Jan AI

Your personal ChatGPT (Self-hosting) that works 100% offline.

Jan AI is an Open Source platform that turns your computer into a local artificial intelligence server. Its key feature is full compatibility with the OpenAI API, allowing you to use local models in any applications that previously required paid ChatGPT keys. Unlike cloud services, Jan stores all data and processes on your device.

Why choose Jan AI

• Complete privacy: Your prompts and data never leave your computer.
• Model flexibility: Install Llama 3, Mistral, or Gemma in one click, or connect to remote APIs (OpenAI, Claude).
• OpenAI compatibility: The built-in local server allows you to use Jan as a replacement for a paid API in other applications.
• Open Source: It is free and transparent.

Technical requirements: Works on Windows, macOS (including M1/M2/M3 processors), and Linux. For stable operation of local models (inference), your hardware resources are critical:

Key use cases

• Local RAGRAG Retrieval-augmented generation: the model answers from passages found in your documents. CustomGPT: a no-code RAG chatbot (Base Knowledge): Integration with your PDFs, documents, and databases through vector indexes (Chroma/FAISS).
• Development and testing: Using the local endpoint http://localhost:1337 to debug code without spending money on cloud provider APIs.
• Private assistant: Configuring custom roles (Coder, Analyst, Editor) while storing chat history on disk.

Detailed guide: how to install Jan AI, load a model, and enable the local API.

Does Jan need an internet connection to work?

Only to download models. After that, you can communicate with the AI in a completely autonomous mode.

Is this program paid?

No, Jan is completely free and distributed under the Apache 2.0 license.

Can Jan AI be used as a replacement for the paid OpenAI API?

Yes. Jan launches a local server that fully mimics the structure of OpenAI requests. You can simply change the base_url in your application to the local one, and it will start working with Llama, Mistral, or Gemma models.

What model formats does Jan support?

Jan works with most modern formats, including GGUF (optimal for CPUs), TensorRT, and other quantized versions that significantly speed up performance on consumer hardware.

Do I need to install additional drivers?

For Windows/Linux users, installing NVIDIA CUDA drivers or ROCm libraries (for AMD) is recommended to utilize the graphics card. On macOS (M1/M2/M3 processors), everything works out of the box thanks to Metal support.