Don't just consume. Flow!
别只是消费信息,让知识流动起来。
AI 资讯推送、公众号文章聚合与知识创作工作台
AI news delivery, curated RSS & WeChat article feeds, and an AI-native knowledge workflow
关键词 / Keywords:AI 资讯推送 · AI news feed · 公众号 AI 文章聚合 · WeChat AI article feeds · 优质信息源 · curated sources · RSS 阅读器 · RSS reader · AI 知识库 · AI knowledge base · AI 写作 · AI writing assistant · 内容创作 · content creation
参考个人网站项目介绍的三栏阅读体验:左侧管理信息源,中间浏览文章,右侧查看 AI 总结与完整正文。
Manage sources on the left, browse articles in the center, and read AI summaries and full text on the right.
把读过的文章拆解为观点、数据、金句和故事卡片,并保留标签、来源与原文上下文。
Turn articles into reusable viewpoint, data, quote, and story cards while keeping tags, sources, and original context.
AtomFlow 是面向内容创作者、产品经理和 AI 从业者的知识工作台。它把分散在 RSS、公众号、科技媒体、博客、YouTube 和播客里的优质内容汇入「今日推送」,再把阅读、收藏、知识沉淀和 AI 辅助创作连接成一条可追溯的工作流。
我们的目标:让每一篇看过的知识,都成为可复用的创作资产。
| 层级 | AtomFlow 做什么 | 你得到什么 |
|---|---|---|
| 输入 | 聚合 AI 资讯、公众号文章、国内外媒体、RSS/RSSHub、视频与播客来源 | 一个更干净、更高质量的信息入口 |
| 沉淀 | 全文阅读、保存原文,并将内容拆解为观点、数据、金句和故事卡片 | 可检索、可回溯、可复用的个人知识库 |
| 输出 | 在魔法写作无限画布中组合文章、卡片、笔记、文件、图片和写作 Agent | 带来源上下文、由你掌控的创作结果 |
- AI 资讯推送:持续聚合中英文 AI、科技、产品和商业动态,减少跨平台刷新带来的信息焦虑。
- 公众号文章聚合:通过可用的 RSS/RSSHub 来源统一阅读公众号与其他内容渠道。
- 优质信息源发现:内置精选来源,并支持自定义 RSS、RSSHub、YouTube 与播客订阅。
- 沉浸式全文阅读:使用 Readability + JSDOM 提取正文,保留原文地址、来源和发布时间。
- 原子知识库:把文章沉淀为观点、数据、金句、故事四类卡片,并保留引用上下文。
- 魔法写作画布:在 tldraw 无限画布中显式连接素材与写作 Agent,决定 AI 可以读取哪些上下文。
- 播客解读:把可用的音频内容纳入同一套发现、理解和知识复用体验。
- 可自托管:React + Express + PostgreSQL 一体化部署,AI 模型使用 OpenAI-compatible 接口。
- 需要持续追踪 AI 行业动态、公众号文章与海外科技内容的从业者
- 想把日常阅读转成选题、素材和文章的内容创作者
- 需要保留信息来源和推理上下文的产品经理、研究者与知识工作者
- 希望掌控数据、模型供应商和部署环境的个人或小团队
AtomFlow is a knowledge workspace for creators, product managers, and AI practitioners. It brings high-quality content from RSS, WeChat Official Accounts, technology media, blogs, YouTube, and podcasts into one daily feed—then connects reading, saving, knowledge extraction, and AI-assisted creation in a traceable workflow.
Our goal: turn everything you read into a reusable creative asset.
| Layer | What AtomFlow does | What you get |
|---|---|---|
| Input | Aggregates AI news, WeChat articles, global media, RSS/RSSHub, video, and podcast sources | One focused, higher-signal information feed |
| Knowledge | Saves full articles and extracts viewpoints, data, quotes, and stories | A searchable, source-linked personal knowledge base |
| Creation | Connects articles, cards, notes, files, images, and writing agents on an infinite canvas | Source-aware drafts with human-controlled context |
- AI news delivery — Follow Chinese and English AI, technology, product, and business updates in one feed.
- WeChat article feeds — Read supported WeChat Official Account sources through available RSS/RSSHub routes.
- Curated sources — Start with selected feeds or add custom RSS, RSSHub, YouTube, and podcast subscriptions.
- Full-text reading — Extract readable article content while preserving its original URL, source, and publication time.
- Atomic knowledge base — Turn articles into reusable viewpoint, data, quote, and story cards with source context.
- Magic Writing Canvas — Explicitly connect source material to writing agents and control exactly what context AI can access.
- Podcast insights — Bring supported audio content into the same discovery and knowledge workflow.
- Self-hostable stack — Run the React, Express, and PostgreSQL application with your preferred OpenAI-compatible model provider.
