Cross-Tab Semantic Search

Search every page you've visited by meaning, locally.

Cross-Tab Semantic Search — search every page you've visited by meaning, not keywords. 100% local. Cross-Tab Semantic Search turns the pages you browse into a private, searchable memory. Instead of trying to remember the exact words on a page, just describe what you're looking for and the extension finds it by meaning. All processing and storage happen inside your browser. Nothing is uploaded. THE PROBLEM IT SOLVES You read something useful last week but can't recall where. Browser history only matches titles and URLs, and full-text search only matches exact words. If the page said "deep learning tutorial" but you search "how to train a neural network", you get nothing. Cross-Tab Semantic Search matches by meaning, so you find it anyway. KEY FEATURES • Semantic search across all visited pages — find pages by concept, not exact wording. • Cross-language — the multilingual model lets an English query find Chinese content, and vice versa. • Fully local and private — no account, no server, no analytics. Your page content and queries never leave your device. • Automatic background indexing — pages are indexed quietly when a tab is idle, so browsing stays fast. • One-click navigation — click a result to jump to the page. If it's already open in a tab, the extension switches to it. • You're in control — set how long pages are kept, block specific sites, turn off auto-indexing, or wipe everything instantly. • Bilingual interface — English or 中文, automatically following your browser by default. HOW IT WORKS 1. Extract. As you browse, the extension pulls the main text out of each page, ignoring menus, ads, and clutter. 2. Embed. A small AI model running entirely inside your browser converts the text into vectors that capture its meaning. 3. Store. Text segments and vectors are saved in your browser's local database (IndexedDB) on your device. 4. Search. Your query is converted to a vector and matched against the stored ones by semantic similarity, returning the closest pages and passages. GETTING STARTED 1. Install the extension. 2. On first use, it downloads a one-time AI model file (a few tens of MB), then caches it for offline use. 3. Just browse normally — pages are indexed automatically. 4. Click the toolbar icon, type what you're looking for, and open any result. PRIVACY AND SECURITY • All indexed content, vectors, and search indexes are stored only in your browser. • The AI runtime is bundled in the extension and runs locally. • The only outbound request is the one-time model download from Hugging Face. It contains none of your browsing data. • Built-in browser pages (chrome://, file://, etc.) are never indexed. • See the full Privacy Policy for details. SETTINGS Interface language — Auto (follow browser), 中文, or English. Retention (days) — Auto-delete pages older than this. Default 30. Host blocklist — Sites you never want indexed. Auto-index — Turn automatic indexing on or off. Chunk size and overlap — Advanced. How page text is split before embedding. Index current tab — Manually index the page you're on right now. Clear all data — Permanently delete everything indexed. FAQ Does my data leave my computer? No. Everything stays in your browser's local storage. There is no server. Does it need an internet connection? Only once, to download the AI model. After that it works offline. Will it slow down my browsing? Indexing runs in the background when a tab is idle, and the AI runs in a separate context, so normal browsing stays responsive. How do I keep sensitive pages out of the index? Add their site to the blocklist, turn off auto-indexing, or use a private window (where the extension does not run by default). How much storage does it use? It depends on how many pages you index. You can cap it with the retention period, and the Stats panel shows current usage. "Clear all data" frees everything instantly. SYSTEM REQUIREMENTS Google Chrome or Microsoft Edge, version 116 or newer. ----------------- 跨标签页语义搜索 —— 按"含义"而非关键词，搜索你访问过的每一个网页。100% 本地运行。 跨标签页语义搜索把你浏览过的网页变成一份私密、可检索的记忆。你不必记住页面上的确切措辞，只要描述你想找的内容，扩展就能按含义帮你找到。所有处理与存储都在你的浏览器内完成，不上传任何数据。 它解决什么问题 你上周读到过有用的东西，却想不起在哪。浏览器历史只能匹配标题和网址，全文搜索只能匹配一模一样的字词。如果页面写的是"深度学习教程"，而你搜"如何训练神经网络"，就什么也搜不到。 跨标签页语义搜索按含义匹配，照样能帮你找到。 核心功能 • 跨所有访问页面的语义搜索 —— 按概念查找，而不是精确字词。 • 跨语言 —— 多语言模型让英文查询也能找到中文内容，反之亦然。 • 完全本地、注重隐私 —— 无账号、无服务器、无统计分析。网页内容和查询绝不离开你的设备。 • 后台自动索引 —— 浏览时在标签页空闲时静默索引，不拖慢你的操作。 • 一键跳转 —— 点击结果即可跳到对应页面；若该页已在某个标签页打开，则直接切换过去。 • 由你掌控 —— 可设置页面保留时长、屏蔽特定网站、关闭自动索引，或一键清空全部数据。 • 双语界面 —— 中文 或 English，默认自动跟随浏览器语言。 工作原理 1. 抽取。浏览时，扩展从每个页面提取正文，自动忽略菜单、广告和杂项。 2. 嵌入。一个完全在你浏览器内运行的小型 AI 模型，把文本转换成能表达语义的向量。 3. 存储。文本片段与向量保存在你设备上浏览器的本地数据库（IndexedDB）中。 4. 搜索。你的查询被转成向量，按语义相似度与已存向量匹配，返回最相关的页面与段落。 快速上手 1. 安装扩展。 2. 首次使用时，会一次性下载 AI 模型文件（数十 MB），随后缓存以供离线使用。 3. 正常浏览即可 —— 页面会自动被索引。 4. 点击工具栏图标，输入你要找的内容，打开任意结果。 隐私与安全 • 所有已索引内容、向量与搜索索引仅保存在你的浏览器中。 • AI 运行时已打包在扩展内部，本地运行。 • 唯一的对外请求，是首次从 Hugging Face 下载模型 —— 其中不含你的任何浏览数据。 • 浏览器内置页面（chrome://、file:// 等）永远不会被索引。 • 详情见隐私政策。 设置项 界面语言 —— 自动（跟随浏览器）、中文 或 English。 保留天数 —— 自动删除早于该天数的页面。默认 30。 域名黑名单 —— 你永远不想被索引的网站。 自动索引 —— 开启或关闭自动索引。 分块字符数与重叠 —— 进阶。嵌入前如何切分页面文本。 索引当前标签页 —— 手动索引你当前所在的页面。 清空全部数据 —— 永久删除所有已索引内容。 常见问题 我的数据会离开我的电脑吗？ 不会。一切都保存在你浏览器的本地存储中，没有任何服务器。 它需要联网吗？ 仅需一次，用于下载 AI 模型。之后即可离线工作。 它会拖慢我的浏览吗？ 索引在标签页空闲时于后台进行，AI 也在独立上下文中运行，因此正常浏览保持流畅。 如何让敏感页面不被索引？ 把对应网站加入黑名单、关闭自动索引，或使用隐身窗口（扩展默认不在其中运行）。 它占用多少存储？ 取决于你索引了多少页面。你可以用保留期来限制，统计面板会显示当前用量。"清空全部数据"可立即释放全部空间。 系统要求 Google Chrome 或 Microsoft Edge，版本 116 或更新。