Leaf Reader

中文 | English

中文

Leaf Reader 是一个原生 macOS 文档阅读器，支持 PDF、EPUB 和 DOCX。它面向长文档阅读、学习和批注场景，提供阅读进度恢复、文档搜索、浅色/护眼/深色主题、AI 问答、选中文本翻译/解释/总结、背单词和本地朗读。

官网：https://leafreader.space/

截图

下载

下载 Leaf Reader 1.7.10 pkg 安装包

系统要求

• macOS 12.0 Monterey 或更高版本。
• 阅读器支持 Apple Silicon 和 Intel Mac；本地 TTS runtime 当前仅支持 Apple Silicon。
• AI 功能需要用户自行配置模型服务和 API Key；普通阅读不需要。
• Piper 本地朗读支持 Apple Silicon Mac 上的 macOS 12.0 Monterey 或更高版本。
• Kokoro 本地朗读需要 Apple Silicon Mac 上的 macOS 14.0 或更高版本。

主要功能

• 打开本地 PDF、EPUB、DOCX 文档，并自动恢复阅读位置。
• 支持文档搜索、PDF 翻页、书架、最近阅读、浅色/护眼/深色主题。
• 选中文本后可让 AI 解释、总结、翻译或继续追问上下文。
• 支持保存单词、复习新词、导出词表。
• 支持 Piper、Kokoro 和 Supertonic 本地朗读；短词和短句可回退到 macOS 系统语音。
• 文档保存在本机；只有使用 AI 功能时，相关文本才会发送到用户配置的模型服务。

可选朗读模型

Leaf Reader 可以使用 Piper、FluidAudio Kokoro Core ML 或 Supertonic Core ML 进行本地朗读。Kokoro 提供英文和中文声音，Piper 用于英文朗读，Supertonic 使用 Supertonic 3 模型并复用 FluidAudio CoreML 运行时。小型运行时已经随安装包提供，大模型文件按需下载。打开“设置 -> AI 分析 -> 朗读”即可下载 Piper、Kokoro 或 Supertonic。

朗读模型优先级会自动处理：优先使用用户选择的可用运行时，中文内容会切换到 Kokoro。短词或短句会直接使用 Apple 系统语音。

语音模型下载目前复用稳定的 v1.5.10 语音资源发布：

• https://github.com/dowellhz/LeafReader/releases/download/v1.5.10/kokoro-coreml-macos-arm64.tar.gz
• https://github.com/dowellhz/LeafReader/releases/download/v1.5.10/piper-tts-macos-arm64.tar.gz
• https://github.com/dowellhz/LeafReader/releases/download/v1.5.10/supertonic-coreml-macos-arm64.tar.gz

常规应用版本会复用这些模型文件。只有模型文件变化时才需要重新发布语音模型归档，并同步更新 SpeechRuntimeResourceManager.Runtime.runtimeAssetsReleaseTag。

更新记录

1.7.10

• 新增 PDF 连续滚动并修复上一页导航，同时加入论文式目录和更完整的词汇学习进度。
• PDF 搜索改为渐进返回，文档索引、词汇持久化和可见高亮恢复不再阻塞交互路径。
• 加入内容寻址的 PDF 文本快照和 DOCX 缓存，DOCX 使用流式解析并可在切换文档时取消旧任务。
• 生词按词元和词性分组，支持相关词形差异高亮、显示开关以及直接查看完整释义。
• 修复 PDF 跨行连字符选词，并强化 Web Reader 搜索、高亮和重复 occurrence 恢复。
• 新增事务式用户数据备份和恢复，同时确保 API 凭据不会进入备份包。
• 加强不可信文档、归档、缓存和 runtime 的安全检查，普通阅读不再触发不必要的凭据访问。

1.7.9

• 改进 PDF 朗读：会学习重复出现的页眉和页脚行，TTS 时自动跳过。
• 页码和页脚混在同一视觉行时会整行跳过，避免只删页码、仍读出页脚文字。
• 增加启动耗时诊断，并延迟初始化部分朗读组件，让日常启动更轻。
• 补充 PDF 页脚过滤、低位置正文误删和启动耗时快照的回归测试。