AI 资讯 / 公众号 / RSS / YouTube / 播客
AI news / WeChat / RSS / YouTube / podcasts
↓
今日推送 · Daily Feed
↓
全文阅读与收藏 · Read & Save
↓
原子知识卡片 · Atomic Knowledge Cards
↓
魔法写作 / 播客解读 · Create & Explore
↓
可追溯输出 · Source-aware Output
| Layer | Technology |
|---|---|
| Frontend | React 19, Vite, TypeScript, Tailwind CSS, tldraw |
| Backend | Express, Node.js 22+, RSS Parser, Readability, JSDOM |
| Database | PostgreSQL with parameterized pg queries; optional pgvector |
| AI | OpenAI-compatible API, OpenAI Agents SDK |
| Auth | express-session, connect-pg-simple, bcryptjs |
| Optional services | Resend/SMTP, Baidu Translate, Volcengine ASR |
| Deployment | Railway + PostgreSQL |
- Node.js 22+
- PostgreSQL 14+
git clone https://github.com/SkyNotSilent/Atom-Flow.git
cd Atom-Flow
npm ci
cp .env.example .env至少配置以下环境变量:
DATABASE_URL=postgresql://user:password@localhost:5432/atomflow
SESSION_SECRET=replace-with-a-random-secret
AI_API_KEY=your-api-key
AI_BASE_URL=https://your-openai-compatible-provider.example/v1
AI_MODEL=your-model启动开发环境:
npm run dev知识原子化默认读取:
AI_API_KEY
AI_BASE_URL
AI_MODEL写作 Agent 默认复用 OpenAI-compatible 配置;如需单独使用官方 OpenAI,可配置:
OPENAI_API_KEY
OPENAI_MODEL其他可选能力:
# Email: Resend is tried first, SMTP is the fallback channel
RESEND_API_KEY=...
SMTP_USER=...
SMTP_PASS=...
# Service health alerts (sent to SECURITY_CONTACT_EMAIL)
DB_PROBE_INTERVAL_SECONDS=30
ERROR_RATE_THRESHOLD_PERCENT=10
ERROR_RATE_MIN_REQUESTS=20
ALERT_MAX_PER_HOUR=10
# Translation
BAIDU_TRANSLATE_APPID=...
BAIDU_TRANSLATE_KEY=...
# Real-time speech recognition
VOLCENGINE_ASR_APPID=...
VOLCENGINE_ASR_TOKEN=...
VOLCENGINE_ASR_CLUSTER=volcengine_streaming_common
# Custom RSSHub
RSSHUB_BASE=https://rsshub.app
# Production Magic Writing Canvas
VITE_TLDRAW_LICENSE_KEY=...不要把真实密钥提交到仓库。Never commit real credentials to the repository.
核心数据保存在 PostgreSQL:
| Tables | Purpose |
|---|---|
users, verification_codes, session |
Accounts, verification, and sessions |
user_subscriptions, user_articles |
Custom feeds and per-user article state |
saved_articles, saved_cards |
Source articles and atomic knowledge cards |
notes, user_preferences |
Notes and preferences |
write_style_skills |
Reusable writing and extraction instructions |
write_canvas_projects, write_canvas_nodes, write_canvas_edges |
Infinite-canvas structure |
write_canvas_assets |
Canvas text, files, images, and extracted content |
write_agent_templates, write_agent_instances, write_canvas_agent_messages |
Writing-agent configuration and conversations |
数据库写入使用参数化查询,并按 user_id 隔离用户数据。
npx tsc --noEmit
npm run lint
npm test
npm run build涉及 RSS 运行时或内存治理时,可额外运行:
npm run test:rss:cache
npm run test:rss:soak- 从 GitHub 创建 Railway 服务并添加 PostgreSQL。
- 配置
DATABASE_URL、强随机SESSION_SECRET、AI 配置、APP_URL和精确的ALLOWED_ORIGINS。 - 如启用魔法写作画布,配置与实际授权匹配的
VITE_TLDRAW_LICENSE_KEY。 - 使用
railway.json的/api/health健康检查,并在部署后验证日志与线上关键流程。 - 在共享状态、队列和全局限流迁移完成前保持单 Web 副本。
AI、翻译、语音和邮件功能可能向部署实例配置的第三方供应商发送完成请求所需的数据。公开部署前,请披露实际供应商并提供清晰的就地告知。
- 安全政策 · Security
- 隐私说明 · Privacy
- 服务条款 · Terms
- 第三方声明 · Third-party notices
- 对外发布说明 · Public release notes
AtomFlow 自有代码默认采用 MIT License。魔法写作无限画布使用 tldraw,生产环境必须遵守 tldraw 自有许可证并配置适当、有效的许可证密钥;AtomFlow 的 MIT License 不会重新许可 tldraw。详见 LICENSES/TLDRAW_LICENSE.md。
www.atomflow.cloud