jan README

The project's README from github.com/janhq/jan. As one markdown file: README.md.

Jan - Open-source ChatGPT replacement

github jan banner

English · 中文 · 日本語

GitHub commit activity Github Last Commit Github Contributors GitHub closed issues Discord

Getting Started - Community - Changelog - Bug reports

Jan is bringing the best of open-source AI in an easy-to-use product. Download and run LLMs with full control and privacy.

Installation

Get it from Microsoft Store Get it on Flathub

The easiest way to get started is by downloading one of the following versions for your respective operating system:

Platform Download
Windows jan.exe
macOS jan.dmg
Linux (deb) jan.deb
Linux (AppImage) jan.AppImage
Linux (Arm64) How-to

Download from jan.ai or GitHub Releases.

Features

Build from Source

For those who enjoy the scenic route:

Prerequisites

Run with Make

git clone https://github.com/janhq/jan
cd jan
make dev

This handles everything: installs dependencies, builds core components, and launches the app.

Available make targets:

Manual Commands

yarn install
yarn build
yarn dev

Building on Windows

Run make dev from Git Bash (installed with Git for Windows) — make dispatches its recipes through sh, so a plain cmd.exe won't work.

You do not need a "Native Tools Command Prompt for VS 2022". The bundled llama.cpp engine builds with Ninja + clang-cl, and clang-cl locates the MSVC toolchain and Windows SDK on its own. What has to be installed (and on PATH for ninja/clang-cl/cmake):

Engine variants are picked with JAN_ENGINE_VARIANT (tokens: cpu, vulkan, metal, cuda12, cuda13, hip/rocm, joined by -), e.g.:

make dev JAN_ENGINE_VARIANT=cuda13

"nvcc fatal : Could not open output file ...fattn-...cu.obj.d" during tauri-plugin-llamacpp(build) means the build path crossed Windows' 260-character MAX_PATH limit — nvcc does not honor the long-path opt-in. The build script now detects this and automatically relocates the llama.cpp build tree to a short directory under %LOCALAPPDATA%\jan-engine. If you hit path-length errors anyway, set JAN_ENGINE_BUILD_DIR to a short path (e.g. C:\jb) or move the checkout closer to the drive root.

System Requirements

Minimum specs for a decent experience:

For detailed compatibility, check our installation guides.

Troubleshooting

If things go sideways:

  1. Check our troubleshooting docs
  2. Copy your error logs and system specs
  3. Ask for help in our Discord #🆘|jan-help channel

Contributing

Contributions welcome. See CONTRIBUTING.md for the full spiel.

Note: Please sign your commits so we can verify your contributions.

Contact

License

Apache 2.0 - Because sharing is caring.

Acknowledgements

Built on the shoulders of giants:

jan (janhq/jan)

Searching Database…

45K ·GitHub
这是什么?Jan 是一款开源、可自托管的 ChatGPT 替代品,可在你自己的电脑上 100% 离线运行,让你在本地下载并运行大语言模型,完全自主可控,也更保护隐私。
这是桌面应用,还是要在终端里运行的程序?这是一款使用 Tauri(Rust 后端)构建的桌面应用,提供 Windows、macOS 和 Linux 的安装包(deb/AppImage/Arm64)。
我该怎样把它运行起来?可以从 jan.ai 或 GitHub Releases 下载适用于你操作系统的预编译安装包;也可以使用 Node.js、Yarn、Make 和 Rust,通过 'make dev' 从源代码构建。
在 Windows 上构建时有什么需要注意的坑吗?必须在 Git Bash 中运行 'make dev',不能用 cmd.exe,因为 make 会通过 sh 来执行各个构建命令;另外,构建路径过长时可能触及 Windows 的 260 字符 MAXPATH 限制,构建脚本会尝试自动绕过这一限制。
要流畅运行较大的模型,需要什么硬件?最低配置随模型大小而提升:3B 模型需要 8GB 内存,7B 需要 16GB,13B 需要 32GB;Windows(NVIDIA/AMD/Intel Arc)和 Linux 支持 GPU 加速。
我的数据会离开我的电脑吗?不会。默认情况下,一切都在本地运行;只有你明确进行设置后,才会使用云端连接(如 OpenAI、Anthropic 等)。
Jan 也能连接云端 AI 服务吗,还是只能本地使用?两者都可以:它既能完全离线运行本地模型,也支持云端集成,可通过 OpenAI 连接 GPT,通过 Anthropic 连接 Claude,还支持 Mistral、Groq 和 MiniMax。
Jan 是否提供 API 供其他应用调用?提供。它会在本地服务器 localhost:1337 上运行一个兼容 OpenAI 的 API,供其他应用与之通信。
Jan 除了聊天,还能使用工具或智能体吗?可以。它集成了模型上下文协议(MCP),具备智能体能力,还允许你创建自定义助手,用于处理专门的任务。
本地模型从哪里来?你可以直接从 HuggingFace 下载并运行 Llama、Gemma、Qwen、GPT-oss 等大语言模型。
Creator
janhq
Size
1.5 GB
Stars
45K
Age
3y
Commits · 90 Days