Jan
Your own ChatGPT as an app – load models with a click, everything stays offline.
Jan is a desktop app for local language models: pick a model in the built-in Hub, download it, start typing – no terminal and no account. Your chats never leave your computer. If you want, Jan also talks to cloud providers and offers a local, OpenAI-compatible server (localhost:1337) that other programs can use.
What's new
- This update fixes several security vulnerabilities, such as when opening links and when accessing the local API, and you should install it. Web apps on other computers must now be allowed under "Trusted Hosts" to talk to Jan. There's also a new version of the built-in llama.cpp engine.
What is Jan?
Jan is a desktop app that lets you run language models directly on your own computer. Language models (also called LLMs) are the AI programs behind chatbots like ChatGPT. With Jan, they run locally. That means the computation happens on your computer, not on someone else's server. Your chats never leave your machine.
How does it work?
You install Jan like any normal program. Downloads are available for Windows, macOS, and Linux. Then you open the built-in Hub. There you pick a model and download it. After that, you can start typing. You need neither a terminal nor an account.
Under the hood, Jan uses the llama.cpp engine. That's widely used software that runs language models on ordinary computers. But you don't have to operate it yourself.
What does Jan do especially well?
- Easy start: Choose a model, download it, chat. That's all it takes.
- Privacy: When you work locally, everything stays on your device.
- Lots of models: According to the README, you can load models like Llama, Gemma, Qwen, or GPT-oss from HuggingFace. HuggingFace is a platform where models are shared.
- GPU support: NVIDIA, AMD, and Apple graphics are used. That speeds up responses.
- Custom assistants: You can create specialized assistants for specific tasks.
- Cloud optional: If you want, you connect Jan to providers like OpenAI, Anthropic, Mistral, or Groq. Of course, your input then goes to that provider.
- Local server: Jan provides an OpenAI-compatible API at
localhost:1337. Other programs that work with the OpenAI API can use Jan's models this way. "localhost" means the server is only reachable from your own computer. - MCP: The Model Context Protocol (MCP) is an integration for agentic features, where the AI can use tools. The README only mentions the integration, with no details.
Where are the limits?
Local models need horsepower. For macOS, the README gives these rules of thumb: 8 GB RAM for 3B models, 16 GB for 7B, and 32 GB for 13B. "3B" means three billion parameters, which is roughly the size of the model. Small models are faster but usually less capable than large cloud models. On weak hardware, responses can be slow.
Models are also several gigabytes in size. So plan for disk space and some download time.
What do you need?
- A computer with Windows 10 or newer, macOS 13.6 or newer, or a common Linux distribution
- 8 GB RAM for small models, ideally 16 GB for 7–8B models
- Optionally, a suitable graphics card for more speed
- Internet for downloading the app and models
Jan is released under the Apache-2.0 license. The source code is open, and the app is free.
Ideas: What you can do with it
You paste a cover letter, a journal entry, or an internal document into the chat and have it polished or shortened. This runs locally, so the text stays on your computer. You get suggestions for improvement without sending sensitive content to a service.
You create an assistant for a fixed task, such as rephrasing emails politely or summarizing notes. You give it instructions once and then use it again and again. That saves you from explaining everything from scratch every time.
You download two or three models in the Hub, such as Llama, Gemma, or Qwen, and ask them all the same question. That shows you which one works best for German, ideas, or explanations. It's a good way to get a feel for AI models.
You start the server at localhost:1337 and enter that address in a program that supports the OpenAI API. The program then uses your local model instead of a cloud. That's great for tinkering projects or tests without ongoing costs.
You use a local model for everyday tasks. For tough tasks, you connect Jan to a cloud provider you have an account with. Both run in the same interface, and you just have to choose deliberately what goes where.
How to try it
Commands from the official guides – just copy them.
Linux
# Set up Flatpak (Ubuntu/Debian; on Fedora and Linux Mint it's already there):
sudo apt install -y flatpak
flatpak remote-add --user --if-not-exists flathub https://dl.flathub.org/repo/flathub.flatpakrepo
flatpak install -y --user flathub ai.jan.Jan
flatpak run ai.jan.Jan
# In Jan, click "Hub" on the left, choose a model (e.g. Gemma or Qwen, to fit your memory), "Download", then "Use"macOS
# Get the app: https://jan.ai (Download → macOS), drag it to "Applications" and launch it
# In Jan, click "Hub" on the left, choose a model, "Download", then "Use"Windows (PowerShell)
winget install Jan.Jan
# Open Jan from the Start menu, click "Hub" on the left, choose a model, "Download", then "Use"Tips
- Start with a small model (3B). It downloads quickly and shows you whether your computer keeps up. Try larger models afterward.
- Close other memory-hungry programs before you use a larger model. The model needs a lot of RAM.
- Create your own assistant for recurring tasks instead of typing the same instruction every time.
- If you don't want to hand over your data, stick with local models. With a cloud connection, your input goes to the provider in question.
- Check the documentation if you want to set up the local server or MCP. The README only gives the basics.
If something goes wrong
- The model responds very slowly or the app stutters: The model is probably too big for your RAM. Switch to a smaller model, such as 3B instead of 7–8B.
- The GPU doesn't seem to be used: Jan supports NVIDIA, AMD, and Apple graphics. Check whether your card is among them and whether your drivers are up to date. Details are in the installation guides on jan.ai.
- Other programs can't reach the local server: The address is localhost:1337. Check that the server is running in Jan and that the program uses exactly this address.
- Something isn't working at all: Copy the error message along with details about your system. Check the troubleshooting docs or ask in the help channel on the community Discord.
Ask the AI assistant
Questions about Jan, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.
The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.
Comments
What do you think? Already tried it? Write it here, no sign-up needed.
Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.