MyLibrary-Plus：一个你真的会回头翻的本地论文库
Published:
问题：库变成了墓地
文献管理器擅长把论文存进去，却不擅长让你再回来看。论文流入的速度永远快过注意力流出的速度，于是库慢慢变成墓地：你记得的往往是某张图，而不是某个文件名，而工具里没有任何一处是围绕「重新遇到自己存过的东西」设计的。
做法：把首页做成图优先的时间线
MyLibrary-Plus 站在文献管理器上再加一层，专门优化回访。首页是按加入时间倒序的卡片时间线，卡片正面就是论文的图，往下滑像刷一条自己曾经在意过的信息流。
主要能力：
- 图优先时间线：每张卡片带第 1 页和 1–3 张图，可全屏灯箱；按标签、按「有没有讲义」筛选，也可搜索
- 批注集中在一处：PDF 和 Markdown 讲义里的选中文字都能批注，同一个侧栏同时放两边，支持标签、时间戳、编辑和回复
- 讲义模式：左边 PDF、右边 Markdown 讲义，支持 KaTeX 公式、内嵌图、提示框、
[[wikilinks]]和反链；一篇论文可以挂多份讲义，讲义里还能嵌入交互式 HTML/JS 组件 - 夜览模式：跟随系统或手动切换，首屏绘制前生效，不闪白
- 离线 Zotero 导入：直接读 Zotero 的 SQLite，复用本地已有 PDF，付费论文的图也不会丢
- 面向 agent 的接口：
GET /api/papers/{id}/annotations/context把论文、批注和指令打包成 JSON，agent 可以按role: "assistant"回进批注线程 - 全部本地：SQLite 加一个
data/目录，只绑127.0.0.1，不需要账号，不上云，没有埋点
项目 fork 自 liusida/MyLibrary，在其之上加了整层阅读与批注，以 MIT 协议开源。
链接
The problem: a library that becomes a graveyard
Reference managers are good at filing papers and bad at bringing you back to them. Papers flow in faster than attention flows out, so the library slowly turns into a graveyard: you remember a figure, not a filename, and nothing in the tool is designed around running into your own saved work again.
The approach: a figure-first timeline as the home page
MyLibrary-Plus sits one layer above the reference manager and optimises for the revisit. The home page is a reverse-chronological timeline of paper cards, and each card shows the paper’s figures, so scrolling feels like a feed of things you once cared about.
Highlights:
- Figure-first timeline: every card carries page 1 plus figures 1–3 in a swipeable carousel with a full-screen lightbox; filter by tag, by “has study notes”, or search
- Annotations in one place: select text in the PDF or in your Markdown study notes and annotate it; one sidebar holds both, with tags, timestamps, edits and replies
- Study view: PDF on the left, Markdown notes on the right, with KaTeX math, embedded figures, callouts,
[[wikilinks]]and backlinks; a paper can carry several notes, and a note can embed interactive HTML/JS components - Night reading: system / light / dark toggle, applied before first paint
- Offline Zotero import: reads Zotero’s SQLite directly and reuses the PDFs already on disk, so paywalled papers keep their figures
- Agent-ready annotations:
GET /api/papers/{id}/annotations/contextreturns the paper, its annotations and instructions as JSON; an agent can reply into the thread asrole: "assistant" - Everything stays local: SQLite plus a
data/directory, binds to127.0.0.1, no account, no cloud, no telemetry
It is a fork of liusida/MyLibrary with a full reading and annotation layer added on top, released under MIT.