# 📚 收藏到就是学到 > 一款 Chrome 浏览器扩展，在你收藏 X (Twitter) 内容时，自动生成 AI 摘要并保存为本地 Markdown 文件，让每次收藏都变成一次学习。 [English](#english) | 中文 <a href="https://x.com/JamesAI/status/2021089989136875810">查看 X 发布帖</a> ### 演示视频 <video src="assets/demo.mp4" width="100%" controls></video> --- ## 功能特点 - **一键摘要** — 点击收藏按钮，自动生成结构化 TLDR 摘要（要点提炼、步骤流程、事实核查评分） - **AI 开关** — 一键关闭 AI 摘要和事实核查，仅保存原文 + 元数据到 Markdown（无需 API Key），默认开启 - **原文模式** — 支持切换为原文模式，跳过 AI 摘要，直接保存完整原文到 Markdown（无需 API Key） - **多模型支持** — 支持 OpenAI (GPT)、Claude (Anthropic)、Kimi (月之暗面)、智谱 (GLM)，以及本地 Claude CLI（无需 API Key） - **自定义 Base URL** — 支持配置中转 API 地址，可走私有网关或代理服务 - **深度内容提取** — 自动展开"显示更多"折叠内容，支持 X Articles 长文、引用/转发长帖的全文抓取 - **卡片堆叠** — 支持连续快速收藏，多张 TLDR 卡片同时显示，互不阻塞 - **历史记录** — 自动保存所有摘要，随时回顾，附带原帖链接 - **Markdown 归档** — 每次收藏自动下载 Markdown 文件到本地，包含 TLDR + 原文，方便知识管理 - **自定义保存路径** — 通过 Native Helper 可选择任意本地文件夹保存 Markdown 文件 - **多语言摘要** — 支持简体中文、繁體中文、English、日本語、한국어 - **深色模式** — 跟随系统偏好自动切换，支持手动切换（自动/浅色/深色） - **事实核查** — 每条摘要末尾附带可信度评分 (1-10) - **安全存储** — API Key 通过 AES-GCM 加密存储在本地，不会同步到云端 ## 安装方法 1. 下载或克隆本仓库： ```bash git clone git@github.com:iamzifei/bookmark-is-learned.git ``` 2. 打开 Chrome，进入 `chrome://extensions/` 3. 开启右上角的 **开发者模式** 4. 点击 **加载已解压的扩展程序**，选择项目文件夹 5. 点击浏览器工具栏中的扩展图标，填写你的 API Key 并保存 ## 使用方法 1. **设置** — 点击扩展图标，选择 AI 模型，填入 API Key，选择摘要语言和保存模式 2. **AI 开关** — 在 AI 服务配置区域右侧有一个开关，关闭后收藏时仅保存原文和元数据，不调用 AI 3. **选择模式** — TLDR 摘要模式（默认）会生成 AI 摘要；原文模式直接保存完整原文 4. **收藏** — 在 X (Twitter) 时间线上，点击任意推文的收藏/书签按钮 5. **阅读摘要** — 页面右下角会弹出 TLDR 卡片，包含要点提炼和事实核查 6. **查看历史** — 点击扩展图标，切换到「历史记录」标签页 7. **本地归档** — 每次收藏自动下载 Markdown 文件到本地 ### 自定义保存路径（可选） 默认保存到 `Downloads/bookmark-is-learned/` 目录。如需保存到其他文件夹： 1. 在高级设置中点击「一键下载安装脚本」 2. 在终端运行 `bash ~/Downloads/install-btl-native.sh` 3. 重启浏览器 4. 在高级设置中点击「选择文件夹」，选择任意本地目录 ## 保存模式 | 模式 | 说明 | |------|------| | AI 开启 + TLDR 摘要模式 | 调用 AI 生成结构化摘要，Markdown 包含 TLDR + 原文（需要 API Key） | | AI 开启 + 原文模式 | 调用 AI 生成摘要，Markdown 包含 TLDR + 完整原文；若帖子含引用链接（含链接预览卡片），会在 `Referenced Links` 小节附上可点击链接（需要 API Key） | | AI 关闭 | 不调用 AI，Markdown 仅保存元数据 + 完整原文；若帖子含引用链接（含链接预览卡片），会在 `Referenced Links` 小节附上可点击链接（无需 API Key） | ## 支持的内容类型 | 类型 | 说明 | |------|------| | 普通推文 | 提取推文全文生成摘要 | | 长推文 | 自动展开"显示更多"获取完整内容 | | X Articles | 后台抓取长文全文，生成详细摘要 | | 引用/转发帖 | 自动获取被引用帖的完整内容一并总结 | | 帖子串 (Thread) | 后台抓取整个 Thread 内容 | ## Markdown 文件格式 每次收藏会自动保存一个 `.md` 文件，文件结构如下： **TLDR 摘要模式：** ```markdown # 作者名 或 文章标题 > **Author**: 作者名 > **Source**: https://x.com/user/status/123456 > **Date**: 2025-01-15 14:30 --- ## TLDR AI 生成的结构化摘要（要点、流程、事实核查评分） --- ## Original Content 原文完整内容 ### Quoted Content (by 被引用作者) 被引用/转发的完整内容（如有） ### Referenced Links - [https://github.com/volcengine/OpenViking](https://github.com/volcengine/OpenViking) ``` **原文模式：** ```markdown # 作者名 或 文章标题 > **Author**: 作者名 > **Source**: https://x.com/user/status/123456 > **Date**: 2025-01-15 14:30 --- ## TLDR AI 生成的结构化摘要 --- ## Original Content 原文完整内容 ### Quoted Content (by 被引用作者) 被引用/转发的完整内容（如有） ### Referenced Links - [https://github.com/volcengine/OpenViking](https://github.com/volcengine/OpenViking) ``` **AI 关闭模式：** ```markdown # 作者名 或 文章标题 > **Author**: 作者名 > **Source**: https://x.com/user/status/123456 > **Date**: 2025-01-15 14:30 --- ## Original Content 原文完整内容 ### Quoted Content (by 被引用作者) 被引用/转发的完整内容（如有） ### Referenced Links - [https://github.com/volcengine/OpenViking](https://github.com/volcengine/OpenViking) ``` ## 工作原理 ``` 用户点击收藏 → 内容脚本检测点击 → 提取推文内容（展开折叠、抓取全文） ↓ 后台脚本接收 → 