微信公众号聚合阅读器开源:WeRSS + Readeck + Obsidian 完整方案WeChat Public Account Aggregation Reader Opensource: Complete WeRSS + Readeck + Obsidian Solution


AI free token/AI免费积分

本文来源:https://mp.weixin.qq.com/s/mwWB6umILs2ryhnpq77DiA
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

为什么做

公众号里有很多值得长期保存的内容,但原生阅读流程有几个明显问题:订阅源分散、历史文章难整理、划线和批注无法稳定沉淀,最后真正有价值的内容还是要手工搬进知识库。

我最初考虑过 Folo,但私密订阅和 Obsidian 集成需要付费,而且它并不负责我真正需要的持久划线批注导出。

所以我把链路重新设计为:WeRSS 负责抓取,Readeck 负责阅读数据,自定义 Reader 负责交互,Obsidian 负责长期知识沉淀。

整套方案已经开源:

GitHub:WeChat RSS Reader[1]

三栏公众号阅读工作台

这套系统解决了什么

它不是单纯把公众号转成 RSS,而是补齐了一条完整的阅读工作流:

  1. WeRSS 按计划抓取公众号新文章和完整正文。
  2. 完整文章自动进入 Readeck,并按 URL 去重。
  3. 自定义 Reader 提供公众号、时间线、正文三栏布局。
  4. 阅读进度、收藏、价值评分、主题标签、高亮和批注持久保存。
  5. 收藏或含划线的文章自动进入 Obsidian。
  6. Obsidian 中的人工笔记不会被后续同步覆盖。

核心原则是:完整文章留在阅读库,只有人工筛选后的内容进入知识库。

这样既保留了完整信息,又不会让 Obsidian 变成一个塞满全文的垃圾场。

阅读体验

Reader 使用接近 Folo 的高密度三栏信息架构,顶部可以快速切换收件箱、未读、收藏、划线笔记、高价值和订阅;左侧按公众号过滤,中间浏览时间线,右侧直接阅读正文。

打开文章后会自动记录阅读进度,滚动到正文 80% 自动标记已读。右侧的 A+ / A- 可以调整字号,并在当前浏览器中记住选择。

阅读状态、价值评分和主题标签都落在 Readeck,而不是只存在于前端内存。 刷新页面或更换设备后仍然存在。

划线和批注

拖动选择文字后,批注框会根据选区和视口空间自动出现在右侧、左侧、上方或下方,不再固定挡在正文中间。

批注框自动避让选区

保存后,Reader 会自动回到对应划线,并在旁边显示一张包含摘录、笔记和来源的小卡片。

顶部“划线笔记”会把所有公众号的摘录集中起来。点击任意一条,就能重新定位到原文对应位置。

划线笔记、原位批注和字号控制

划线笔记还可以直接下载成 Markdown,内容包含:

  • 文章标题
  • 公众号名称
  • 发布时间
  • 划线原文
  • 我的批注
  • 原文链接

支持 File System Access API 的浏览器,还可以直接选择一个 Obsidian 文件夹写入汇总笔记。

Obsidian 联动:机器更新,人工内容不覆盖

后台同步器每 5 分钟运行一次,默认只处理两类文章:

  • 已收藏的文章
  • 包含高亮或批注的文章

文件名由发布日期、文章标题和 Readeck 短 ID 组成,因此同标题文章不会相互覆盖。

每篇笔记包含人工区和机器区。机器区可以更新原文、高亮和批注;人工区保存阅读状态、评分、主题、提升方向和“我的笔记”。

重复同步只刷新机器管理区,不会覆盖人工 frontmatter 和笔记。 我用真实文章连续同步两次,文件内容和修改时间都保持稳定。

Obsidian 中的原文、高亮、批注和人工笔记

完整公众号原文始终留在 Archive。真正值得进入 Knowledge 或 Output 的内容,再新建一篇自己的综合条目并引用原文,而不是直接移动整篇文章。

新文章如何更新

微信没有文章发布 Webhook,所以这套系统是安全近实时,而不是秒级实时。

当前默认策略是:

