Our verdict: Jan is one of the friendliest ways to chat with open models entirely on your own computer. It downloads models from inside the app, runs them on your GPU or CPU, and keeps all data local. You need a reasonably modern PC: an AVX2 CPU, 8 GB RAM at minimum and a GPU with 6 GB VRAM for good speed.
About Jan
Jan is an open-source desktop app under the Apache License 2.0, developed by Menlo Research. It runs open models such as Llama, Gemma, Mistral, Qwen and DeepSeek locally through a bundled llama.cpp engine, with GPU acceleration on NVIDIA, AMD and Intel Arc cards. Models, chats and settings are stored in a local data folder, and nothing is sent to the cloud.
If you want, you can also add cloud providers with your own API keys and use them in the same chat window. Jan includes a local API server compatible with the OpenAI format, so other programs on your PC can use your local models, and it supports MCP servers for tools.
What it does well
- Local models. Downloads and runs open LLMs on your own hardware, offline.
- GPU acceleration. Uses NVIDIA (CUDA), AMD and Intel Arc (Vulkan) graphics cards.
- Local API server. Serves your models through an OpenAI-compatible API on your PC.
- Cloud providers. Optionally adds online models with your own API keys.
- MCP support. Connects Model Context Protocol servers for tools.
Who it is for
Good fit
Privacy-minded users who want a ChatGPT-like assistant that runs offline on their own PC.
Look elsewhere
Older PCs without AVX2 or with little RAM, and people who need top cloud-model quality.
Pros and cons
Pros
- Free and open source
- Runs offline with data kept on your PC
- Simple in-app model downloads
- OpenAI-compatible local API
Cons
- Needs an AVX2 CPU and plenty of RAM
- Local models are weaker than top cloud models
- Model files take a lot of disk space
How to install Jan
Run the x64 setup file; Windows on ARM PCs need the ARM64 installer from the same page. Jan needs a CPU with AVX2, at least 8 GB RAM (16 GB recommended) and 10 GB of free space. Turn on your GPU in Settings, Hardware, and download a model from the Hub; model files are several gigabytes each.
Questions about Jan
Is Jan free?
Yes. Jan is open source under the Apache License 2.0 and free to use. Cloud models you add with your own API key are billed by that provider.
Does Jan work offline?
Yes. Once a model is downloaded, Jan runs it locally, and your chats and settings stay in a data folder on your PC.
