Jan is a free, open-source desktop application that runs large language models entirely offline on personal computers. Users can chat with AI models locally, so your data stays on your machine — you don’t need an internet connection.
Run LLMs Offline with a Desktop App
Jan allows users to download and run various open-source LLMs directly on their machines, providing a chat interface for interaction. It includes a model hub for easy model downloads and supports popular models like Llama 3, Gemma 3, Mistral, Qwen, and GPT-oss, typically in GGUF format. The tool also offers its own models, such as Jan-v2 (multimodal) and Jan-v1 (with web search capabilities). Once models are downloaded, Jan can operate completely offline, keeping data on the user’s machine.
Privacy-First Users, Remote Workers, and Edge Computing
Jan’s designed for those who prioritize data privacy and sovereignty. This includes professionals like lawyers, doctors, and researchers who handle confidential information, as well as businesses with strict compliance requirements. Developers can use Jan as a local AI server — it’s got for projects, while AI enthusiasts can experiment with open-source models without subscription costs. Specific use cases involve running LLMs for privacy-sensitive tasks, customizing AI workflows, and developing AI applications without cloud dependencies.
llama.cpp Backend, GPU Acceleration, and HuggingFace Hub
Jan is cross-platform, available on Windows, macOS (supporting both Intel and Apple Silicon), and Linux. It provides an OpenAI-compatible local API server on localhost:1337, enabling direct integration with existing tools and workflows. The application also supports the Model Context Protocol (MCP) for advanced agentic features and can optionally connect to cloud AI providers like OpenAI, Anthropic, Mistral, Groq, and Cohere for hybrid workflows.
8GB RAM Minimum, GPU Recommended for Models Above 7B
Running LLMs locally requires adequate hardware. Jan’s minimum requirements include 8GB of RAM for 3-billion-parameter models, 16GB for 7B models, and 32GB for 13B models. A CPU from 2013 or newer with AVX2 support is generally sufficient, but a dedicated GPU (NVIDIA, AMD, or Intel Arc) is recommended for optimal performance. You’ll want to assess your system’s capabilities before attempting to run larger models.
Free and Open-Source; Only Hardware Costs Apply
Jan is completely free and open-source, operating under the Apache 2.0 license. There aren’t any subscription fees, usage limits, or premium tiers for its core functionalities. You’ll only pay for their hardware and electricity for running models locally, or API costs if they choose to connect to paid cloud AI services. An optional 100% offline voice dictation feature is available for a one-time payment of $24 on Mac.
Slow on CPU-Only Machines, Limited Model Selection vs Cloud
Jan has certain limitations around initial setup and hardware requirements:
Zero Data Leaving Your Machine
Jan excels in scenarios where data privacy and sovereignty are paramount. It offers a completely local AI experience, so sensitive information never leaves the user’s device. That’s why it’s particularly suitable for professionals handling confidential data and businesses with strict compliance needs. Its open-source nature provides users with complete control over model behavior, free from content filters or vendor lock-in — you’re in control, allowing for inspection, auditing, and modification of the AI stack. Furthermore, Jan’s a cost-effective solution by eliminating recurring API fees and usage limits associated with cloud services, offering unlimited responses once a model is downloaded. Its combination of a polished desktop user interface and an OpenAI-compatible local API server makes it a reliable option — it’s user-friendly for integrating local AI into various workflows. You can learn more and download the application at jan.ai.