如有长文/引用帖，后台标签页抓取完整内容 ↓ ┌─ AI 开启 ────────────────────────────────────┐ │ 调用 LLM API → 生成结构化 TLDR 摘要 │ └──────────────────────────────────────────────┘ ┌─ AI 关闭 ────────────────────────────────────┐ │ 跳过 API 调用，仅保存原文 + 元数据 │ └──────────────────────────────────────────────┘ ↓ ┌─────────────────────────────────────────┐ │ ① 页面右下角弹出卡片 │ │ ② 保存到插件历史记录 │ │ ③ 下载 Markdown 文件到本地 │ └─────────────────────────────────────────┘ ``` ## 默认模型 | 模型提供商 | 默认模型 | |-----------|---------| | OpenAI | `gpt-4o-mini` | | Claude | `claude-sonnet-4-20250514` | | Kimi | `moonshot-v1-8k` | | 智谱 | `glm-4-flash` | | 本地 Claude | 自动（使用本机 Claude CLI） | 可在设置中自定义模型版本（如 `gpt-4o`、`claude-opus-4-20250514` 等）。 本地 Claude 使用已安装的 Claude Code CLI (`npm install -g @anthropic-ai/claude-code`)，无需 API Key，需先完成 CLI 登录认证并安装 Native Helper。 可选配置 `Base URL` 以使用中转服务： - 填写 `https://your-proxy.com/v1` 时，将自动补全为对应模型接口 - 也可直接填写完整接口地址，如 `https://your-proxy.com/v1/chat/completions` - 首次保存会弹出权限授权，用于访问你填写的域名 ## 项目结构 ``` bookmark-is-learned/ ├── manifest.json # Chrome 扩展配置 (Manifest V3) ├── background.js # 后台 Service Worker（API 调用、内容抓取、历史保存、Markdown 下载） ├── content.js # 内容脚本（收藏检测、DOM 提取、卡片 UI） ├── content.css # 内容脚本样式（卡片堆叠、深色模式） ├── popup.html # 弹出页面（设置 + 历史记录） ├── popup.js # 弹出页面逻辑（标签切换、历史浏览） ├── popup.css # 弹出页面样式 ├── native-host/ # Native Messaging Host（自定义文件夹写入） │ └── btl_file_writer.py └── icons/ # 扩展图标 ├── icon16.png ├── icon48.png └── icon128.png ``` ## API Key 获取 | 模型 | 获取地址 | |------|---------| | OpenAI | https://platform.openai.com/api-keys | | Claude | https://console.anthropic.com/settings/keys | | Kimi | https://platform.moonshot.cn/console/api-keys | | 智谱 | https://open.bigmodel.cn/usercenter/apikeys | ## 许可证 MIT License --- <a name="english"></a> # 📚 Bookmark Is Learned > A Chrome extension that automatically generates AI-powered TLDR summaries and saves local Markdown files when you bookmark content on X (Twitter) — turning every bookmark into a learning moment. [中文](#) | English <a href="https://x.com/JamesAI/status/2021089989136875810">View on X</a> ### Demo <video src="assets/demo.mp4" width="100%" controls></video> --- ## Features - **One-Click Summaries** — Bookmark a post and instantly get a structured TLDR (key points, step-by-step processes, fact-check scoring) - **AI Toggle** — Disable AI summarization and fact-checking with one click — saves only original text + metadata to Markdown (no API Key needed), enabled by default - **Original Text Mode** — Switch to Original mode to save the full original text directly to Markdown without AI summarization (no API Key required) - **Multi-Model Support** — Choose between OpenAI (GPT), Claude (Anthropic), Kimi (Moonshot), Zhipu (GLM), and Local Claude CLI (no API Key needed) - **Custom Base URL** — Route requests through your API proxy or private gateway - **Deep Content Extraction** — Auto-expands "Show more" truncated text, fetches full X Articles, and retrieves complete quoted/retweeted long posts - **Card Stacking** — Bookmark multiple posts in rapid succession — each TLDR loads independently as a stacked card - **History** — All summaries are saved automatically with links back to the original posts - **Markdown Export** — Each bookmark is automatically saved as a local Markdown file (TLDR + original content) for knowledge management - **Custom Save Path** — Install the Native Helper to save Markdown files to any local folder - **Multi-Language** — Summaries available in Simplified Chinese, Traditional Chinese, English, Japanese, and Korean - **Dark Mode** — Follows your system preference automatically, with manual toggle (auto/light/dark) - **Fact Check** — Every summary includes a credibility score (1-10) - **Secure Storage** — API Keys are encrypted via AES-GCM and stored locally only (never synced to the cloud) ## Installation 1. Clone this repository: ```bash git clone git@github.com:iamzifei/bookmark-is-learned.git ``` 2. Open Chrome and navigate to `chrome://extensions/` 3. Enable **Developer mode** (top right toggle) 4. Click **Load unpacked** and select the project folder 5. Click the extension icon in the toolbar, enter your API key, and save ## Usage 1. **Configure** — Click the extension icon, select your AI model, enter your API key, choose the summary language and save mode 2. **AI Toggle** — Use the toggle switch in the AI config section to enable/disable AI; when off, bookmarks save only original text and metadata 3. **Choose Mode** — TLDR mode (default) generates AI summaries; Original mode saves full original text 4. **Bookmark** — On the X (Twitter) timeline, click the bookmark button on any post 5. **Read** — A TLDR card appears at the bottom-right corner with key insights and a fact-check score 6. **Browse History** — Click the extension icon and switch to the "History" tab 7. **Local Archive** — Each bookmark is automatically saved as a Markdown file locally ### Custom Save Path (Optional) By default, files are saved to `Downloads/bookmark-is-learned/`. To use a custom folder: 1. Click "Download install script" in Advanced Settings 2. Run `bash ~/Downloads/install-btl-native.sh` in Terminal 3. Restart your browser 4. Click "Choose Folder" in Advanced Settings to select any local directory ## Save Modes | Mode | Description | |------|-------------| | AI On + TLDR Mode | Calls AI to generate structured summaries; Markdown includes TLDR + original text (requires API Key) | | AI On + Original Mode | Calls AI to generate summaries; Markdown includes TLDR + full original text. If the post contains referenced links (including link-preview cards), they are appended under `Referenced Links` as clickable URLs (requires API Key) | | AI Off | Skips AI entirely; Markdown saves only metadata + full original text. If the post contains referenced links (including link-preview cards), they are appended under `Referenced Links` as clickable URLs (no API Key required) | ## Supported Content Types | Type | Description | |------|-------------| | Regular tweets | Extracts full tweet text for summarization | | Long tweets | Auto-expands "Show more" to get complete content | | X Articles | Fetches the full long-form article in a background tab | | Quoted/Retweeted posts | Fetches the complete quoted post and summarizes both | | Threads | Fetches the full thread content from the background | ## Markdown File Format Each bookmark automatically saves a `.md` file: **TLDR Mode:** ```markdown # Author Name or Article Title > **Author**: Author Name > **Source**: https://x.com/user/status/123456 > **Date**: 2025-01-15 14:30 --- ## TLDR AI-generated structured summary (key points, processes, fact-check score) --- ## Original Content Full original text ### Quoted Content (by Quoted Author) Full quoted/retweeted content (if applicable) ### Referenced Links - [https://github.com/volcengine/OpenViking](https://github.com/volcengine/OpenViking) ``` **Original Mode:** ```markdown # Author Name or Article Title > **Author**: Author Name > **Source**: https://x.com/user/status/123456 > **Date**: 2025-01-15 14:30 --- ## TLDR AI-generated structured summary --- ## Original Content Full original text ### Quoted Content (by Quoted Author) Full quoted/retweeted content (if applicable) ### Referenced Links - [https://github.com/volcengine/OpenViking](https://github.com/volcengine/OpenViking) ``` **AI Off Mode:** ```markdown # Author Name or Article Title > **Author**: Author Name > **Source**: https://x.com/user/status/123456 > **Date**: 2025-01-15 14:30 --- ## Original Content Full original text ### Quoted Content (by Quoted Author) Full quoted/retweeted content (if applicable) ### Referenced Links - [https://github.com/volcengine/OpenViking](https://github.com/volcengine/OpenViking) ``` ## How It Works ``` User clicks bookmark → Content script detects click → Extract tweet (expand, fetch full text) ↓ Background receives → If article/quoted post, fetch full content via background tab ↓ ┌─ AI On ──────────────────────────────────────┐ │ Call LLM API → Generate structured TLDR │ └───────────────────────────────────────────────┘ ┌─ AI Off ─────────────────────────────────────┐ │ Skip API call, save original text + metadata │ └───────────────────────────────────────────────┘ ↓ ┌──────────────────────────────────────────────────┐ │ ① Show card at bottom-right of page │ │ ② Save to extension history │ │ ③ Download Markdown file to local disk │ └──────────────────────────────────────────────────┘ ``` ## Default Models | Provider | Default Model | |----------|--------------| | OpenAI | `gpt-4o-mini` | | Claude | `claude-sonnet-4-20250514` | | Kimi | `moonshot-v1-8k` | | Zhipu | `glm-4-flash` | | Local Claude | Auto (uses local Claude CLI) | You can override the model version in settings (e.g. `gpt-4o`, `claude-opus-4-20250514`). Local Claude uses the installed Claude Code CLI (`npm install -g @anthropic-ai/claude-code`). No API Key required — just complete CLI authentication first and install the Native Helper. Optional `Base URL` for proxy routing: - `https://your-proxy.com/v1` will be expanded to the model-specific endpoint - Full endpoint is also supported, e.g. `https://your-proxy.com/v1/chat/completions` - The first save triggers a permission prompt for the custom domain ## Project Structure ``` bookmark-is-learned/ ├── manifest.json # Chrome extension config (Manifest V3) ├── background.js # Service worker (API calls, content fetching, history, Markdown download) ├── content.js # Content script (bookmark detection, DOM extraction, card UI) ├── content.css # Content script styles (card stacking, dark mode) ├── popup.html # Popup page (settings + history tabs) ├── popup.js # Popup page logic (tab switching, history browsing) ├── popup.css # Popup page styles ├── native-host/ # Native Messaging Host (custom folder writing) │ └── btl_file_writer.py └── icons/ # Extension icons ├── icon16.png ├── icon48.png └── icon128.png ``` ## Getting API Keys | Provider | URL | |----------|-----| | OpenAI | https://platform.openai.com/api-keys | | Claude | https://console.anthropic.com/settings/keys | | Kimi | https://platform.moonshot.cn/console/api-keys | | Zhipu | https://open.bigmodel.cn/usercenter/apikeys | ## License MIT License