jan

Jan drops a chat window on your desktop and points it at models running on your own machine instead of someone else's server. Pull a GGUF model from Hugging Face — Llama, Gemma, Qwen, GPT-oss — and it loads through llama.cpp or TensorRT-LLM, no account, no API key, no data leaving the box. Once it's running, Jan opens a local server at `localhost:1337` that speaks the OpenAI API format, so anything built to talk to ChatGPT can talk to Jan instead. It also wires into MCP tools for web search and code execution, and you can still bolt on OpenAI, Anthropic, Mistral, or Groq when a local model isn't enough for the task at hand. Menlo Research, the team behind it, ships builds for Windows, macOS, and Linux, plus a Microsoft Store listing and a Flathub package for anyone who'd rather not touch the command line. The catch is hardware: a 7B model wants 16GB of RAM to feel usable, and 13B pushes that to 32GB, so older laptops end up stuck with the smaller, less capable models. Jan isn't the only player here either (Ollama, LM Studio, and Open WebUI all cover similar ground), but its pitch is the full chat app bundled in, rather than a bare model server you point a separate interface at.