一个类似「知云文献翻译」的桌面 PDF 阅读翻译工具，使用 Python + PyQt6 开发。
A desktop PDF reader & translator similar to "Zhiyun Literature Translation", built with Python + PyQt6.
左侧查看 PDF，右侧显示翻译结果；用鼠标刷选 PDF 内容即可翻译，支持记录阅读位置，下次从上次位置继续阅读。
View the PDF on the left and the translation on the right. Select text in the PDF with your mouse to translate it. Your reading position is remembered so you can resume where you left off.
- 📄 PDF 阅读：按需渲染，打开大文件也不卡顿；支持适合页面 / 适合宽度 / 百分比三种缩放模式，也可用
Ctrl + 滚轮快速缩放 - 🖱️ 鼠标刷选翻译：按段落刷选文本，保持句子连续性，翻译效果更好
- 🎯 浮动翻译按钮：刷选文本后，翻译按钮自动出现在鼠标附近，点击即可翻译，无需移动鼠标
- 🌐 多翻译引擎：内置微软 Edge、通用 OpenAI 兼容大模型、MyMemory 三个接口，可自动回退或手动选择
- 🤖 全文总结：对整篇论文调用大模型生成结构化总结（背景、方法、数据、结论、优缺点、后续方向），流式输出实时显示，独立可缩放窗口，支持 Markdown 渲染与公式显示
- 🔍 全文搜索：关键词全文搜索，结果黄色高亮并自动垂直居中定位，支持上一个/下一个循环跳转
- 📂 拖拽打开：直接把 PDF 文件拖到窗口即可打开
- 🖥️ 窗口状态记忆：记住上次关闭时的窗口大小、位置与最大化状态
- ⚙️ 大模型设置：支持任意 OpenAI 兼容服务商（OpenAI、DeepSeek、智谱、硅基流动、本地 Ollama 等），模型名可从服务商下拉拉取
- 📌 阅读位置记忆：自动记录每个 PDF 的阅读位置，下次打开自动跳转
- 🕘 最近打开历史：记录最近打开的 10 个文件，支持一键清空
- 🔗 PDF 链接可点击：悬停显示手型光标，点击用系统浏览器打开；正文中的参考文献引用（内部链接）点击可直接跳转到文末对应条目
- 🌙 日间/夜间主题：设置菜单可切换日间与夜间模式，夜间模式下 PDF 页面自动反色（白底变黑底），晚上读论文不刺眼
- ⌨️ 快捷键：
Ctrl + / Ctrl -调节缩放，Ctrl + L切换全屏，Ctrl + T翻译 - 🔐 API Key 安全存储：大模型 Key 存入 Windows 凭据管理器（keyring），不落盘明文
- 🕐 状态栏日期时间：右下角实时显示当前日期和时间，左侧显示当前翻译引擎状态
- 🎨 界面美化：右侧面板卡片化（圆角 + 阴影）、翻译结果 Markdown 渲染、浮动按钮淡入动画、可调翻译字体、内置 Logo、关于对话框（含官网链接）
- 📄 PDF reading: on-demand rendering, smooth even for large files; supports Fit Page / Fit Width / Percentage zoom modes
- 🖱️ Select-to-translate: selects text by paragraph to preserve sentence continuity for better translations
- 🎯 Floating translate button: appears near the mouse after selecting text; click to translate without moving the mouse
- 🌐 Multiple translation engines: built-in Microsoft Edge, generic OpenAI-compatible LLM, and MyMemory, with auto-fallback or manual selection
- 🤖 Full-document summary: summarizes the entire paper via LLM (background, methods, data, findings, pros/cons, future work) with streaming output, shown in a resizable standalone window with Markdown rendering and formula display
- 🔍 Full-text search: keyword search across the whole document, results highlighted in yellow and auto-centered vertically, with prev/next cyclic navigation
- 📂 Drag & drop: drag a PDF file onto the window to open it
- 🖥️ Window state memory: remembers window size, position, and maximized state from the last session
- ⚙️ LLM settings: supports any OpenAI-compatible provider (OpenAI, DeepSeek, Zhipu, SiliconFlow, local Ollama, etc.); model names can be fetched from the provider
- 📌 Reading position memory: automatically remembers the position of each PDF and resumes there next time
- 🕘 Recent files: remembers the last 10 opened files, with one-click clear
- 🔗 Clickable PDF links: shows a hand cursor on hover, opens in the system browser on click; in-text reference citations (internal links) jump directly to the corresponding entry at the end of the document
- 🌙 Light/Dark theme: switch between light and dark modes in the Settings menu; in dark mode the PDF pages are automatically inverted (white background becomes black) for comfortable night reading
- ⌨️ Shortcuts:
Ctrl + / Ctrl -to zoom,Ctrl + Lto toggle fullscreen,Ctrl + Tto translate - 🔐 Secure API key storage: LLM key is stored in the Windows Credential Manager (keyring), never in plaintext
- 🕐 Status bar clock: shows the current date and time in the bottom-right corner, plus the active translation engine on the left
- 🎨 Polished UI: card-based right panel (rounded corners + shadow), Markdown rendering for translation results, fade-in animation for the floating button, adjustable translation font, built-in logo, About dialog (with website link)
需要 Python 3.9+。
Requires Python 3.9+.
