Blab Translation

🌐 Blab Translation - Chrome Extension

An AI-powered Chrome browser translation extension that supports selection translation and full-page translation, making web translation smarter and more natural.

✨ Features

Smart Translation

Math Formula Preservation: Automatically detects and preserves MathJax/KaTeX formulas without translation
Code Block Protection: Code snippets remain untouched during translation
Elegant Menu Translation: Sidebar translations align perfectly with original text (not icons)
Custom Prompts: Customize translation style with your own prompts (formal, casual, technical, etc.)

Selection Translation

A small translate icon appears next to the last line of a selection; click it to open the translation card
Or press the modifier key (⌘ on Mac, Ctrl elsewhere); Settings → Selection trigger picks icon, modifier key, or both
The card sits beside the selection instead of covering it, and shrinks and scrolls when the text is long
Card actions: retranslate, switch between Chrome built-in and your AI for this card, copy, and read aloud
The card shows which engine translated it; errors show on the card, never as a translation
The modifier key, float ball and right-click menu follow the display setting (card or inline)

Hover Translation

Hover a paragraph and press the hotkey (default: Shift) to translate inline
Translation appears directly below the paragraph as bilingual text
Press the hotkey again to restore the original view
Hotkey is configurable in Settings
Press Esc to clear inline translations, or right-click a paragraph/translation to cancel it

Full-Page Translation

Translate the entire webpage with one click
Translations appear below original text, preserving layout
Inherits original styling (font, color, size)
Toggle show/hide translations
Site translation rules: choose, per site, which areas to translate, which to leave out, which to keep as the original, extra CSS for the translations, and the engine
Batched translation: at most 40 paragraphs or 9,000 characters per batch, 4 batches at a time on the built-in engine and 12 on an AI service

Translation Styles and Display

Six looks for page translations: Default, Underline, Dashed box, Highlight, Quote bar, and Blur until hovered (hover, focus or tap a blurred translation to read it). Switching restyles the page at once and translates nothing again
Bilingual or translation only, switched from any of four places: the Display row in the popup, the float ball menu, the Settings page, or Alt+T (rebind it at chrome://extensions/shortcuts)
In translation-only mode, point at (or tap) a translation to see its original in a small card; move away, tap again or press Esc to close it

Image Text (OCR)

Right-click any image to read the text in it — screenshots, signs, menus, comics
Recognition runs on your own device by default: free, offline, no API key
Translating what it read is a separate step you can switch off — sometimes "what does this say" is the whole question
Or point it at your own vision model when the image is a photograph or stylized type the local engine struggles with

Video Subtitles

Bilingual subtitles on YouTube, and on any video whose player carries a standard subtitle track — no per-site support needed
Only translates the subtitles you already have switched on, and hands the track back untouched when you switch them off
Drag to move, drag the edges to resize; position and size are remembered
Survives fullscreen

UI Polish

Inline translations inherit original typography for a clean, consistent look
Inline loading indicator is more visible to show translation progress
Settings controls aligned for consistent spacing and visual hierarchy

Float Ball

Draggable quick action button
Supports translating selection, page, and toggling translations
Position auto-saves, persists across page navigation

Other Features

Right-click context menu translation
Dark/Light theme toggle
76 target languages (39 translated free on-device by Chrome's built-in Translator, the rest through your AI service)
Right-to-left targets (Arabic, Hebrew, Persian, Urdu) are laid out right-to-left, and a right-to-left page translated into a left-to-right language is laid out left-to-right
Input text translation dialog

🚀 Installation

1. Download

git clone https://github.com/FutrixDev/translator.git
cd translator

2. Load in Chrome

Open Chrome browser
Navigate to chrome://extensions/
Enable "Developer mode" in the top right
Click "Load unpacked"
Select the translator folder

3. Configure API

Click the extension icon in the browser toolbar
Click "Settings"
Fill in API configuration:
- Provider: pick your service; Custom shows the API Endpoint field
- API Endpoint (Custom only): e.g., https://api.openai.com/v1/chat/completions
- API Key: Your API key. Leave it empty for a local model server (see Local models)
- Model Name: e.g., gpt-4.1-mini (pick from the dropdown, or type any model your endpoint serves)
Select target translation language
Click Test Connection to check it. Settings save as you type; there is no save button

