Jan
A desktop chat app that runs open LLMs on your own machine and also connects to cloud models through your own API keys.
Download open models from HuggingFace, chat locally, or plug in OpenAI, Anthropic and Groq keys.
Jan is a desktop app for chatting with large language models that run on your own computer. It suits people who want a ChatGPT-style assistant without sending prompts to a vendor, and who are happy to pick and download models themselves.
What it is
Jan downloads open models such as Llama, Gemma, Qwen and GPT-oss from HuggingFace and runs them through a bundled llama.cpp engine. It can also connect to OpenAI, Anthropic, Mistral, Groq and MiniMax, and it exposes an OpenAI-compatible server on localhost:1337 for other programs.
Who it's for
It fits professionals who draft, summarise or research with sensitive text and want it kept on the laptop. It is a poor fit if you need a team workspace with shared document libraries, because the README describes a single-user desktop app, not a multi-user server.
Key features
Local model downloads
Pull Llama, Gemma, Qwen or GPT-oss models from HuggingFace and run them on your own hardware.
Cloud keys alongside local
Connect OpenAI, Anthropic, Mistral, Groq or MiniMax in the same app as your local models.
Custom assistants
Create specialised assistants for repeated tasks instead of re-typing instructions.
Local API on port 1337
Other applications can call Jan through an OpenAI-compatible server at localhost:1337.
MCP agent tools
Model Context Protocol support lets models call external tools for agentic work.
GPU engine variants
The bundled llama.cpp engine ships CPU, Vulkan, Metal, CUDA and ROCm variants.
Up and running in 3 steps
Our honest verdict
What's great
- Installers exist for Windows, macOS, Linux (deb, AppImage, Flatpak) and the Microsoft Store, so you do not need a terminal to start.
- One interface covers both offline local models and paid cloud models, so you can compare them on the same prompt.
- The localhost:1337 OpenAI-compatible server lets scripts and other apps reuse the models you already downloaded.
- The README lists approximate memory needs per model size, so you can judge your machine before downloading anything.
Where it falls short
- Local models need real memory: the README suggests 8GB RAM for 3B models, 16GB for 7B and 32GB for 13B on macOS.
- Quality depends on the model you choose; small local models will not match the large cloud models Jan can also connect to.
- It is a single-user desktop app, so there is no shared team workspace, permissions or document library.
- Building from source needs Node.js, Yarn, Make and Rust, and CUDA builds need extra toolchains on Windows.
Pricing & license
Jan is free under the Apache 2.0 licence. The README and jan.ai show no paid Jan plan; if you connect cloud providers such as OpenAI or Anthropic, you pay those providers through your own API keys.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Does Jan work without an internet connection?
Once you have downloaded a model, the README says everything runs locally when you want it to. Downloading models and using cloud providers still need a connection.
How much RAM do I need?
The README gives macOS guidance of 8GB for 3B models, 16GB for 7B and 32GB for 13B. Windows needs version 10 or later, with GPU support for NVIDIA, AMD and Intel Arc.
Can other apps use the models in Jan?
Yes. Jan runs an OpenAI-compatible local server at localhost:1337 that other applications can call.
