内置一键安装仅覆盖 Gemma3、Llama 3.2、Phi-4、Qwen3，其他模型需自行下载 GGUF 文件；README 未给出源码构建与启动命令，需自行查阅项目页面；默认会联网检查更新，需在设置中手动关闭才能完全离线
README 未提供安装或运行命令，仅说明可从项目页面下载应用后直接打开使用 硬件要求按 README 说明
README 未说明具体显存要求；应用会自动探测并选用 Metal、CUDA、Vulkan 等后端，实际可跑性取决于所选模型大小与本地硬件。
需要 CUDA
用 90 年代 Clippy 界面在本地跑 GGUF 大模型的 Electron 桌面聊天应用
把 Windows 经典助手 Clippy 用 Electron 重制，并接入本地 llama 大模型，让回形针能真正对话回答问题。
核心能力
- 通过 llama.cpp 与 node-llama-cpp 本地运行 GGUF 模型
- 自动选择 Metal、CUDA、Vulkan 等最高效推理后端
- 一键安装 Gemma3、Llama 3.2、Phi-4、Qwen3 等模型
- 支持加载自备模型并自定义提示词与参数
- 完全离线运行，仅检查更新时联网且可关闭
容易踩的坑
内容由模型根据 README 生成，可能不完整或不准确，以官方文档为准。看原文 README
- v0.4.3最新正式版2025-05-08 · 2025-05-08 14:41 UTC
We still had issues with the incorrect native binaries making it into the build, breaking Clippy for Macs with Intel processors. This should finally be fixed.
- v0.4.22025-05-08 · 2025-05-08 00:37 UTC
查看更新说明
This small updated adds additional debugging features. Users without problems shouldn't see any changes.
- v0.4.12025-05-06 · 2025-05-06 22:18 UTC
查看更新说明
A small bug-fix: The Intel macOS release didn't include the correct binaries because the build machine is running on Apple Silicon. That's fixed now. Windows, Linux, and macOS builds are identical with
0.4.0. - v0.4.02025-05-06 · 2025-05-06 01:17 UTC
这个版本没有写更新说明。
- v0.3.02025-05-03 · 2025-05-03 20:48 UTC
这个版本没有写更新说明。
用户点评
登录用户的真实使用记录，可按版本和硬件筛选共 0 条
还没有人点评。第一个说出真实体验的人最有价值。
Clippy
Clippy let's you run a variety of large language models (LLMs) locally on your computer while sticking with a user interface of the 1990s. Through Llama.cpp, it supports models in the popular GGUF format, which is to say most publicly available models. It comes with one-click installation support for Google's Gemma3, Meta's Llama 3.2, Microsoft's Phi-4, and Qwen's Qwen3.
It's a love letter and homage to the late, great Clippy, the assistant from Microsoft Office 1997. The character was designed by illustrator Kevan Atteberry, who created more than 15 potential characters for Microsoft's Office Assistants. This app is not affiliated, approved, or supported by Microsoft. Consider it software art. If you don't like it, consider it software satire.
It is also meant to be a reference implementation of @electron/llm, hoping to help other developers of Electron apps make use of local language models.
Features
- Simple, familiar, and classic chat interface. Send messages to your models, get a response.
- Batteries included: No complicated setup. Just open the app and chat away. Thanks to llama.cpp and
node-llama-cpp, the app will automatically discover the most efficient way to run your models (Metal, CUDA, Vulkan, etc). - Custom models, prompts, and parameters: Load your own downloaded models and play with the settings.
- Offline, local, free: Everything runs on your computers. The only network request Clippy makes is to check for updates (which you can disable).
Non-Features
Countless little chat apps for local LLMs exist out there. Many of them are likely better - and that's okay. This project isn't trying to be your best chat bot. I'd like you to enjoy a weird mix of nostalgia for 1990s technology paired with one the most magical technologies we can run on our computers in 2025.
Downloading More Models
Clippy supports (thanks to Llama.cpp) most GGUF models. You can find GGUF models in plenty of online sources - I tend to go with models quantized by TheBloke or Unsloth.
Acknowledgements
Thanks to:
- I am so grateful to Microsoft - not only for everything they've done for Electron, but also for giving us one of the most iconic characters and designs of computing history.
- Kevan Atteberry for Clippy
- Jordan Scales (@jdan) for the Windows 98 design
- Pooya Parsa (@pi0) for being the (as far as I know) person to extract the length of each frame from the Clippy spritesheet.
- node-llama-cpp for squeezing llama.cpp into Node.js