📖 Usage

Selection Translation

Select text on any webpage
Click the translate icon next to the selection, or press the modifier key
Read the translation in the card; retranslate, switch engine, copy or listen from its action row
Press Esc or click × to close/clear

Hover Translation

Move the mouse over a paragraph
Press the hover hotkey (default: Shift)
Translation appears below the paragraph
Press the hotkey again to restore the original view
Press Esc to clear inline translations, or right-click a paragraph/translation to cancel it

Full-Page Translation

Method 1: Float Ball

Click the float ball in the bottom-right corner
Select "Translate Page"

Method 2: Extension Menu

Click the extension icon in toolbar
Click "Translate Page"

Method 3: Context Menu

Right-click on the page
Select "Translate this page"

Show/Hide Translations

After translation:

Click the float ball
Select "Hide Translations" or "Show Translations"
Translations are preserved, no need to re-translate

Site Translation Rules

On a page, open the float ball's ··· menu (or the toolbar popup) and choose Adjust what gets translated here. Point at an area and click it, press ↑ Parent to take the enclosing block, check the Matches count, and edit the selector if you like. Then choose:

Don't translate here: a whole block is not translated; a word or link inside a paragraph is left out of that paragraph's translation.
Keep original: a whole block is not translated; a word or link inside a paragraph stays, unchanged, inside the translation.
Only translate here: translate just this area on this site.

Every rule is listed under Settings → Site translation rules, where you can also write one by hand: the sites it matches (example.com, *.example.com, example.com/docs/*), the areas as CSS selectors, custom CSS for the translations (anything that loads from the network is refused), and the engine for that site. Rules sync with your browser account (at most 50 rules, 24 KiB), take effect on open pages as soon as they are saved, and win over the built-in site list. Export rules saves blab-site-rules-YYYYMMDD.json; Import rules… previews what it adds and replaces first.

Document Translation

Runs on our servers against your monthly free page allowance, so it needs a signed-in account.

Open the upload page: the popup's Translate a Local Document…, or the same item in the toolbar icon's right-click menu.
Drop a file on the page or click to pick one. Accepted: PDF (.pdf), Word (.docx), EPUB (.epub), MOBI (.mobi, .azw3), plain text (.txt) and Markdown (.md, .markdown).
Size limits: PDF 30 MB; Word, EPUB and MOBI 50 MB; TXT and Markdown 10 MB. Word, EPUB, TXT and Markdown are measured on your computer first (one page = 3,000 characters); anything over 800 pages, an empty file, or a file whose contents do not match its extension is refused on the spot, before anything is uploaded.
The file goes straight to storage through a one-time signed address, and the page shows the job's progress. You can close it: the job keeps running, the popup lists it, and a notification says when it is done.
If the document turns out longer than estimated, the job stops and asks. The popup's row reads "Needs your confirmation" with a Review button (the notification is clickable too); the page says how many more pages continuing will use. Continue carries on; Cancel Task, or doing nothing until the stated time, cancels it with a full refund.
When it is done:
- PDF: Open Bilingual PDF / Open Translated PDF open the result in a tab. Open in the popup does the same.
- Word, EPUB, TXT, Markdown: Save Bilingual File / Save Translated File save it as <name> (bilingual).<ext> / <name> (translated).<ext>. Open in the popup brings you back to the job's page.
- MOBI: the result is read on the website (View on the web); there is no file to download.

Web PDFs (an open PDF tab, or a link to one) can also be translated from the popup's Translate This PDF and the right-click menu; that path takes PDFs only.

First Run

Installing the extension opens a welcome page (only on a fresh install, never on an update). Every choice on it is saved the moment you make it:

Offline translation shows whether Chrome's on-device language pack for your target language is ready. If it can be downloaded, a download button appears; nothing is downloaded until you press it.
Translate into picks the target language.
Translation engine: keep Built-in (recommended), or choose Set up AI and then where the model runs: Ollama or LM Studio on this computer (the connection is filled in for you), or Cloud API. Each opens Settings at the connection card so you can pick a model and test it.
Keyboard shortcuts lists the extension's commands and their keys. Change them at chrome://extensions/shortcuts.

