searches

• Computer use

• Other APIs

• Calendar and third-party services

affects. OpenAI also states that the Realtime API cost is calculated based on usage measurement and that backend tools may also create costs.

Is a cheaper method possible?

Yes.

One of the most logical methods is the hybrid system.

For example:

Sound

Local Whisper

Simple commands

Local small model

Memory

SQLite + native embedding

Web

Web API when needed

Complex thinking

Cloud AI

Thus:

“Open Chrome.”

.

But:

“Analyze the performance of my YouTube videos this week, compare them with previous videos, and come up with a content plan for next week.”

.

The current open source JARVIS project itself implements a multi-layer modeling approach that may use a smaller model for faster tasks and a more capable model for more powerful tasks.

How to connect a calendar to JARVIS?

For example, you can use Google Calendar API.

User:

“Add a meeting tomorrow at 15.00.”

says:

1. AI decodes the date.

2. Decodes the time.

3. Specifies the title.

4. Calls the Calendar API.

5. Creates the event.

6. Reports the result to the user.

Later:

“What is my schedule today?”

can read events from the same API.

How to make a reminder system?

On Windows:

• Task Scheduler

• Python timers

available.

In more advanced systems:

• APScheduler

• n8n

• Celery

• Redis

can be used.

For example:

“Summarize my work today at 08.00 every morning.”

command:

Scheduler → JARVIS → Calendar → Tasks → Weather → AI summary → Voice reply

.

Can JARVIS read your emails?

Technically yes.

Gmail or Outlook API can be connected.

For example:

“Summarize important emails received today.”

, the system:

1. Retrieves emails.

2. Classifies the important ones.

3. Summaries.

4. Explains out loud.

One can go even further:

“Prepare a reply to Ahmet's e-mail saying 14.00 tomorrow is convenient.”

The system can prepare the e-mail.

But here human approval rather than direct automatic sending authorization.

Should JARVIS take full control of your computer?

I would definitely hit the brakes here.

Because artificial intelligence:

“Do anything you want on the computer.”

is a bad idea.

For example, if the AI gets it wrong:

“Clear files.”

may misinterpret the command.

Safer system:

Low risk jobs

Auto.

• Open Chrome

• Open Spotify

• Read weather

• Fetch calendar

• Take a screenshot

Medium risk jobs

Request confirmation.

• Move file

• Create email draft

• Add event to calendar

High risk jobs

Strictly human approval.

• Send email

• Delete file

• Money transfer

• Change password

• Account closure

• Purchase

This approach is also seen in current JARVIS projects; örneğin açık kaynaklı JARVIS'in açıklamasında riskli işlemler için insan onayı mekanizmaları bulunuyor.

Türkiye'de gerçekçi olarak ne kadar sürede yapılır?

Burada sosyal medya videolarının en çok abarttığı bölüm burası.

“Birkaç saatte Iron Man JARVIS yaptım.”

videolarına fazla inanmayın.

Temel sürüm

Şunları yapabilen:

• Sesli konuşma

• AI cevapları

• Basit bilgisayar komutları

• Web araması

Deneyimli bir geliştirici için: 1–3 gün

Orta seviye JARVIS

• Ses

• Hafıza

• Takvim

• Hatırlatmalar

• Web

• Dosyalar

• Uygulama kontrolü

• MCP

• Sesli planlama

Yaklaşık: 1–2 hafta

Gelişmiş sistem

• Ekranı görebilme

• Mouse/klavye kontrolü

• Tarayıcı otomasyonu

• Çok aşamalı görev planlama

• Hata yönetimi

• Yetki sistemi

• Onay mekanizması

• Kalıcı hafıza

• Güçlü masaüstü arayüzü

• Log sistemi

• Açılışta otomatik başlama

• Paketleme ve dağıtım

Yaklaşık: 4–8 hafta

Gerçek anlamda “Iron Man JARVIS”

Yani:

“Bütün bilgisayarı anla, tüm uygulamalarla konuş, beni hatırla, verdiğim uzun görevleri kendi başına çöz, hataları kendi düzelt, gerektiğinde interneti kullan, sürekli aktif ol.”

gibi bir sistem:

Birkaç günlük proje değildir.

Tek kişi için aylar sürebilir.

Çünkü zor olan yalnızca AI'ı bağlamak değil.