  • WeRSS 每小时第 17 分钟检查全部公众号。
  • reading-sync 每 5 分钟检查正文、阅读数据和 Obsidian。
  • Reader 右上角提供“检查新文章”按钮。

主动刷新包含 10 分钟冷却、single-flight、连续失败暂停和微信 200013 风控保护。浏览器不会拿到 WeRSS Access Key,也不能直接操作 Docker 或 SQLite。

实测样本通常在几分钟到一小时内进入 Reader。Mac 睡眠或关机期间不会抓取;如果要求全天运行,可以把相同 Compose 和数据迁移到 NAS 或常开服务器。

手机和外网怎么访问

本地所有服务都只绑定 127.0.0.1。外网通过 Cloudflare Tunnel 接入,并分别为 Reader 和 WeRSS 管理端配置两个 Cloudflare Access 应用。

两个入口使用不同的 AUD 和精确邮箱白名单。 未授权请求无法读取文章、API 或管理页面,浏览器中也没有硬编码 Token。

新设备第一次打开时完成一次邮箱验证码授权,之后在会话有效期内可以直接进入 Reader。

390px 移动端阅读界面

如何部署

当前版本主要在 Apple Silicon macOS 上完成验证,需要 Docker Desktop、Chrome,以及公众号或服务号的运营权限。

git clone https://github.com/summerchaserwwz/wechat-rss-reader.gitcd wechat-rss-reader./scripts/doctor.sh./scripts/init-secrets.shdocker compose pulldocker compose up -d./scripts/init-readeck.sh

启动后,本机入口分别是:

  • WeRSS:http://127.0.0.1:8001
  • Readeck:http://127.0.0.1:8002
  • 自定义 Reader:http://127.0.0.1:8082

完成 WeRSS 扫码、公众号添加和 Readeck API Token 配置后,运行:

./scripts/save-readeck-token.sh./scripts/install-reading-sync.sh

Cloudflare 和 Obsidian 的完整配置、备份恢复、安全边界都已经写在仓库文档中。

一些必须说明的限制

  • 必须拥有公众号或服务号运营权限,普通微信关注列表不能直接授权。
  • 微信风控出现 200013 时不要高频重试。
  • WeRSS 上游镜像当前在 Apple Silicon 上通过 Rosetta 运行,Readeck 为原生 ARM64。
  • 不要提交 .env、数据库、微信授权、API Token、Cloudflare AUD 或 Tunnel 凭据。
  • WeRSS 可能在容器日志中打印环境变量,不要公开原始日志。

这个项目最重要的是把订阅、阅读、判断、摘录和知识沉淀放进了一条可恢复、可迁移、可自己控制的链路。

如果你也想把公众号从信息流变成真正的个人知识来源,可以直接从仓库开始:

开源地址:summerchaserwwz/wechat-rss-reader[2]

往期推荐

个人开源项目 | PULSE/AI 全球 AI 中文情报雷达

你的Codex降智了吗:用Codex雷达监控重置、额度、降智情况

Skill 推荐 | 63k Star的Taste Skill 让你的设计少点AI味

Codex宠物社区Petdex:3700+宠物任你挑选

Obsidian自动图床配置:PicList + Cloudflare R2 完整教程

开源Skill | GoalLint:生成更明确、可验证的Goal,让Codex长程任务更稳定进行

个人开源工具:Obsidian 笔记同步飞书知识库插件

插件解锁?会话导出?试试Codex增强神器CodexPlusPlus

“

大家好,我是小柴,C9 计算机本硕,做过产品,现在是 AI 算法工程师,持续跟进和分享 AI 资讯、智能体、开源实测、经验分享、个人工具分享等内容,欢迎加好友进群交流,您的点赞、关注和分享是我持续创作的动力。

引用链接

[1]GitHub:WeChat RSS Reader: https://github.com/summerchaserwwz/wechat-rss-reader

[2]开源地址:summerchaserwwz/wechat-rss-reader: https://github.com/summerchaserwwz/wechat-rss-reader

Source: https://mp.weixin.qq.com/s/mwWB6umILs2ryhnpq77DiA. For copyright inquiries, please contact: 13951431913#139.com or submit a comment on the right.