Open Full Settings and Done are at the bottom of the page.

Import and Export Settings

The Import & Export card is the last card on the Settings page.

Export Settings saves blab-settings-YYYYMMDD.json with your settings, site rules and site translation rules. Your API key is left out unless you tick Include my API key. Caches, usage statistics, your account sign-in, per-site prompt counters and panel positions stay on this device.
Import Settings… reads such a file and shows what it would change before anything is written: which settings change, which are skipped (unknown or invalid), how many site rules are merged in, how many site translation rules are added or replaced, and any part of the file this version does not recognise. Press Import to apply it or Cancel to leave everything as it is. Settings are merged over yours; site rules are merged into your list, the file winning for a site in both.
The preview warns when importing would let AI translate without a click (for example, turning the automatic engine to AI, or a site translation rule whose engine is AI), and when the file points the API endpoint somewhere new while your saved key stays.
A file that is not JSON, not a Blab Translation settings file, from another file version, damaged in any part, over 1 MB, or that would give selection and hover translation the same hotkey is refused as a whole, and nothing is written.

⚙️ Supported APIs

Works with any OpenAI-compatible API endpoint. Just configure the endpoint URL, API key, and model name.

Service	Example API Endpoint	Notes
OpenAI	`https://api.openai.com/v1/chat/completions`	GPT-4o, GPT-4o-mini, etc.
Anthropic Claude	`https://api.anthropic.com/v1/messages`	Claude Sonnet, Opus, Haiku
Azure OpenAI	`https://your-resource.openai.azure.com/...`
Google Gemini	`https://generativelanguage.googleapis.com/v1beta/openai/chat/completions`	Gemini Pro, Flash, etc.
DeepSeek	`https://api.deepseek.com/v1/chat/completions`	DeepSeek-V3, etc.
OpenRouter	`https://openrouter.ai/api/v1/chat/completions`	Multiple providers
Ollama (Local)	`http://localhost:11434/v1/chat/completions`	Local models
LM Studio (Local)	`http://localhost:1234/v1/chat/completions`	Local models

Auto-detection: The extension automatically detects Anthropic Claude API (by domain or /v1/messages path) and uses the correct request/response format.

Local models (Ollama, LM Studio)

A model server on your own machine or local network needs no API key. Choose Ollama (Local) or LM Studio (Local) as the Provider, or choose Custom and enter an endpoint on localhost, 127.0.0.1, a private LAN address (10.x, 172.16-31.x, 192.168.x) or a .local name, and leave API Key empty. No authorization header is sent when the key is empty; if your server does require one (LM Studio's "Require Authentication"), fill it in and it is sent as usual.

Both servers have to be told to accept requests from a browser extension:

Ollama only accepts browser extensions listed in OLLAMA_ORIGINS, and answers anything else with HTTP 403. Set OLLAMA_ORIGINS=chrome-extension://* in Ollama's environment and restart it (on macOS: launchctl setenv OLLAMA_ORIGINS "chrome-extension://*", then quit and reopen the Ollama app).
LM Studio: turn on CORS in the server settings of the Developer tab, or start the server with lms server start --cors.

If a request fails, the error names the fix: a 403 from a local server shows these instructions, and a server that is not running is named by its address.

🌍 Supported Languages

Translation targets: 76 languages, named in your interface language. 39 of them can be translated free on-device by Chrome's built-in Translator; the other 37 are marked "AI only" in the language menus and go through your AI service.

Interface: 简体中文 • 繁体中文 • English • 日本語 • 한국어 • Français • Deutsch • Español • Português • Русский

📁 Project Structure

translator/
├── manifest.json          # Chrome extension configuration
├── background/            # Background script
├── content/               # Content script & styles
├── popup/                 # Popup menu
├── options/               # Settings page
├── i18n/                  # Internationalization
└── icons/                 # Extension icons

🧱 Technical Architecture

The extension follows a content-first architecture: content scripts collect and batch text, background scripts handle API calls, and UI surfaces manage user settings.

Content Script: scans DOM, filters code/table/math, batches text with token estimation, inserts translations.
Background Worker: builds prompts, calls OpenAI-compatible or Claude APIs, parses errors.
Options/Popup UI: manages API key, model, prompt, theme, and quick actions.
Storage: settings persisted in chrome.storage.sync.