1.7.8

• 改进主题覆盖，工具栏、搜索、书架、背单词和 AI 界面更一致地跟随白色、护眼和深色主题。
• 加强发布和文档检查，本地检查会在临时目录构建手册，避免生成 HTML 噪声。
• 清理语音 runtime 发布维护逻辑，移除废弃下载包装，并把 Piper worker 支持代码拆得更聚焦。
• 增加主题调色板回归测试，保持 UI theme 扫描无警告。

1.7.7

• 新增难句解析入口：阅读笔记选中文本和正文选中句子后都可以直接请求 AI 拆解难句。
• 难句解析会按句子结构、主谓宾/从句/修饰关系、逐层翻译、常见表达和写法原因输出。
• 整理选中文本 AI 动作的共用流程，减少总结、翻译和难句解析之间的重复逻辑。

1.7.6

• 改进 AI 对话导出，支持 Markdown、HTML 和 PDF，并让导出按钮跟随当前主题配色。
• 合并阅读笔记导出流程到系统保存面板，隐藏标签栏，并把范围和文件类型放在同一个选择区域。
• 修复 AI 对话和阅读笔记导出的 HTML/PDF Markdown 渲染，标题、列表和 **加粗** 等内容会显示为实际样式。
• 将阅读笔记导出入口改名为“导出笔记”，并整理导出相关代码和测试。

1.7.5

• 新增阅读笔记导出选项，支持 Markdown、HTML 和 PDF，并可只导出收藏笔记。
• 改进 PDF 朗读分段匹配，重复文本和跨页/跨句朗读时能更准确地高亮当前句子。
• 修复 PDF 断行连字符单词保存和高亮，例如 Nine-\ntenths 会作为 Nine-tenths 处理。
• 收窄 PDF 单词下划线范围，避免跨行选词时生成过长下划线。
• 更新代码索引和相关逻辑测试。

1.7.4

• 优化单词查询路径：AI 请求时不再同步触发本地词典慢查询，本地词典 metadata 改为后台补齐，减少浮动菜单点词时的卡顿。
• 精简 ECDICT 查询策略：多词选区不再检索本地词典，未命中时不做全表扫描，检索依据会跟随主词条删除。
• 改进朗读浮动控制：上一条、下一条、重播和快捷键会聚焦到当前朗读句子所在页面与位置。
• 放大选中单词和句子的浮动菜单文字，让解释、总结、翻译等操作更容易点击和阅读。
• 整理 AI 气泡、单词持久化、朗读导航和词典查询相关代码，移除临时性能监控逻辑。

1.7.3

• 阅读笔记支持图片资源持久化，插入图片会保存到应用资源目录，重新打开笔记后不再只剩文件名或丢失图片。
• 改进阅读笔记的 AI 整理流程：向 AI 发送内容时保留图片占位符，整理结果会自动还原图片，避免图片在改写、总结或润色后消失。
• 重构阅读笔记 Markdown 渲染、序列化、样式策略和 slash 范围处理，减少从富文本反推 Markdown 带来的样式漂移。
• 放大阅读笔记窗口、正文、元信息和工具栏相关字号，并修复图片按钮可能卡住的问题。
• 优化朗读选区上下文：没有普通选区时，AI 总结、翻译和问问题会使用当前朗读选中的句子；朗读工具条增加当前句重播按钮。
• 改进单词查询：AI 结果会复用本地词典 Tags，标签展示更紧凑；网络不可用或模型不可达时更可靠地回退到本地词典。
• 统一历史单词标注和新标注的下划线颜色，并修复朗读选中句子解析后缺少下划线标记的问题。
• 调整 AI 气泡正文和标题字号，并整理 AI 面板、词汇、朗读和阅读笔记相关代码边界。

1.7.2

• 改进阅读笔记：支持保存按钮、复制/粘贴快捷键，并修复重新打开后样式和字数统计不同步的问题。
• 修复朗读手动/自动接续和下一页流程中的状态遗漏，让 PDF/Web 朗读翻页更稳定。
• 整理阅读笔记、朗读、文档、AI 和 embedding 状态结构，降低后续维护风险。

1.7.1

• 增强离线和未配置模型时的选词体验：本地 ECDICT 查询、朗读、复制和模型设置入口的状态更清晰。
• 优化背单词流程，默认词频优先，并避免一次性查询整批单词导致卡顿。
• 整理网络/模型状态、浮动窗口、本地词典和复习卡片加载结构。

1.7.0

• 增加阅读笔记入口，并改进笔记编辑、Ask AI、粘贴和右键菜单行为。
• 优化朗读控制状态、停止图标、语速滑块和浮动播放器布局。
• 增加词汇复习优先级选项，并整理词汇复习/列表结构，让每日复习更稳定。

1.6.6

• 降低 Piper runtime 的最低 macOS 标记到 macOS 12.0，并加强发布前 bundle 校验。
• 修复朗读模型状态判断，区分缺少运行时和缺少模型。
• 同步本地与远端语音模型清单，并明确本地 TTS runtime 的 Apple Silicon 要求。

1.6.5

• 修复安装版 Piper 朗读 runtime 缺少动态库搜索路径导致启动失败的问题。
• 细分朗读错误提示，区分模型未下载、中文内容需要 Kokoro，以及 runtime 启动失败。
• 发布检查增加 Piper runtime bundle 校验，避免缺失 LC_RPATH 的包进入发布流程。

1.6.4

• 增加 Piper 作为本地英文朗读模型选项，并随安装包提供 Piper runtime。
• 朗读浮动控制器增加停止和设置入口。
• 优化朗读模型下载列表、Piper 语速处理和发布资源上传流程。

1.6.3

• 朗读浮动播放器增加“下一页”按钮，可直接接续朗读下一页内容。
• 修复 PDF 双页模式下朗读从左页跳到右页时不应翻到下一屏的问题。
• 底部工具栏“书架”和“背单词”按钮增加与主题一致的 SF Symbol 图标。

1.6.2

• 增加朗读时的浮动播放器，支持上一句、暂停/继续、下一句，以及自动/手动接续模式。
• 优化朗读队列，点击下一句会立即停止当前句并播放下一句，手动模式每次只播放一句。
• 保留最近两句 wav 缓存，让上一句回退更流畅。
• 修复朗读浮动播放器与 PDF 单词标记、AI 提示气泡和页面点击之间的层级/点击冲突。

1.6.1

• 缩减安装包内置语音资源，只保留本地朗读所需的最小 runtime 资源。
• 删除重复 Kokoro 英文声音，保留 Bella、Heart、Adam、Emma、George。
• 保持本地朗读在更小的 runtime 资源下正常工作。

1.6.0

• 使用统一主题色重新打磨阅读器、AI 聊天、设置、最近文档和词汇界面。
• 给选中文本浮动工具栏动作增加 SF Symbol 图标，并让工具栏布局更紧凑。
• 改进 AI 聊天气泡持久化、关联单词处理和跨阅读会话的来源标注。
• 增强 embedding 在回填、重建和切换文档时的生命周期与状态处理。
• 优化词汇复习/列表导航和最近文档清理流程。

更早版本见 GitHub Releases。

许可证

Leaf Reader 使用 Apache License 2.0 许可发布。

第三方朗读模型和运行时版权归各自项目所有：

• FluidAudio Kokoro Core ML / Kokoro 模型：Apache License 2.0。
• Piper：MIT License；Piper 语音模型资源版权和许可遵循上游模型包随附的元数据。

English

Leaf Reader is a native macOS reader for PDF, EPUB, and DOCX documents. It is built with Swift, PDFKit, and WebKit, and focuses on a quiet reading experience with fast navigation, document search, reading progress restore, light and dark reader themes, and an optional AI panel for working with selected passages.