Why do this?

Public accounts contain a lot of content worth long-term preservation, but the native reading workflow has several obvious issues: fragmented subscription sources, difficult to organize historical articles, and unreliable accumulation of highlights and annotations. Eventually, the truly valuable content still has to be manually moved into a knowledge base.

批注框自动避让选区

So I redesigned the pipeline to be: WeRSS for fetching, Readeck for storing reading data, a custom Reader for interaction, and Obsidian for long-term knowledge accumulation.

The entire solution has been open-sourced:

GitHub: WeChat RSS Reader [1]

三栏公众号阅读工作台

What does this system solve?

It’s not just about converting public accounts to RSS; it completes a full reading workflow: 10. WeRSS fetches new articles and full texts on a schedule. 11. Complete articles are automatically sent to Readeck and deduplicated by URL. 12. The custom Reader provides a tri-pane layout for public accounts, timelines, and article bodies. 13. Reading progress, favorites, value ratings, topic tags, highlights, and annotations are permanently stored. 14. Favored or highlighted articles are automatically entered into Obsidian. 15. Manual notes in Obsidian are not overwritten by subsequent synchronizations. 16. The core principle is to keep the full text in the reading repository and only move manually selected content to the knowledge base, preventing Obsidian from becoming a dump for full articles.

WeRSS fetches new articles and full content from WeChat public accounts as scheduled.

Complete articles are automatically stored in Readeck, with duplicates removed based on URL.

A custom reader provides a three-column layout for public accounts, timelines, and article text.

Reading progress, favorites, value ratings, tags, highlights, and annotations are persistently saved.

Favored or annotated articles are automatically transferred to Obsidian.

Manual notes in Obsidian are not overwritten by subsequent synchronizations.

The core principle is that full articles remain in the reading library, while only manually selected content enters the knowledge library.

Reading Experience

The Reader uses a high-density, Folo-like tri-pane architecture. The top bar allows quick switching between the inbox, unread, favorites, highlighted notes, high value, and subscriptions. The left side filters by public account, the middle shows the timeline, and the right side displays the article body.

Opening an article automatically records reading progress, and scrolling past 80% of the body marks it as read. The A+ / A- buttons adjust font size, with the selection remembered in the current browser.

Reading status, value ratings, and topic tags are stored in Readeck, not transient in the browser memory. Data persists after refreshing or switching devices.

Highlighting and Annotating

After selecting text, the annotation box intelligently positions itself on the right, left, top, or bottom of the visible area, rather than obstructing the center of the text.

After dragging to select text, the annotation box automatically appears on the right, left, top, or bottom depending on the selection and viewport space, no longer blocking the center of the text.

批注框自动避让选区

The top-level “Highlighted Notes” tab concentrates all clips from all public accounts. Clicking any entry relocates the reader to the exact position in the original text.

划线笔记、原位批注和字号控制

划线笔记、原位批注和字号控制

Annotated notes can also be downloaded directly as Markdown, including content from the public account, article title, URL, creation date, tags, ratings, and detailed annotations.

文章标题

Pub Name

Publish Time

Strikethrough Original

My Annotations

Original Link

Browsers that support the File System Access API can directly select an Obsidian folder to write the summary note.

Obsidian Integration: Machine updates, human content not overwritten

The background synchronizer runs every 5 minutes, defaulting to processing two types of articles:

Collected articles

Articles containing highlights or annotations

The filename is composed of the publication date, article title, and Readeck short ID, ensuring that articles with the same title do not overwrite each other.

Each note contains a human area and a machine area. The machine area can update the original text, highlights, and annotations; the human area saves reading status, rating, theme, improvement direction, and "my notes".

Repeated synchronization only refreshes the machine-managed area and will not overwrite the human frontmatter and notes. I performed two consecutive synchronizations on a real article, and the file content and modification time remained stable.

