Jan

Jan is a free, open-source desktop application, licensed under Apache 2.0, that describes itself as an open alternative to ChatGPT that runs 100% offline on your own computer. You download and run open models locally, so your conversations stay on your machine by default.

It downloads open models — Llama, Gemma, Qwen, and others — from Hugging Face and runs them locally through a clean chat interface, with everything processed on your device when you choose local models. For flexibility it can also connect out to cloud models (OpenAI, Anthropic, Mistral, Groq, and more) using your own API keys, so one app covers both private local use and hosted models.

Beyond chat, Jan runs a local OpenAI-compatible server (on localhost:1337) that your own tools can call, supports the Model Context Protocol (MCP) for agent-style capabilities, and lets you build custom assistants. It is available for Windows, macOS, and Linux, including the Microsoft Store and Flathub.

It fits privacy-conscious users who want a ChatGPT-style assistant that keeps data on their machine, developers who want a local OpenAI-compatible endpoint to build against, and anyone who wants one interface for both local and cloud models. It sits alongside tools like Ollama as one of the more approachable ways to run AI locally.

The honest limits, stated plainly. Running capable models locally needs a reasonably powerful computer, and larger models need substantial memory; performance depends on your hardware. Local models are private, but the cloud connectors send your prompts to those third-party providers under their own terms. As open-source software, review the Apache 2.0 licence for any commercial or work use.