开发者Christian发布了一款名为Hob的开源桌面应用程序，该软件采用MIT许可证，支持Mac、Windows和Linux三大操作系统1。Hob定位为Claude Cowork和ChatGPT桌面版的本地替代方案，其核心设计理念是将所有文件处理及聊天数据保留在用户设备本地，确保数据不上传至云端1。该应用利用Gemma 4和gpt-oss等开源权重模型在本地运行，并基于Electron桥接llama.cpp构建技术栈1。
为了提供完整的离线工作能力，Hob集成了多项高级功能，包括自动匹配硬件性能的最佳模型、沙盒化Python代码执行以及通过SearXNG支持的Web搜索1。此外，该应用还具备MCP连接器支持、文件系统CRUD操作能力，并在图像处理方面使用Qwen3-VL模型，语音听写则依赖Whisper模型1。针对现有本地AI应用在解析大文件或图像时的局限性，Hob试图通过避免依赖托管模型或账户体系来解决这些问题，从而强调全离线环境下的隐私保护1。
Developer Christian has launched Hob, a free, MIT-licensed open-source desktop application designed to serve as a local alternative to the desktop apps of Claude and ChatGPT 1. The software operates on Mac, Windows, and Linux systems, leveraging open-weight models such as Gemma 4 and gpt-oss to process files and images directly on the user's device 1. This architecture ensures that all chat interactions and file data remain within the local system, avoiding uploads to the cloud and eliminating dependencies on hosted models or account systems 1.
Hob distinguishes itself by addressing limitations in existing local AI applications regarding large file handling and image parsing 1. Built on an Electron stack bridging llama.cpp, the app utilizes the Qwen3-VL model for image processing and Whisper for voice dictation 1. Key features include automatic selection of the best model suited to the user's hardware, sandboxed Python code execution, offline web search via SearXNG integration, MCP connector support, and full filesystem CRUD operations 1. The application emphasizes privacy protection and the capability to function entirely offline 1.
评论
还没有评论，欢迎留下第一条。