Jan is an open-source ChatGPT alternative that runs large language models entirely on your own hardware, with no data sent to the cloud. It supports models like Llama, Mistral, Qwen, DeepSeek, and Gemma via the llama.cpp engine, and also allows optional connections to cloud providers via API keys. Built for privacy-conscious individuals, developers, and regulated businesses, it offers a local OpenAI-compatible API at localhost:1337 and MCP support for agentic use cases. The project has over 43,000 GitHub stars and 4 million downloads. Key features: - Fully offline local LLM inference via llama.cpp (GGUF models) - OpenAI-compatible local API at localhost:1337 for third-party app integration - Optional cloud model connections (OpenAI, Claude, Gemini, etc.) via API keys - LocalDocs: augment chats with local files, data never leaves the machine - Model Context Protocol (MCP) support for agentic capabilities - Cross-platform: macOS, Windows, Linux; AGPLv3 licensed
Free and open source. No paid tiers; enterprise support options available on request.