jan: Run AI Models Locally or Connect to Cloud Models

Overview

Jan is a desktop application for using AI models from one place. It can download and run language models locally, giving users control over their models and data, or connect to cloud model providers when they need those services.

It is a fit for individuals and developers who want a desktop chat interface with local inference, while retaining options for hosted models and integrations with other applications.

Key Features

• Download and run local language models, including models from Hugging Face.
• Connect to cloud providers including OpenAI, Anthropic, Mistral, Groq, and MiniMax.
• Create custom assistants for specialized tasks.
• Provide an OpenAI-compatible local API at localhost:1337.
• Integrate with the Model Context Protocol for agentic capabilities.
• Run as a desktop app on Windows, macOS, and Linux.

Use Cases

• Privacy-conscious users can run supported models locally when they prefer not to send prompts to a cloud service.
• People exploring local models can download and use models through a desktop interface rather than setting up a separate inference workflow.
• Developers can connect applications that support the OpenAI API format to Jan's local server.
• Users who need both local and hosted models can switch between local inference and supported cloud providers.
• Teams or individuals can configure custom assistants for recurring tasks.

Project Facts

• Language: Rust
• License: NOASSERTION
• Stars: 44.8k
• Forks: 3.1k
• Topics: chatgpt, gpt, llamacpp, llm, localai, open-source, self-hosted, tauri
• Archived: false

Getting Started

Download an installer for your platform from jan.ai or GitHub Releases. For build instructions and platform-specific setup, see the README.

Alternatives

• cherry-studio: Cherry Studio focuses on chatting with local and cloud models, model comparison, assistants, and documents rather than downloading and serving local models.
• deepchat: DeepChat emphasizes agent workflows, tools, and resumable sessions, while Jan centers on running local models and providing an OpenAI-compatible API.

Considerations

Local inference depends on the selected model and the computer's available memory and hardware. The README gives example memory guidance for macOS model sizes and notes GPU support for Windows, but performance will vary by device. Building from source requires Node.js, Yarn, Make, and Rust, with additional platform-specific tooling; consult the README before attempting a source build. Cloud integrations require using the respective provider rather than running those models locally.