Jan

Что такое Jan?

Jan is an open-source desktop AI chat application for using language models on your own computer or through cloud providers. Its default workflow runs a local model, which can operate offline without an API key; users can also connect providers such as OpenAI, Anthropic, and Google, or configure compatible endpoints. The app combines model selection and conversation with workspaces for organizing chats, custom assistants, autonomous agents, and external tool connections. Jan is intended for people who want to choose where inference happens and configure their AI workflow in one desktop application.

Что умеет Jan?

Local model inference

Download and run models on the computer. The desktop quickstart says local inference is the default and works offline without a provider API key.

Cloud provider connections

Configure services including OpenAI, Anthropic, and Google with an API key. Jan also supports OpenAI- or Anthropic-compatible endpoints.

Model Hub

Browse or search available local models, download them, and select one for a chat. The Hub indicates whether a model fits the user's hardware.

Projects and assistants

Projects group conversations with shared instructions and files; custom Assistants provide a setup tailored to recurring work.

Agents and MCP connectors

Agents can take actions such as reading files, while MCP connections extend the app with external tools. Cowork is listed as a preview feature, with tasks handled in a sandbox or directly on a project.

Built-in web tools

The desktop quickstart describes built-in web search and fetch tools for asking about current events without additional setup.

Варианты использования

“Chat offline with a local model”

Download a model that fits the computer's hardware and use it for questions or drafting without sending inference requests to a cloud provider.

“Use a hosted frontier model”

Connect a cloud provider with an API key when a task calls for a provider-hosted model or local hardware is not suitable.

“Keep project conversations organized”

Use Projects to bring related chats together with shared instructions and files for ongoing work.

“Delegate file-based development tasks”

Use an agent to inspect a code project, answer questions, or propose changes. Jan Agent asks permission before file writes and shell commands; its documented quickstart requires a remote model provider.

Часто задаваемые вопросы

Which operating systems are supported?

The desktop app is listed for macOS, Windows, and Linux.

Can Jan run without an API key?

Yes. The desktop app defaults to local inference, which runs on the user's machine and does not require an API key. Cloud providers require their own API key.

Can I connect a model endpoint that is not listed?

Yes. Jan supports custom OpenAI- or Anthropic-compatible endpoints.

Is Jan Agent the same as the desktop app?

No. Jan Agent is distributed separately as a terminal-based agent. Its quickstart says it uses a remote model provider and can reuse providers configured in Jan Desktop.