🔁 Architecture Flowchart

flowchart LR
  U[User Action] --> C[Content Script]
  C --> D[DOM Scan + Filters]
  D --> P[Priority Split]
  P --> B[Token-Aware Batching]
  B -->|TRANSLATE / TRANSLATE_BATCH_FAST| W[Background Worker]
  W -->|Prompt Build + Placeholder Rules| A[LLM API]
  A --> W --> R[Translations]
  R --> C --> I[Insert Translations into DOM]
  S[Options/Popup] --> K[chrome.storage.sync]
  K --> C
  K --> W

📄 License

MIT License

🌐 叭叭翻译 - 智能翻译插件

一款基于 AI 的 Chrome 浏览器翻译插件，支持划词翻译和全文翻译，让网页翻译更智能、更自然。

✨ 功能特性

智能翻译

数学公式保留：自动识别并保留 MathJax/KaTeX 数学公式，不会被翻译破坏
代码块保护：代码片段在翻译过程中保持原样不变
优雅的菜单翻译：侧边栏译文与原文精确对齐（而非与图标对齐）
自定义 Prompt：支持自定义翻译风格（正式、口语化、技术文档等）

划词翻译

选中文字后，末行旁边出一个小翻译图标，点它打开翻译卡片
也可以按修饰键（Mac 上是 ⌘，其他系统是 Ctrl）；设置 → 划词触发方式可选图标、修饰键或两者都要
卡片贴着选区放，不盖住选中的字；原文很长时卡片缩高、内容区内部滚动
卡片操作：重译、在 Chrome 内置和我的 AI 之间换引擎（只对这张卡片）、复制、朗读
卡片标出这次是哪个引擎译的；出错显示在卡片上，不会当成译文
修饰键、悬浮球和右键菜单按显示方式设置出卡片或段落下方译文

悬停翻译

鼠标悬停段落并按下快捷键（默认：Shift）触发翻译
译文显示在段落下方，呈双语形式
再次按快捷键可恢复原文
快捷键可在设置中自定义
按 Esc 清除所有段落译文，或右键段落/译文选择取消

全文翻译

一键翻译网页：默认只翻正文，网站的导航、侧栏、菜单和页眉页脚不翻；想要某一页全部翻译，用「翻译整个页面」（悬浮球菜单或 Alt+W），或在设置里把「整页翻译范围」改成整页
iframe 里的内容（嵌入的文章、评论区、帖子）、Shadow DOM 组件里的文字（open 与 closed 都算）一起翻，之后新长出来的内容也跟着翻
页面标了 translate="no" 或 class="notranslate" 的部分（品牌名、代码、人名）原样保留
译文显示在原文下方，保持原网页布局
继承原文样式（字体、颜色、大小）
支持显示/隐藏译文切换
站点翻译规则：按网站指定翻哪些区域、哪些不翻、哪些保留原文，给译文加 CSS，指定翻译引擎
批量翻译：每批最多 40 段、9000 字符；内置引擎同时跑 4 批，AI 服务 12 批

译文样式与显示

网页译文有六种样式：默认、下划线、虚线框、高亮、引用竖线、模糊（悬停显示；悬停、聚焦或轻点模糊的译文即可看清）。切换立即生效，不会重新翻译
双语 / 仅译文可从四处切换：弹窗里的「显示」一行、悬浮球菜单、设置页，或快捷键 Alt+T（可在 chrome://extensions/shortcuts 改键）
仅译文模式下，指向（或轻点）一段译文会弹出小卡片显示原文；移开、再点一次或按 Esc 关闭

图片文字（OCR）

右键任意图片即可读出其中的文字——截图、路牌、菜单、漫画
默认在本机识别：免费、离线、不需要 API Key
识别之后是否翻译是可以单独关掉的一步——很多时候你只想知道“这上面写了什么”
照片和艺术字本地引擎读不动时，也可以改用你自己的视觉模型

视频字幕

YouTube，以及任何自带标准字幕轨的播放器，都能显示双语字幕——无需逐站适配
只翻译你已经打开的字幕；你把字幕关掉，字幕轨也原样交还
可拖动位置、拉边缘改大小，位置与尺寸都会记住
全屏下依然显示