# 克隆仓库 / Clone the repository
git clone https://github.com/fangvv/PDF-AI-Viewer.git
cd PDF-AI-Viewer
# 安装依赖 / Install dependencies
pip install -r requirements.txt
# 国内用户可使用清华源加速 / Chinese users can use the Tsinghua mirror
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simplepython main.py- 点击「打开 PDF」或使用
Ctrl + O打开一个 PDF 文件（也可直接把 PDF 拖到窗口） - 在左侧 PDF 中用鼠标刷选要翻译的文本，翻译按钮会出现在鼠标附近，点击即可翻译
- 翻译结果显示在右侧；也可点击「翻译选中内容」或按
Ctrl + T - 可在右上角选择翻译引擎（自动 / 微软 Edge / 大模型 / MyMemory）
- 使用「A- / A+」按钮调节翻译字体大小
- 缩放：工具栏选择适合页面 / 适合宽度 / 百分比，或按住
Ctrl滚动鼠标滚轮快速缩放 - 点击工具栏「全文总结」对整篇论文生成总结，结果在独立窗口中流式显示
- 在工具栏搜索框输入关键词回车，即可全文搜索并高亮定位，支持「上一个 / 下一个」跳转
- 首次使用大模型翻译或总结前，请到「设置 → 大模型设置」填写接口地址、API Key 和模型名
- 夜间阅读可到「设置 → 主题」切换夜间模式，PDF 页面会自动反色
- 若 PDF 自带内部链接，点击正文中的参考文献引用（如 [1]）可直接跳转到文末对应条目
- 关闭程序后，下次打开同一 PDF 会自动跳转到上次阅读位置，并恢复窗口状态
- Click "Open PDF" or press
Ctrl + Oto open a PDF file (or drag a PDF onto the window) - Select the text you want to translate in the PDF on the left; a translate button appears near the mouse — click it to translate
- The result appears on the right; you can also click "Translate Selected" or press
Ctrl + T - Choose a translation engine in the top-right (Auto / Microsoft Edge / LLM / MyMemory)
- Use the "A- / A+" buttons to adjust the translation font size
- Zoom: choose Fit Page / Fit Width / Percentage in the toolbar, or hold
Ctrland scroll the mouse wheel for quick zoom - Click "Full-document Summary" in the toolbar to summarize the entire paper; the result streams into a standalone window
- Type a keyword in the toolbar search box and press Enter to search the whole document with highlighted, centered results; use "Prev / Next" to navigate
- Before using LLM translation or summary for the first time, configure the base URL, API key, and model under "Settings → LLM Settings"
- For night reading, switch to dark mode under "Settings → Theme"; PDF pages are automatically inverted
- If the PDF has internal links, click an in-text reference citation (e.g. [1]) to jump directly to the corresponding entry at the end of the document
- After closing, reopening the same PDF resumes at your last reading position and restores the window state
本软件支持任意 OpenAI 兼容 的大模型接口，由用户自行选择服务商并填写：
This software supports any OpenAI-compatible LLM endpoint. You choose the provider and fill in:
-
接口地址 (Base URL)：例如
https://api.openai.com/v1、https://api.deepseek.com/v1、https://api.siliconflow.cn/v1等 -
API Key：从服务商获取，安全存入 Windows 凭据管理器
-
模型名 (Model)：可手动输入，或点击「刷新模型」从服务商自动拉取下拉列表
-
Base URL: e.g.
https://api.openai.com/v1,https://api.deepseek.com/v1,https://api.siliconflow.cn/v1, etc. -
API Key: obtained from the provider, securely stored in the Windows Credential Manager
-
Model: can be typed manually, or fetched from the provider via the "Refresh Models" button
pdf_translator/
├── main.py # 主窗口：分栏布局、工具栏、菜单栏、翻译/总结线程、浮动按钮
├── pdf_viewer.py # PDF 阅读器：按需渲染、刷选、缩放、链接点击
├── translator.py # 翻译引擎：Edge / OpenAI 兼容大模型 / MyMemory，流式总结
├── settings.py # 配置存储：阅读位置、最近历史、界面设置、大模型配置
├── make_logo.py # Logo 生成脚本
├── requirements.txt # 依赖清单
├── AGENTS.md # 给 AI 助手/开发者的工程说明与约定
├── LICENSE # MIT 许可证
├── dist/ # 打包产物（PDF阅读翻译器.exe）
└── logo.ico / logo.png # 应用图标
- PyQt6 — 桌面 GUI 框架
- PyMuPDF (fitz) — PDF 渲染与文本提取
- requests — 调用翻译 / 大模型 API
- keyring — 系统凭据管理器（安全存储 API Key）
| 引擎 / Engine | 说明 / Description | 需要 Key |
|---|---|---|
| 微软 Edge / Microsoft Edge | 官方翻译接口，质量稳定 | 否 |
| 大模型 / LLM | 通用 OpenAI 兼容大模型，翻译质量高，可做 AI 总结 | 是（可选） |
| MyMemory | 免费在线翻译服务 | 否 |
选择「自动」时，程序会依次尝试各引擎，直到成功为止。
When "Auto" is selected, the program tries each engine in turn until one succeeds.
- 灵感来自「知云文献翻译」/ Inspired by "Zhiyun Literature Translation"
- 感谢 PyQt6、PyMuPDF 等开源项目 / Thanks to PyQt6, PyMuPDF and other open-source projects