Asıl zor kısım:

güvenilirlik + yetki kontrolü + hafıza + araç entegrasyonları + hata yönetimi + güvenliktir.

Türkiye'de benim önerdiğim JARVIS sistemi

Bugün sıfırdan kuracak olsam üç aşamada ilerlerdim.

Aşama 1 — Ucuz prototip

Windows 11

•  

Python

•  

Whisper

•  

Ollama

•  

yerel küçük model

•  

SQLite

•  

MCP

Bu aşamada amaç:

“Konuşuyorum ve bilgisayar beni anlıyor.”

Aşama 2 — Gerçek asistan

Buna:

• Google Calendar

• Gmail

• Web arama

• Hava durumu

• Hatırlatmalar

• Dosya erişimi

• Chrome kontrolü

eklenir.

Artık:

“Bugün ne yapacağım?”

sorusuna gerçekten cevap verebilen bir sisteminiz olur.

Aşama 3 — Sosyal medyadaki JARVIS seviyesi

Sonra:

• Ekranı görme

• Bilgisayar kullanma

• Mouse kontrolü

• Klavye kontrolü

• Browser automation

• Uzun görev planlama

• Kişisel hafıza

• Sabah brifingi

• Otomatik rapor

• Onay sistemi

eklenir.

İşte videolarda gördüğünüz “JARVIS gerçekten bilgisayarı kullanıyor” görüntüsü esas olarak bu aşamada ortaya çıkıyor.

Peki bunun için güçlü bir bilgisayar şart mı?

Bulut tabanlı sistemde:

Hayır.

Normal bir Windows 11 laptop bile yeterli olabilir.

Çünkü asıl yapay zekâ bulutta çalışır.

Yerel AI tarafında ise bilgisayarın RAM'i ve varsa GPU'su önem kazanır.

Güncel açık kaynak Jarvis projesi küçük modelleri daha mütevazı donanımlarda çalıştırabilecek bir başlangıç seviyesi olarak veriyor; daha büyük yerel modeller için daha fazla belleğe ihtiyaç duyuluyor.

Pratik olarak:

16 GB RAM: temel yerel asistan için başlanabilir.

32 GB RAM: daha rahat.

64 GB+ / güçlü GPU: büyük yerel modeller ve daha ağır görüntü/AI görevleri için daha uygun.

Bunlar katı minimumlar değil; seçilen modelin boyutu, sıkıştırma yöntemi, bağlam uzunluğu ve aynı anda çalışan işlemler sonucu ciddi biçimde değişir. Açık kaynak JARVIS'in kendi dokümanı da model boyutu, quantization, context ve konuşma tanımanın bellek ihtiyacını etkilediğini belirtiyor.

Sonuç: Sosyal medyadaki JARVIS sahte mi?

Hayır, ama videoda gördüğünüz şey muhtemelen tek bir “mucize program” değil.

Gerçek sistem genellikle:

Ses tanıma

•  

Büyük dil modeli

•  

hafıza

•  

araç kullanımı

•  

MCP/API bağlantıları

•  

bilgisayar otomasyonu

•  

sesli cevap

•  

zamanlama

şeklinde birleştirilmiş bir AI ajanıdır.

Bugün Türkiye'de bunu yapmak teknik olarak mümkün.

Üstelik iki uç nokta da var:

Hazır çözüm: Aylık birkaç yüz TL seviyesinden başlayan abonelik.

Kendi sisteminiz: Açık kaynak ve yerel modellerle yazılım tarafında 0 TL'ye kadar indirilebilir; ancak kurulum ve bilgisayar gücü maliyeti size kalır.

Bulut API ile özel JARVIS: Kullanıma bağlı olarak aylık birkaç yüz TL'den birkaç bin TL'ye kadar çıkabilir.

Ve en önemli nokta:

Gerçek JARVIS'i oluşturan şey yapay zekânın kendisi değil, yapay zekânın bilgisayarınıza ve günlük hayatınızdaki araçlara kontrollü şekilde bağlanmasıdır.

Bugün bu teknoloji “Iron Man seviyesine geldi” demek için hâlâ erken.

Ama:

“Bilgisayarımla konuşayım, beni hatırlasın, günlük planımı çıkarsın, takvimimi okusun, web'de araştırsın ve bazı işleri gerçekten benim yerime yapsın.”

dediğiniz noktaya artık gerçekten çok yakınız.