Jan is an open-source desktop application that provides a ChatGPT alternative running entirely offline on users' computers. Built with TypeScript and Tauri, it enables users to download and execute large language models locally with full control and privacy. The application supports multiple operating systems including Windows, macOS, and various Linux distributions, with installers available through the Microsoft Store, Flathub, and direct downloads from jan.ai.
The core functionality centers on local AI model execution, allowing users to run models like Llama, Gemma, Qwen, and GPT-oss downloaded from HuggingFace. Beyond local models, Jan provides cloud integration capabilities, enabling connections to GPT models via OpenAI, Claude models through Anthropic, and services from Mistral, Groq, and MiniMax. Users can create custom AI assistants tailored to specific tasks and access an OpenAI-compatible API running on localhost:1337 for integration with other applications. The platform includes Model Context Protocol integration for agentic capabilities, emphasizing privacy-first design where everything runs locally when desired.
The repository demonstrates active development and community engagement.
System requirements vary by platform, with macOS requiring version 13.6 or later and RAM recommendations ranging from 8GB for 3B models to 32GB for 13B models. Windows 10 and later are supported with GPU acceleration available for NVIDIA, AMD, and Intel Arc cards. Linux compatibility spans most distributions with GPU acceleration options. The project uses Node.js 20.0.0 or higher, Yarn 4.5.3 or higher, Make 3.81 or higher, and Rust for Tauri compilation.
Development is facilitated through Make targets including make dev for full development setup, make build for production builds, make test for testing and linting, and make clean for resetting the environment. The project maintains comprehensive documentation at jan.ai/docs, an API reference, and a changelog tracking updates. Community support occurs through Discord channels, with dedicated spaces for help and general discussion. The codebase is licensed under Apache 2.0 and acknowledges dependencies on Llama.cpp, Tauri, and Scalar. The repository overlaps with contributors from microsoft/vscode, prisma/prisma, and microsoft/typescript, indicating cross-project collaboration within the developer ecosystem.
