Jan

Jan

Jan is an open-source desktop app that works as a ChatGPT replacement running on your own computer. On first launch Jan downloads a default model, so you can chat offline with no account or API key, and its Hub lists other open models with a note on whether each fits your hardware. The same app can also connect to OpenAI, Anthropic, Google and other cloud providers with your own key, so local and frontier models sit side by side. The desktop app adds Projects, custom Assistants, web search, MCP connectors and a local API server.

Production credibility: Jan is built in public by Menlo Research and released under the Apache 2.0 licence. Its GitHub repository had more than 45,000 stars on 8 November 2026, and the homepage reports millions of downloads. The team also publishes its own small open models on Hugging Face, such as Jan-v3-4B for general use and Jan-Code-4B for coding, which ship as options inside the app. Alongside the desktop app, Jan Agent is a separate command-line coding agent, currently a nightly preview, that can point at a cloud provider or at Jan's own local API server. The desktop app is free; cloud models are paid for directly to their providers through your own API keys.

Key Features

• Runs open models offline with no account or API key
• Hub that shows whether each model fits your hardware
• Connects to OpenAI, Anthropic, Google and any OpenAI-compatible endpoint
• Projects, custom Assistants, file upload and built-in web search
• MCP connectors and a local API server for other apps
• Apache 2.0 open-source code for Mac, Windows and Linux

Ideal Use Case

People who want a ChatGPT-style app that keeps conversations on their own machine, such as journalists, lawyers or developers handling sensitive text, and anyone who wants one interface for both local models and paid cloud models. Developers can also run Jan's local API server as an OpenAI-compatible endpoint for their own tools.

How Jan differentiates

Jan's case against Msty and the LM Studio app is its licence: the whole desktop app is open source under Apache 2.0, so it can be audited, forked and self-built. Compared with Ollama, which is mainly a model server driven from the command line, it is a finished chat app aimed at non-developers. AnythingLLM leans further into document retrieval and team workspaces, while this app stays closer to a private personal assistant.

FAQ

Q: What is Jan? A: Jan is an open-source desktop app for Mac, Windows and Linux that works as a ChatGPT replacement. It runs AI models locally on your computer and can also connect to cloud providers.

Q: Is Jan free? A: Yes. Jan is free and open source under the Apache 2.0 licence. If you connect cloud models such as OpenAI or Anthropic, you pay those providers directly through your own API key.

Q: Does Jan work offline? A: Yes. Jan downloads a default model on first launch and runs local models entirely on your machine, with no internet connection or API key required.

Q: Who makes Jan? A: Jan is built in public by Menlo Research, with its code on GitHub under the janhq organisation and its own small models published on Hugging Face.

Q: Can Jan act as a local API server? A: Yes. Jan includes a local API server, so other apps and tools, including Jan Agent, can call the models it runs through an OpenAI-compatible endpoint.

tl;dr

Jan is a free, Apache 2.0 open-source ChatGPT replacement for Mac, Windows and Linux built by Menlo Research. It runs models offline, connects to cloud providers with your own keys, and serves a local OpenAI-compatible API.