Jan

What is Jan?

Jan is an open-source alternative to hosted chat assistants that runs models on the user's own computer. Jan Desktop supports macOS, Windows, and Linux, can download and run compatible local models through optimized inference engines, and stores local conversations and configuration on the device by default.

Users can also connect cloud models with their own provider keys, expose local models through an OpenAI-compatible Jan Server, use the Jan CLI, and extend assistants with Model Context Protocol tools. This makes Jan useful as both a private desktop assistant and a local model backend for developer tools.

How to Use Jan

1. Download Jan for the supported operating system and verify the installer source.
2. Select a local model that fits available RAM, storage, CPU, and GPU resources.
3. Download the model, start a chat, and confirm it can run offline.
4. Optionally add cloud provider API keys for models that cannot run locally.
5. Enable only trusted MCP servers and review every requested tool permission and action.
6. For development, start Jan Server or the CLI and connect an OpenAI-compatible client to the local endpoint.

Core Features

• Local model runtime: Runs compatible open models on personal hardware.
• Offline operation: Keeps prompts and responses on-device when using local models and tools.
• Model choice: Supports Jan models, compatible GGUF models, and selected cloud providers.
• Desktop applications: Provides macOS, Windows, and Linux clients.
• OpenAI-compatible server: Exposes local inference to compatible applications and developer tools.
• Jan CLI: Serves models and launches supported autonomous-agent workflows from a terminal.
• MCP integration: Connects models to external tools and data through a standard protocol.
• Open-source stack: Uses an Apache 2.0 license and can be inspected, modified, and built from source.

Use Cases

• Chat with an AI model locally without sending prompts to a hosted assistant.
• Analyze sensitive drafts or documents with an appropriate offline model.
• Compare local and cloud models in one interface.
• Provide a local OpenAI-compatible endpoint to coding or research tools.
• Build tool-using personal agents with carefully scoped MCP servers.

Pricing

Jan's core desktop software, local server, and CLI are free and open source. Local inference has no per-message fee, but uses the user's hardware and electricity. Cloud models are billed directly by their providers when API keys are connected.

Frequently Asked Questions

Is Jan always offline?

It can operate offline after the app and local models are downloaded. Cloud providers, remote MCP servers, web features, and updates require network access and have separate privacy terms.

Will every model run well on every computer?

No. Model size, quantization, context length, and hardware determine speed and memory use. Users should start with a model sized for their machine.

Are MCP tools automatically safe because Jan is local?

No. An MCP server can read data or perform actions according to its permissions. Install only trusted servers and require approval for sensitive tool calls.