Obsidian 中的原文、高亮、批注和人工笔记

The complete original public account article always remains in the Archive. Only content truly worth entering the Knowledge or Output sections should prompt the creation of a new, standalone entry that references the original article, rather than moving the entire piece directly.

How new articles are updated

Since WeChat does not offer a publication webhook, this system operates on a safe, near-real-time basis rather than with millisecond precision.

The current default strategy is:

WeRSS checks all public accounts every hour at the 17-minute mark.

reading-sync checks the body text, reading data, and Obsidian every 5 minutes.

The Reader provides a "Check for New Articles" button in the top right corner.

Manual refreshes include a 10-minute cooldown, single-flight execution, pause upon consecutive failures, and protection against WeChat's 200013 rate-limiting. The browser does not receive the WeRSS Access Key and cannot directly manipulate Docker or SQLite.

In practice, sample articles generally enter the Reader within a few minutes to an hour. No fetching occurs while the Mac is asleep or turned off; if 24/7 monitoring is required, the same Compose and data can be migrated to a NAS or a constantly running server.

Access from mobile and external networks

All local services are bound only to 127.0.0.1. External access is granted via Cloudflare Tunnel, with distinct Cloudflare Access applications configured for both the Reader and the WeRSS management interface.

These two entry points utilize different audiences (AUD) and precise email whitelists. Unauthorized requests cannot read articles, access APIs, or view management pages, and no tokens are hardcoded into the browser.

The first time a new device connects, it must complete a single email verification step before being granted access within the session validity period.

390px 移动端阅读界面

Deployment requirements

The current version has been validated primarily on Apple Silicon macOS and requires Docker Desktop, Chrome, and operational privileges for a public or service account.

After startup, the local entry points are:

WeRSS: http://127.0.0.1:8001

Readeck: http://127.0.0.1:8002

Custom Reader: http://127.0.0.1:8082

After completing the WeRSS QR code scan, public account addition, and Readeck API Token configuration, run:

The full configuration, backup recovery, and security boundaries for Cloudflare and Obsidian are already documented in the repository.

Some necessary limitations

Must have operational permission for a public account or service account; ordinary WeChat follow lists cannot be directly authorized.

Do not retry frequently when WeChat risk control shows 200013.

WeRSS upstream image is currently running on Apple Silicon via Rosetta; Readeck is native ARM64.

Do not submit .env files, databases, WeChat authorizations, API Tokens, or Cloudflare AUD/Tunnel credentials.

WeRSS may print environmental variables in container logs; do not publish raw logs.

The most important part of this project is putting subscription, reading, judgment, extraction, and knowledge accumulation into a recoverable, migratory, and self-controllable pipeline.

If you also want to turn your public account from an information stream into a true personal knowledge source, you can start directly from the repository:

Open source address: summerchaserwwz/wechat-rss-reader [2]

Recommended readings

Personal open source project | PULSE/AI Global AI Chinese Intelligence Radar

Is your Codex downgrading? Use Codex Radar to monitor resets, quotas, and degradation

Skill recommendation | 63k star Taste Skill makes your design less AI-ish

Codex pet community Petdex: 3700+ pets for you to choose from

Obsidian automatic image host configuration: PicList + Cloudflare R2 complete tutorial

Open source Skill | GoalLint: Generate clearer, verifiable Goals to stabilize long-term Codex tasks

Personal open source tool: Obsidian note synchronization to FeiShu knowledge base plugin

Plugin unlock? Session export? Try the Codex enhancement artifact CodexPlusPlus

"

Hello everyone, I am Xiaocai, C9 undergraduate and master's in computer science, former product manager, current AI algorithm engineer. I continuously follow up and share AI news, agents, open source tests, experience sharing, and personal tool sharing. Welcome to add me as a friend and join the group for discussion. Your likes, follows, and shares are the driving force for my continuous creation.

[1] GitHub: WeChat RSS Reader: https://github.com/summerchaserwwz/wechat-rss-reader

[2] Open source address: summerchaserwwz/wechat-rss-reader: https://github.com/summerchaserwwz/wechat-rss-reader

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry