一款不绑定任何书城的本地阅读器，接入你自己的 AI——边读边讲、边读边问，划过的每一笔都自动变成笔记。
A local-first e-book reader tied to no bookstore, powered by your own AI — explain, translate, and chat while you read; every highlight becomes a note automatically.
BYOB（Bring Your Own Book）× BYOK（Bring Your Own Key）。 书和笔记全部存在本机，AI 请求直达你自己配置的接口，不经过任何第三方服务器。无账号、无统计、无订阅——你的书、你的 key、你的笔记。
BYOB (Bring Your Own Book) × BYOK (Bring Your Own Key). Books and notes live entirely on your device. AI requests go straight to the endpoint you configure — no third-party server in between. No accounts, no analytics, no subscriptions.
读非虚构类书籍时，我们常卡在三个地方：读不懂、记不住、看不下去。 BookFriend 把 AI 做成一个坐在你旁边的读书伙伴：
- 读不懂 → 选中任何一段话，让它结合上下文讲给你听
- 记不住 → 划线和问答自动沉淀为笔记卡片，读完一键导出
- 看不下去 → 不懂就问，追问到底，阅读从单向输入变成对话
- 导入 EPUB / PDF / TXT（MOBI 请先用 Calibre 转 EPUB）
- EPUB 精细排版：字号 / 主题 / 深色模式 / 目录跳转
- PDF 双模式：原版渲染 + 重排阅读模式（抽取文字、去页眉页脚、碎行合并回段落、识别标题层级，大多数文字版 PDF 可以像读书一样读）
- 扫描版 PDF 导入时自动 Vision OCR 加文字层，图片书也能选中、能问 AI
- TXT 自动识别 UTF-8 / GB18030 编码，启发式识别章节标题生成目录
- 选中文字 → 讲解 / 翻译 / 展开背景，流式输出，随时可停
- 多轮对话：对回答不满意就追问，面板里完整保留上下文
- 上下文自动携带：书名、章节、选段前后文，AI 知道你在读什么
- 划线 + AI 问答自动沉淀为笔记卡片，按书 → 章节归档
- 多轮对话结束后 AI 自动提炼要点更新卡片，而不是堆砌问答记录
- 一键导出 Markdown，直接导入 Obsidian / Notion
- 支持任何 OpenAI 兼容接口（DeepSeek / 通义 / Kimi / 中转 / Ollama……）+ Anthropic 官方接口
- 自定义 base URL、模型名；key 只存本机 Keychain，绝不上传
- 没有账号体系，没有自有服务器，没有任何埋点统计
- iOS / iPadOS 与 macOS 双 target，SwiftUI + SwiftData 共享绝大部分代码
- iOS 用 sheet 弹 AI 面板，macOS 用侧边栏，各自贴合平台习惯
（待补充：书架 / AI 面板 / 笔记列表）
brew install xcodegen
xcodegen # 生成 ShuYou.xcodeproj
open ShuYou.xcodeproj在 Xcode 中选择你的 Development Team 即可运行（iOS 17+ / macOS 14+）。 依赖：ZIPFoundation（SwiftPM 自动拉取）。
命令行构建（免签名）：
# macOS
xcodebuild -project ShuYou.xcodeproj -scheme ShuYouMac \
-destination 'generic/platform=macOS' -configuration Debug build \
CODE_SIGN_IDENTITY="-" CODE_SIGNING_REQUIRED=NO CODE_SIGNING_ALLOWED=NO
# iOS 模拟器
xcodebuild -project ShuYou.xcodeproj -scheme ShuYou \
-destination 'generic/platform=iOS Simulator' build \
CODE_SIGN_IDENTITY="-" CODE_SIGNING_REQUIRED=NO CODE_SIGNING_ALLOWED=NOSources/
├── App/ iOS 入口
├── Mac/ macOS 入口与 Mac 版阅读器
├── Models/ Book / NoteCard（SwiftData）、主题
├── Services/ AIService（SSE 流式，OpenAI 兼容 + Anthropic）
│ AISettings（BYOK，key 存 Keychain）
│ PDF 重排 / 字体扫描 / Vision OCR / TXT 章节识别
│ EPUB 导入 / Markdown 导出 / 本地存储约定
├── EPUB/ EPUB 解析（container.xml → OPF → spine → NCX/Nav 目录）
├── Reader/ 三个阅读器 + 选段中枢（SelectionCenter）
│ EPUB：WKWebView 渲染，注入 JS 实现选段定位与划线
│ PDF：PDFKit，选段 → AI；划线 = PDF 高亮批注
│ TXT：TextView，选段 → AI
├── Shared/ 跨平台补丁、Markdown 渲染、选中动作条
└── Views/ 书架 / AI 面板（流式多轮）/ 笔记列表 / 设置 / 目录侧栏
- 选段链路：三种阅读器统一经
SelectionCenter取选段，每次新选段 = 新AIRequest= 新面板实例 = 干净会话，无需手动清状态 - 划线定位：EPUB 用「文本 + 章节内偏移」锚定，重开章节后 JS 重新上色
- 笔记提炼：仅 1 轮对话存 AI 原始回答；≥2 轮结束后由 AI 提炼要点就地更新卡片，提取失败兜底存问答转录
下一步：断档续接（重开书时 AI 回顾上次进度）、章节钩子（带着问题读）、全书问答。
When reading non-fiction, we usually get stuck in three places: can't understand, can't remember, can't keep going. BookFriend turns AI into a reading companion sitting next to you:
- Can't understand → select any passage and get it explained in the context of the book
- Can't remember → highlights and Q&A automatically become note cards; export everything when you finish
- Can't keep going → ask questions and follow up until it's clear — reading becomes a conversation, not a one-way intake
- Import EPUB / PDF / TXT (convert MOBI to EPUB with Calibre first)
- Refined EPUB typography: font size / themes / dark mode / table of contents
- Dual PDF modes: original rendering + reflow reading mode (text extraction, header/footer removal, broken-line merging, heading detection — most text-based PDFs read like real books)
- Scanned PDFs get an automatic Vision OCR text layer on import — image-only books become selectable and askable
- TXT auto-detects UTF-8 / GB18030 encodings and heuristically builds a chapter outline
- Select text → Explain / Translate / Expand background, streamed, stoppable anytime
- Multi-turn chat: not satisfied? Follow up — the panel keeps full context
- Context comes for free: book title, chapter, and surrounding text, so the AI knows what you're reading
- Highlights + AI Q&A settle into note cards, filed by book → chapter
- After multi-turn chats, the AI distills key points into the card instead of dumping raw transcripts
- One-tap Markdown export, ready for Obsidian / Notion
- Works with any OpenAI-compatible endpoint (DeepSeek / Qwen / Kimi / relays / Ollama…) plus the official Anthropic API
- Custom base URL and model name; keys live only in the local Keychain, never uploaded
- No account system, no first-party server, no analytics of any kind
- Dual targets for iOS / iPadOS and macOS; SwiftUI + SwiftData share most of the code
- AI panel shows as a sheet on iOS and a sidebar on macOS — native to each platform
(Coming soon: bookshelf / AI panel / notes list)
brew install xcodegen
xcodegen # generates ShuYou.xcodeproj
open ShuYou.xcodeprojPick your Development Team in Xcode and run (iOS 17+ / macOS 14+). Dependency: ZIPFoundation (fetched automatically via SwiftPM).
Command-line builds (no signing):
# macOS
xcodebuild -project ShuYou.xcodeproj -scheme ShuYouMac \
-destination 'generic/platform=macOS' -configuration Debug build \
CODE_SIGN_IDENTITY="-" CODE_SIGNING_REQUIRED=NO CODE_SIGNING_ALLOWED=NO
# iOS Simulator
xcodebuild -project ShuYou.xcodeproj -scheme ShuYou \
-destination 'generic/platform=iOS Simulator' build \
CODE_SIGN_IDENTITY="-" CODE_SIGNING_REQUIRED=NO CODE_SIGNING_ALLOWED=NOSources/
├── App/ iOS entry point
├── Mac/ macOS entry point and Mac readers
├── Models/ Book / NoteCard (SwiftData), themes
├── Services/ AIService (SSE streaming, OpenAI-compatible + Anthropic)
│ AISettings (BYOK, keys in Keychain)
│ PDF reflow / font scanning / Vision OCR / TXT chapter detection
│ EPUB import / Markdown export / local storage conventions
├── EPUB/ EPUB parsing (container.xml → OPF → spine → NCX/Nav TOC)
├── Reader/ Three readers + selection hub (SelectionCenter)
│ EPUB: WKWebView rendering, injected JS for selection & highlights
│ PDF: PDFKit, selection → AI; highlights = PDF annotations
│ TXT: TextView, selection → AI
├── Shared/ Cross-platform shims, Markdown rendering, selection action bar
└── Views/ Bookshelf / AI panel (streaming, multi-turn) / notes / settings / TOC sidebar
- Selection pipeline: all three readers go through
SelectionCenter; each new selection = newAIRequest= fresh panel instance = clean session, no manual state clearing - Highlight anchoring: EPUB highlights anchor by "text + offset within chapter"; JS re-applies them when the chapter reopens
- Note distillation: single-turn chats save the raw AI answer; after 2+ turns the AI distills key points and updates the card in place, falling back to a Q&A transcript if extraction fails
Next up: resume-with-recap (AI reviews your last progress when you reopen a book), chapter hooks (questions to read with), whole-book Q&A.
MIT © 2026 luqiguan80-debug — 可自由使用、修改、商用，保留版权声明即可。 Free to use, modify, and distribute, commercially or otherwise, with attribution.