UI 美化

译文继承原始排版，整体视觉更统一
内嵌加载提示更清晰，便于感知翻译进度
设置页控件对齐，层级更清晰

悬浮球

可拖动的快捷操作球
支持翻译选中文本、翻译页面、显示/隐藏译文
位置自动保存，跨页面保持

其他功能

右键菜单快速翻译
深色/浅色主题切换
支持 76 门目标语言：其中 39 门可用 Chrome 内置翻译免费在本机完成，其余 37 门走 AI
阿拉伯语、希伯来语、波斯语、乌尔都语等右到左语言的译文按右到左排版；右到左的页面译成左到右语言时同样按译文自己的方向排版
输入文本翻译对话框

🚀 安装使用

1. 下载插件

git clone https://github.com/FutrixDev/translator.git
cd translator

2. 加载到 Chrome

打开 Chrome 浏览器
地址栏输入 chrome://extensions/
开启右上角「开发者模式」
点击「加载已解压的扩展程序」
选择 translator 文件夹

3. 配置 API

点击浏览器工具栏中的插件图标
点击「打开设置」
填写 API 配置：
- 服务商: 选择你用的服务；选 Custom 才会出现 API 地址 一栏
- API 地址（仅 Custom）: 如 https://api.openai.com/v1/chat/completions
- API Key: 你的 API 密钥。用本地模型服务时留空即可（见本地模型）
- 模型名称: 如 gpt-4.1-mini（可从下拉列表选择，也可直接填写你的接口支持的任意模型）
选择目标翻译语言
点「测试连接」检查一下。设置边填边自动保存，没有保存按钮

📖 使用方法

划词翻译

在网页中选中需要翻译的文字
点选区旁边的翻译图标，或按修饰键
在卡片里看译文；操作行可以重译、换引擎、复制、朗读
按 Esc 或点击 × 关闭/清除

悬停翻译

将鼠标移动到段落上
按下悬停快捷键（默认：Shift）
译文显示在段落下方
再次按快捷键恢复原文
按 Esc 清除所有段落译文，或右键段落/译文选择取消

全文翻译

默认只翻正文（设置里的「整页翻译范围」：「只翻译正文（推荐）」/「翻译整个页面」）。

方式一：悬浮球

单击页面右下角的悬浮球直接翻译（再点一下还原）
或者打开球上的 ··· 菜单，选「翻译此页面」（按设置的范围）或「翻译整个页面」（这一页连导航、侧栏一起翻，重新加载页面后回到设置的范围）

方式二：插件菜单

点击浏览器工具栏的插件图标
点击「翻译页面」

方式三：右键菜单

在页面空白处右键
选择「翻译此页面」

方式四：快捷键

Alt+W：翻译整个页面（同悬浮球菜单里的那一项；可在 chrome://extensions/shortcuts 改键）

显示/隐藏译文

翻译完成后：

点击悬浮球
选择「隐藏译文」或「显示译文」
译文会被保留，再次显示无需重新翻译

站点翻译规则

在网页上打开悬浮球的 ··· 菜单（或工具栏弹窗），选 调整本站翻译区域。把鼠标移到某块区域上点一下，点 ↑上一层 换成外面那一块，看一眼 匹配 N 处 的数，需要的话直接改选择器，然后选：

不翻译这里：整块不翻译；段落里的一个词或链接会从这段的译文里拿掉。
保留原文：整块不翻译；段落里的一个词或链接原样留在译文里。
只翻译这里：这个网站只翻这块区域。

所有规则都列在 设置 → 站点翻译规则，也可以在那里手写：匹配的网站（example.com、*.example.com、example.com/docs/*）、用 CSS 选择器写的区域、给译文加的 CSS（会从网络加载东西的写法一律拒绝），以及这个网站用的翻译引擎。规则随浏览器账号同步（最多 50 条、24 KiB），保存后已打开的页面立即生效，优先于内置的网站列表。导出规则 保存为 blab-site-rules-YYYYMMDD.json；导入规则… 先预览会新增和替换几条。

