What Jan does
Jan (jan.ai) is a desktop chat app for macOS, Windows and Linux that its docs describe as "an open-source replacement for Claude and ChatGPT". It is built by Menlo Research and licensed under Apache 2.0. On first launch Jan downloads a default model and you can start chatting with no account. A built-in Hub lists local models and shows whether each one fits your hardware; Jan's own models include Jan-v3-4B (general purpose), Jan-Code-4B (coding) and Jan-v2-VL (image understanding). Local models run through the bundled llama.cpp engine, or MLX on Apple Silicon Macs, and stay on your machine. You can also add cloud models from OpenAI, Anthropic, Google, Groq, Mistral, Hugging Face, OpenRouter and Azure OpenAI, or any OpenAI- or Anthropic-compatible endpoint, by pasting your own API key; those requests go to the provider, not just your computer.
Beyond chat, Jan has Projects (threads with shared instructions and files), custom Assistants, built-in web search and fetch tools, file upload, and MCP connectors with inline tool approval. The Local API Server exposes your local models as an OpenAI-compatible API at http://127.0.0.1:1337/v1 by default, so other apps can use them. Two newer pieces are previews on the nightly channel only: Cowork, which works through multi-step tasks by reading and writing files and running commands, and Jan Agent, a separate terminal coding agent that has no local inference engine and calls a cloud provider or an endpoint such as Jan Desktop's API server. The latest stable release is v0.8.8 (October 28, 2026), a security release that also moved to llama.cpp 0.7.0.
