Why it exists

Cloud chat assistants are astonishing — and they mean every prompt, document and mistake sits on someone else's servers. Jan's premise is that the local hardware most people already own can run surprisingly capable open models, and that the blocker is UX, not software. So Jan wraps llama.cpp-grade local inference in a polished desktop app with a model hub (one-click downloads of Llama, Mistral, Qwen, DeepSeek, Gemma…), a familiar chat interface, document context, and — for developers — a drop-in OpenAI-compatible local server. Nothing leaves the machine unless you plug in a cloud provider like OpenAI or Anthropic.

Main features

• Run any model, your way — pick from a built-in hub of open models or point Jan at online providers (OpenAI, Anthropic, Gemini, Mistral, DeepSeek…) per conversation; local and cloud models mix freely.
• 100% offline capable — models, chat history and document context all stay on disk; Jan works on a plane.
• OpenAI-compatible local API server — point any tool or script at localhost and swap cloud calls for local ones without code changes.
• Hardware-aware model recommendations — suggests what fits your RAM/GPU; GPU acceleration on Apple silicon and NVIDIA.
• Model control — adjust context length, temperature, system prompts per assistant; multiple assistants with different personas.
• Memory (coming) — context and preferences that carry over between sessions.
• Extensible — open plugin architecture and an active community; 7M+ downloads and 15k+ community members.

Who it's for

• Privacy-conscious users who want an AI assistant without uploading their notes, contracts or code.
• Developers who want a free local endpoint for prototyping LLM features before paying for tokens.
• Laptop owners with Apple silicon or a decent GPU who discover modern open models run fine locally.
• Tinkerers who want the weights, the logs and the settings — not a web wrapper.

How it compares

• Dot — smaller, macOS-only local RAG app built specifically for chatting with your documents; Jan is the broader Swiss-army assistant.
• Ollama + LM Studio — Ollama is the developer CLI workhorse and LM Studio the polished closed-source desktop; Jan splits the difference: open source and GUI-first.
• ChatGPT/Claude desktop apps — stronger frontier models, zero privacy; Jan's Ultimate trick is doing both in one app while defaulting to local.
• GPT4All — a similar open-source desktop; Jan's API server and model-hub ergonomics are generally considered a step ahead.

Notes

• Model quality depends on what your hardware can hold: 7–8B models run on most modern laptops, 70B-class needs a fat GPU or lots of unified memory.
• Actively developed under Menlo Research (the project moved from its original org); frequent releases and a fast-moving feature set — pin a version if you want stability.
• Cloud providers remain optional BYOK integrations; the privacy posture holds as long as you don't add them.
• License is Apache-2.0, so forking, embedding and commercial use are all fine.

Useful links

• Official site: https://www.jan.ai/
• Source code: https://github.com/janhq/jan
• Documentation: https://jan.ai/docs
• Community (Discord): https://discord.gg/FTk2MvZwJH