文档翻译

跑在我们的服务器上、按月度免费页数计，所以需要先登录账号。

打开上传页：弹窗里的「翻译本地文档…」，或工具栏图标右键菜单里的同名一项。
把文件拖到页面上，或点击选择。支持 PDF（.pdf）、Word（.docx）、EPUB （.epub）、MOBI（.mobi、.azw3）、纯文本（.txt）、Markdown（.md、 .markdown）。
大小上限：PDF 30 MB；Word、EPUB、MOBI 50 MB；TXT、Markdown 10 MB。Word、EPUB、 TXT、Markdown 会先在本机数篇幅（3000 字符算一页）；超过 800 页、空文件、内容与扩展名对不上的文件当场拒收，什么都不上传。
文件经一次性签名地址直接传到存储，页面显示任务进度。可以关掉页面：任务照常跑，弹窗里能看到它，完成时有通知。
文档比预计长时，任务会停下来问你。弹窗那一行显示「需要你确认」和「查看」按钮（通知也可以点）；页面上写明继续要多用几页。点「继续」接着翻；点「取消任务」，或到页面写的时间还没处理，任务取消并全额退回。
完成后：
- PDF：「打开双语 PDF」/「打开译文 PDF」在新标签页打开结果；弹窗里的「打开」也一样。
- Word、EPUB、TXT、Markdown：「保存双语文件」/「保存译文文件」存成 <文件名> (双语).<扩展名> / <文件名> (译文).<扩展名>；弹窗里的「打开」回到这个任务的页面。
- MOBI：结果在网站上阅读（「在网页中查看」），没有可下载的文件。

网页上的 PDF（打开着的 PDF 标签页，或指向 PDF 的链接）也可以从弹窗的「翻译此 PDF」和右键菜单翻译；这条路只接 PDF。

首次安装

装好扩展会打开一个欢迎页（只在全新安装时打开，更新不会）。页面上的每个选择都会立即保存：

离线翻译：显示 Chrome 端上翻译所需的目标语言语言包是否就绪。可以下载时会出现下载按钮，不点就不会下载。
翻译成：选择目标语言。
翻译引擎：保留 内置（推荐），或选 配置 AI 再选模型在哪里运行：本机的 Ollama 或 LM Studio（连接会替你填好），或 云端 API。三者都会打开设置页并定位到连接卡片，在那里选模型、测试连接。
快捷键：列出扩展的命令和对应按键，在 chrome://extensions/shortcuts 修改。

页面底部是 打开完整设置 和完成。

导入与导出设置

设置页最后一张卡片是 导入与导出。

导出设置：保存为 blab-settings-YYYYMMDD.json，包含设置、站点规则和站点翻译规则。API 密钥默认不导出，勾选 包含我的 API 密钥 才会写进文件。缓存、使用统计、账号登录、各站点的询问计数和面板位置只留在这台设备上。
导入设置…：读取这样的文件，写入之前先列出会发生什么：哪些设置会改变、哪些被跳过（不认识或取值不合法）、多少条站点规则会并入、站点翻译规则会新增和替换几条、文件里有哪些部分此版本不认识。点 确认导入 才写入，点取消一切不变。设置按项合并到你现有的设置上；站点规则并入你的列表，同一站点以文件为准。
导入后 AI 会在无人点击时翻译（例如自动翻译引擎改成 AI，或某条站点翻译规则的引擎是 AI），或文件把 API 接口地址改到别处而你已保存的密钥会发往那里时，预览里会给出提示。
文件不是 JSON、不是叭叭翻译的设置文件、来自另一个文件版本、任何一部分损坏、超过 1 MB，或会让划词翻译和悬停翻译用同一个快捷键时，整个文件被拒绝，什么都不会写入。

⚙️ 支持的 API

支持所有 OpenAI 兼容的 API 接口，只需配置接口地址、API Key 和模型名称即可。

服务	API 地址示例	说明
OpenAI	`https://api.openai.com/v1/chat/completions`	GPT-4o, GPT-4o-mini 等
Anthropic Claude	`https://api.anthropic.com/v1/messages`	Claude Sonnet, Opus, Haiku
Azure OpenAI	`https://your-resource.openai.azure.com/...`
Google Gemini	`https://generativelanguage.googleapis.com/v1beta/openai/chat/completions`	Gemini Pro, Flash 等
DeepSeek	`https://api.deepseek.com/v1/chat/completions`	DeepSeek-V3 等
OpenRouter	`https://openrouter.ai/api/v1/chat/completions`	多种模型提供商
Ollama (本地)	`http://localhost:11434/v1/chat/completions`	本地模型
LM Studio (本地)	`http://localhost:1234/v1/chat/completions`	本地模型

