Free hands-on workshop · about an hour
Your own private ChatGPT.
Run a ChatGPT-style AI on your own laptop. No subscription, no API key, no account. Keep its chats on a drive you chose, switch the internet off and keep chatting.
- No code
- Windows
- macOS
- Offline
- Qwen3.5-2B
- Privacy
Explain compound interest with an example in rupees.
Put ₹10,000 in at 8% a year. After one year you have ₹10,800. In year two you earn 8% on ₹10,800, not on ₹10,000, so you end at ₹11,664. Interest earning interest: that is compounding.
No internet. Answered on this laptop. Saved on the drive you chose.
Watch first
Watch it once, then do it with us
A short walk-through: how ChatGPT and local AI differ, what a model actually is, then the whole setup on a real laptop, step by step. The follow-along page matches the video, so you can pause and catch up.
- ChatGPT vs local AI
- What a model is
- Install Jan
- Data off the C: drive
- First chat
- Internet off
- What is private
The idea
Restaurant or home kitchen
Cloud AI cooks your order in someone else’s kitchen. Local AI cooks it in yours. The kitchen is smaller, but nobody else sees what you made. Flip between them.
The model is a file on your drive. Jan loads it into your memory, and your own processor writes the answer.
- Where the thinking happens
- on your laptop
- Works without internet
- yes, after setup
- Who can see your messages
- whoever can open your laptop
- Cost
- free
- How clever
- less: the model is small
How it works
What a model actually is
No magic and no internet. Here is what happens on your laptop when you press send.
- 1 / 4A file on your driveThe model is one .gguf file in your data folder.
- 2 / 4Loaded into memoryWhen you chat, Jan copies it into RAM, or into graphics memory if you have a GPU.
- 3 / 4Your chip does the mathsThe CPU or GPU predicts the next small piece of a word, again and again.
- 4 / 4Words appear one by oneThat is why answers stream in, and why a faster laptop writes faster.
What the “2B” means
A model is billions of numbers, called parameters, tuned while it was trained on huge amounts of text. “2B” means about 2 billion of them. More parameters give it more room to hold what it learned, so it is usually cleverer, but the file is bigger, needs more memory and writes more slowly.
Q4_K_M is the compressed version: each number is squeezed to about 4 bits. A much smaller file, a little less clever. The sweet spot for laptops.
- File to download (Q4_K_M)
- 1.28 GB
- Runs well on
- Almost any laptop with 8 GB. This is the one we use.
Rule of thumb: you need at least the file size in free memory, plus room for the chat. Jan’s Hub labels each model Fits, May be slow or Won’t fit for your laptop.
Check first
Will my laptop run it?
Pick what you have. Everything runs in your browser; nothing is sent anywhere. Details in chapter 02.
Hands-on
Four stops to your own AI
One page, every step in order, for your system and your drive. Tick each one off. When the last box is ticked, the AI is running on your laptop.
- 01Install JanFree, open source, about 10 minutes.
- 02Move its data off C:To a folder on D:, E:, F:, any drive that is not your system drive.
- 03Download one small modelQwen3.5-2B, 1.28 GB. Runs on 8 GB laptops.
- 04Chat with the internet offAirplane mode on. It still answers.
The one rule: move Jan’s data off the system drive right after setup, before the first download. A Windows reset usually wipes C:, and models fill it up fast. Only one drive? Keep the default or use an external SSD.
Start following alongWhy bother
Where a small local model helps
Where it does not help: recent news, exact sums, and anything medical or legal. Chapter 06 shows it failing on purpose.
Before you start
Questions people ask first
Is it really free?
Yes. Jan is free and open source, and the model in this workshop, Qwen3.5-2B, is free to download under the Apache-2.0 licence. No subscription, no account, no API key.
Does it work without internet?
Chatting does, once setup and the model download are done. The first-time setup and downloading models need the internet, and Jan checks for updates when it is online.
Will it run on my laptop?
You need 8 GB of memory (16 GB is better) and 10 GB of free disk. On Windows, a processor with AVX2, which most from 2013 onwards have. On a Mac, Apple Silicon, M1 or newer. Intel Macs cannot run local models in Jan.
Is it as good as ChatGPT?
No. A small model on a laptop is less clever, does not know recent news, and can be confidently wrong. It is good for everyday writing, explaining and practice, and it never sends your chats to an AI company.
Is it completely private?
Your chats with a local model are not sent to an AI company. But setup and downloads use the internet, Jan checks for updates by default, and chats are saved as plain-text files on your disk. Lock your laptop and turn on disk encryption.
Why move the data off the C: drive?
Models are big and fill the system drive, and a Windows reset usually wipes C:. Keeping Jan’s data in a folder on another drive, such as D:, E: or F:, keeps it safe and easy to find.
Do I need to know how to code?
No. Everything is clicking in an app. The follow-along page gives every step for Windows or Mac, with a checklist.
Go deeper
10 short chapters
The why behind every step: privacy, what to do when it breaks, and Jan vs LM Studio. A few minutes each, Windows and Mac side by side. Nothing to code.
Read them on GitHub