开源 DigUp:本机按内容检索照片、PDF、音频与代码,跳到页码与时间戳
开发者盘点十月开源工具时提到 DigUp(ARahim3/DigUp):用 EmbeddingGemma 2 在设备端按内容检索本地文件,可直达 PDF 页、音频时间戳或代码行;帖文称约 5.7k star、MIT,支持 Mac/Windows/Linux。
在一份盘点十月开源开发工具的帖文中,开发者 @WuRichard_ 把 DigUp(仓库 ARahim3/DigUp)列为代表「按内容而非文件名检索本机资料」方向的项目。按帖文描述,DigUp 面向照片、PDF、音频、视频与代码等本地文件,检索完全在设备端完成,底层使用 EmbeddingGemma 2;命中后可直接跳到 PDF 页码、音频时间戳或代码行。帖文给出的社区信号是约 5.7k star、MIT 许可,并声称覆盖 Mac、Windows 与 Linux。
同一帖还并列提到 OpenAI 放出未发布模型相关的数学手稿集合,以及面向 AI 应用界面的设计系统 solid-glass。后两者分别偏研究材料与 UI 主题,与 DigUp 的本地知识检索主线不同;本稿只采纳 DigUp 这一条可核对的工具事实,不把列表里其他项目并成同一产品新闻。
对办公与知识管理场景,DigUp 的卖点是「材料散落在本机各处时,用语义检索代替记住路径」。是否满足合规(敏感文件索引范围、是否出网、索引落盘位置)仍取决于具体安装与配置,帖文未给出企业部署说明。采用前建议直接查看仓库 README 与权限提示,并以本机实测确认索引范围。
来源帖:https://x.com/WuRichard_/status/2114179704894665193 ;仓库名以帖文所写 ARahim3/DigUp 为准。
Three October open-source repos for AI tooling
3 open-source repos from October that show where dev tooling is going:
1️⃣ openai/math — OpenAI dropped 722 math manuscripts from an unreleased model at once: 4,000 candidate problems → 372 result families, ~3 hours of Pro-grade compute per result. 719 remain (3 withdrawn over a sign error). The catch: the model itself isn't public, so you can read the proofs but can't replay them.
2️⃣ ARahim3/DigUp — search your computer by content, not filename: photos, PDFs, audio, video, code. Fully on-device via EmbeddingGemma 2, jumps straight to the PDF page, audio timestamp, or code line. ~5.7k stars, MIT, Mac/Win/Linux.
3️⃣ deeplethe/solid-glass — a design system built for AI apps: one CSS sheet that themes your existing shadcn/ui components into calm glass, plus agent-workspace blocks (chat, tool steps, background tasks). Install: npx shadcn@latest add deeplethe/solid-glass/solid-glass
The pattern: AI is now building the tools that build AI. Which one lands in your stack this week?