自动检测：插件会自动检测 Anthropic Claude API（通过域名或 /v1/messages 路径），并使用正确的请求/响应格式。

本地模型（Ollama、LM Studio）

跑在本机或局域网里的模型服务不需要 API Key。服务商选 Ollama (Local) 或 LM Studio (Local)，或者选 Custom 并填一个 localhost、127.0.0.1、局域网私有地址（10.x、172.16-31.x、192.168.x）或 .local 主机名的地址， API Key 留空即可。Key 为空时不发送任何认证头；如果你的服务确实要 Key （LM Studio 的「Require Authentication」），照常填上就会照常发送。

两种服务都要先允许浏览器扩展访问：

Ollama 只接受 OLLAMA_ORIGINS 里列出的浏览器扩展，其余一律回 HTTP 403。在 Ollama 的环境里设置 OLLAMA_ORIGINS=chrome-extension://* 后重启（macOS： launchctl setenv OLLAMA_ORIGINS "chrome-extension://*"，再退出并重新打开 Ollama 应用）。
LM Studio：在 Developer 标签页的服务器设置里打开 CORS，或者用 lms server start --cors 启动服务。

请求失败时，错误提示会直接给出做法：本地服务回 403 时显示上面的设置方法，服务没启动时会写明它的地址。

🌍 支持的语言

目标语言： 76 门，名字按界面语言显示。其中 39 门可用 Chrome 内置翻译免费在本机完成；其余 37 门在语言菜单里标「仅 AI」，走你配置的 AI 服务。

界面语言： 简体中文 • 繁体中文 • English • 日本語 • 한국어 • Français • Deutsch • Español • Português • Русский

📁 项目结构

translator/
├── manifest.json          # Chrome 扩展配置
├── background/            # 后台脚本
├── content/               # 内容脚本 & 样式
├── popup/                 # 弹出菜单
├── options/               # 设置页面
├── i18n/                  # 国际化
└── icons/                 # 插件图标

🧱 技术架构说明

插件采用内容脚本驱动的架构：内容脚本负责收集与分批，后台负责调用 API，UI 管理用户配置。

Content Script：扫描 DOM，过滤代码/表格/公式，基于 token 估算分批并插入译文。
Background Worker：构建 Prompt，调用 OpenAI 兼容或 Claude API，统一错误处理。
Options/Popup UI：管理 API Key、模型、Prompt、主题与快捷操作。
Storage：配置持久化在 chrome.storage.sync。

🔁 技术架构流程图

flowchart LR
  U[用户触发] --> C[Content Script]
  C --> D[DOM 扫描与过滤]
  D --> P[首屏优先分组]
  P --> B[Token 估算分批]
  B -->|TRANSLATE / TRANSLATE_BATCH_FAST| W[Background Worker]
  W -->|构建 Prompt + 占位符规则| A[LLM API]
  A --> W --> R[译文结果]
  R --> C --> I[插入译文到页面]
  S[Options/Popup] --> K[chrome.storage.sync]
  K --> C
  K --> W

📄 License

MIT License

Made with ❤️

Name		Name	Last commit message	Last commit date
.githooks		.githooks
_locales		_locales
background		background
brand/blab-translation-2026-09-19		brand/blab-translation-2026-09-19
content		content
docs		docs
i18n		i18n
icons		icons
offscreen		offscreen
onboarding		onboarding
options		options
pdf		pdf
popup		popup
scripts		scripts
shared		shared
test		test
vendor/tesseract		vendor/tesseract
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CHANGELOG.md		CHANGELOG.md
CLAUDE.md		CLAUDE.md
README.md		README.md
WIKI.md		WIKI.md
manifest.json		manifest.json
package-lock.json		package-lock.json
package.json		package.json
playwright.config.js		playwright.config.js

Latest commit

History

Folders and files

Repository files navigation

Blab Translation