HackerNews中文版

嗨，HN，在过去的几年里，我一直对一个问题感到着迷：我的硬盘是过去十年知识的宝藏（PDF、文档、笔记），我想利用现代人工智能来真正使用它。但每个解决方案都迫使我做出取舍：要么方便（将所有内容上传到云端），要么隐私（让我的文件在本地静默存放）。我厌倦了这种“捏鼻子”的妥协。随着强大的小型语言模型（SLMs）迅速崛起，以及设备端计算（苹果硅芯片、神经处理单元）终于迎头赶上，我相信隐私和智能不再是一个虚假的选择。因此，我（和我的团队）构建了 KnowledgeFocus（仅支持苹果硅芯片）。这是一个开源（Apache-2.0）知识引擎，使用 Tauri（Rust + Python + TS）构建，100% 本地优先。在 v0.6.4 中，它专注于一件事：解锁你的本地文件“宝藏”。 - 扫描与索引：它扫描你指定的本地文件夹（PDF、.md、.txt、.docx 等）。 - 自动标签：使用本地模型自动为文件打标签，以便你可以聚合和发现它们。 - 本地 RAG：你可以与所有本地文件“聊天”。它在设备上 100% 运行 RAG。没有任何数据（包括向量）会离开你的机器。网站（下载）：[https://github.com/huozhong-in/knowledge-focus](https://github.com/huozhong-in/knowledge-focus) 这只是我们“数据工作台”愿景的第一步。我还有很多关于“本地优先代理”、“数据聚合”（例如将你的云端 AI 聊天记录拉取到本地存储）以及为知识工作者构建真正的“第二大脑”的想法。我会在评论区中进一步阐述这些想法和我们的未来路线图。我期待所有的反馈——尤其是批评意见。谢谢，HN。

查看原文

Hi HN,For the past few years, I've been obsessed with a problem: my hard drive is a treasure chest of knowledge (PDFs, docs, notes) from the last decade, and I want to use modern AI to actually use it.But every solution forces a trade-off: either convenience (upload everything to the cloud) or privacy (let my files sit dormant locally).I got tired of this "hold your nose" compromise.With capable SLMs (Small Language Models) exploding and on-device compute (Apple Silicon, NPUs) finally catching up, I believe privacy and intelligence is no longer a false choice.So I (and my team) built KnowledgeFocus (Apple Silicon chips only)It's an open-source (Apache-2.0) knowledge engine built with Tauri (Rust + Python + TS), and it's 100% local-first.In v0.6.4, it focuses on one thing: unlocking your local file 'treasure chest'.- Scans & Indexes: It scans your designated local folders (PDFs, .md, .txt, .docx, etc.).- Auto-tagging: Uses a local model to auto-tag files, so you can aggregate and discover them.- Local RAG: You can 'chat' with all your local files. It runs RAG 100% on-device. No data (vectors included) ever leaves your machine.Website (Download): <a href="https://github.com/huozhong-in/knowledge-focus" rel="nofollow">https://github.com/huozhong-in/knowledge-focus</a>This is just the first step of our "Data Workbench" vision.I have a lot more thoughts on 'local-first agents', 'data aggregation' (like pulling in your cloud AI chat logs to store them locally), and building a real 'second brain' for knowledge workers.I'll be in the comments to expand on these ideas and our future roadmap.I'm here for all the feedback—especially the critical kind. Thanks, HN.

展示HN：我不想将我的私人文件上传到人工智能上