Jan

Jan is an open-source desktop app that provides a local interface for interacting with large language models, allowing users to run AI on their own hardware or connect to external cloud providers. The software is available as a downloadable application for Windows, macOS, and Linux, supporting various installation formats including .exe, .dmg, .deb, and AppImage.

Users utilize Jan to maintain privacy by keeping conversations off third-party servers. It serves as both a chat interface and a local server, enabling other applications to interact with models through a standardized API. The application supports various hardware configurations, including GPU acceleration for NVIDIA, AMD, and Intel Arc, ensuring that model inference is optimized based on the available system resources.

Key features

• Local LLM execution for models such as Llama, Gemma, and Qwen
• Integration with HuggingFace for model downloads
• Cloud connectivity for OpenAI, Anthropic, Mistral, and Groq
• Creation of specialized custom AI assistants
• OpenAI-compatible local API server on port 1337
• Model Context Protocol integration for agentic capabilities
• Local-first architecture for data privacy

Jan is built using Tauri and Node.js, requiring Rust for those who choose to build from source. It is designed for users who need full control over their AI environment, ranging from casual users to developers who require a local API for other software. The software includes a build system via Make targets for development, production builds, and testing. System requirements vary by model size, with RAM needs increasing from 8GB for 3B models to 16GB for 7B models and 32GB for 13B models on macOS.

This software is a privacy-focused hub for both local and remote AI models, bridging the gap between cloud-based AI services and self-hosted hardware.