跳到正文
GitHub·· 7 天前AI 评分30

KabanGPT-2:可连接本地模型与 MCP 的浏览器 AI 聊天工具

xorgchain/KabanGPT-2

AI 导读

KabanGPT-2 是一款在浏览器中运行的本地模型 AI 聊天工具,无需云端即可连接模型对话。它支持接入 OpenAI、Anthropic 或任意 OpenAI 兼容 API 的第三方提供商,并可连接 MCP 服务器为模型扩展工具与数据,同时完整支持 HuggingFace 浏览、下载和使用模型。

正文

KabanGPT-2 is an AI chat for local models. You run it in your browser, connect a model, and talk to it right away. No cloud needed

What it can do

  • Chat with local models (runs fully on your machine)
  • Add your own third-party providers (OpenAI, Anthropic, or any custom API)
  • Connect MCP servers to give the model extra tools and data
  • Full HuggingFace support (browse models, download them, use them in chat)
  • Desktop app on the way: an Electron version you can download and install on your PC is coming soon

Desktop app

A full Electron version is coming soon. It will be a normal desktop app for Windows, macOS, and Linux. You will be able to download it and install it right on your PC, no browser needed. Check the releases page later to get it

Quick start

  1. Clone the repo:
    git clone
    
  2. Open index.html in your browser.
  3. Go to settings and pick your model or provider

That's it. Everything runs locally

Electron version:

  1. Open Releases in this repository
  2. Find the newest version, which is compatible to your OS
  3. Download it and open file

Providers

KabanGPT-2 works out of the box with local models. You can also add outside providers:

  1. Open the settings panel
  2. Click "Add provider"
  3. Enter the name, API key, and base URL
  4. Save and select it in the chat

Any provider with an OpenAI-compatible API works

MCP servers

You can hook up MCP servers to extend what the model can do:

  1. Open settings
  2. Find the MCP section
  3. Add the server address or config
  4. Enable the tools you want to use

The model will then be able to call those tools during the chat

HuggingFace

HuggingFace is fully supported. You can:

  • Search for models right inside the app
  • Download models to your machine
  • Start a chat with any model you download

System requirements (approximate, for Electron, using API)

Minimum Recommended
CPU 2 cores 4+ cores
RAM 8 GB 16 GB
SSD 1 GB 2 GB
Windows 11 64-bit 11 64-bit
MacOS 14 Sonoma 15 Sequoia (Apple Silicon)
Linux Ubuntu 22.04 LTS Ubuntu 24.04 LTS

P.S: If you are going to use local models, your PC must be more powerful (depends on a model)

License

MIT

来源:GitHub · github.com