Jan is a desktop application for running large language models locally, with optional connections to cloud providers. It downloads open-weight models such as Llama, Gemma, Qwen, and GPT-oss from Hugging Face and runs them on the local machine, or connects to hosted models from OpenAI, Anthropic, Mistral, Groq, MiniMax, and others. Everything can run locally when you choose.
The app is built with Tauri on top of llama.cpp and ships for Windows, macOS, and Linux (deb, AppImage, Flathub, and Microsoft Store). It exposes an OpenAI-compatible local server at localhost:1337 so other applications can call models through Jan, and it integrates the Model Context Protocol for agentic tool use. Minimum memory guidance is 8 GB for 3B models, 16 GB for 7B, and 32 GB for 13B, with GPU acceleration for NVIDIA, AMD, and Intel Arc on Windows and GPU support on Linux.
Features
- Local models: downloads and runs Llama, Gemma, Qwen, GPT-oss, and other models from Hugging Face
- Cloud providers: connects to OpenAI, Anthropic, Mistral, Groq, MiniMax, and others
- Custom assistants: creates specialised assistants for particular tasks
- OpenAI-compatible API: local server at localhost:1337 for other applications
- MCP integration: Model Context Protocol support for agentic capabilities
- Cross-platform: Windows, macOS, and Linux builds, including Flathub and Microsoft Store listings
- GPU acceleration: NVIDIA, AMD, and Intel Arc support
