Jan

Jan is an open source, local-first alternative to ChatGPT that downloads and runs large language models directly on your own hardware. It's built for people who want to use AI assistants without sending every prompt to a third-party API, while still being able to plug in cloud models like GPT or Claude when needed. Jan ships as a desktop app for Windows, macOS, and Linux, built on Tauri and llama.cpp, and is distributed through jan.ai, GitHub Releases, the Microsoft Store, and Flathub.

Key features

• Local model execution: download and run models like Llama, Gemma, and Qwen from Hugging Face entirely on-device, with no internet connection required after the download completes.
• Cloud model integration: connect to OpenAI, Anthropic, Mistral, Groq, MiniMax, and other cloud providers from the same interface when a larger model is needed for a specific task.
• Custom assistants: configure specialized assistants for different tasks instead of relying on a single generic chat window for everything.
• OpenAI-compatible local API server: Jan exposes a server at localhost:1337 so other applications can call locally running models the same way they'd call OpenAI's API.
• Model Context Protocol support: MCP integration adds agentic, tool-using capabilities to local or cloud models running inside Jan.
• Privacy by default: when running local models, no data leaves the device, which the project frames as a core design principle rather than an optional setting.

Getting started

Jan is distributed through jan.ai, GitHub Releases, the Microsoft Store, and Flathub, with prebuilt binaries for Windows, macOS, and Linux, including deb, AppImage, and Arm64 packages. Building from source requires Node.js 20 or later, Yarn 4.5.3 or later, Make, and Rust for the Tauri shell, plus the Metal toolchain on Apple Silicon Macs. Running make dev from a clone of the repository handles dependency installation, builds, and launches the app in one step, and make build, make test, and make clean cover the rest of the development workflow.

System requirements

Jan recommends at least 8GB of RAM for 3B-parameter models, 16GB for 7B models, and 32GB for 13B models on macOS 13.6 or later; Windows 10 or later with GPU support for NVIDIA, AMD, or Intel Arc; and most Linux distributions, with GPU acceleration available where supported.

Contributing and support

Contributions follow the project's CONTRIBUTING.md guide, and troubleshooting starts with the documented desktop guides before falling back to the Discord #jan-help channel with logs and system specs attached. The project credits llama.cpp, Tauri, and Scalar as the open source foundations it builds on, and bug reports are tracked through GitHub Issues.

Tech stack

Jan is written in TypeScript on top of Tauri for the desktop shell and llama.cpp for local inference, and is licensed under Apache 2.0.