Website: https://leafreader.space/

Screenshots

Download

Leaf Reader 1.7.10 pkg installer

System Requirements

• macOS 12.0 Monterey or later.
• The reader supports Apple Silicon and Intel Mac; local TTS runtimes currently require Apple Silicon.
• An API key is optional and only needed for AI features.
• Piper local speech supports macOS 12.0 Monterey or later on Apple Silicon Macs.
• Kokoro local speech requires macOS 14.0 or later on Apple Silicon Macs.

Highlights

• Open local PDF, EPUB, and DOCX files in one macOS app.
• Restore the last opened document, page, zoom level, and reading position.
• Navigate PDFs with toolbar controls, keyboard paging, scroll paging, and direct page-number entry.
• Search documents with Command+F, next and previous result controls, and visible result positioning.
• Switch between light and dark reader themes for the document area, search overlay, recent files panel, and AI chat panel.
• Select text and ask the built-in AI assistant to explain, summarize, or translate passages.
• Read selected English or Chinese text with optional downloadable Piper, Kokoro, or Supertonic output where supported; otherwise Leaf Reader falls back to macOS system voices.
• Keep documents local; AI requests are only sent when the assistant is used with the configured API key.

Optional Speech Runtimes

Leaf Reader can use Piper, FluidAudio Kokoro Core ML, or Supertonic Core ML for local text-to-speech. Kokoro provides English and Chinese voices, Piper is used for English read aloud, and Supertonic provides multilingual local speech. Small speech runtime executables are bundled in the installer; large model files are downloaded on demand. Open Settings -> AI Analysis -> Speech to download Piper, Kokoro, or Supertonic.

Runtime priority is automatic: Leaf Reader uses the selected runnable runtime, and Chinese content switches to Kokoro. Short word or phrase selections use Apple TTS directly.

Speech model downloads currently point to the stable v1.5.10 speech asset release:

• https://github.com/dowellhz/LeafReader/releases/download/v1.5.10/kokoro-coreml-macos-arm64.tar.gz
• https://github.com/dowellhz/LeafReader/releases/download/v1.5.10/piper-tts-macos-arm64.tar.gz
• https://github.com/dowellhz/LeafReader/releases/download/v1.5.10/supertonic-coreml-macos-arm64.tar.gz

Regular app releases reuse those files. Regenerated speech archives should only be published when model files change, then SpeechRuntimeResourceManager.Runtime.runtimeAssetsReleaseTag should be updated to the new asset tag.

Changelog

1.7.3

• Made Reading Note images persistent: inserted images are copied into the app's Reading Note asset store and reopen correctly instead of degrading to filenames.
• Improved AI organize/polish flows for Reading Notes by protecting image placeholders before sending Markdown to AI and restoring them after the response.
• Refactored Reading Note Markdown rendering, serialization, inline style policy, slash ranges, and line-prefix handling to reduce rich-text-to-Markdown drift.
• Enlarged the Reading Note window, body text, metadata text, and toolbar sizing, and fixed the image picker action that could freeze the note editor.
• Improved read-aloud context fallback so Summary, Translate, and Ask use the currently spoken sentence when there is no explicit text selection; added replay for the current read-aloud segment.
• Improved vocabulary lookup by reusing local dictionary tags in AI word answers, compacting tag display, and falling back to the local dictionary more reliably when the model is unavailable.
• Kept historical and newly-created vocabulary underline colors consistent, including read-aloud sentence parsing highlights.
• Adjusted AI bubble body/header font sizing and continued splitting AI panel, vocabulary, read-aloud, and Reading Note code into clearer boundaries.

1.7.2

• Improved Reading Notes with a save button, copy/paste shortcuts, and fixes for reopened styling and word-count refresh.
• Fixed read-aloud continuation edge cases across manual/automatic advance and next-page flows for more reliable PDF/Web playback.
• Refactored Reading Notes, read-aloud, document, AI, and embedding state boundaries to reduce maintenance risk.

1.7.1

• Improved selection behavior when offline or when no model API key is configured, with clearer local ECDICT, speech, copy, and model-settings states.
• Refined vocabulary review with frequency-first ordering by default and current-card dictionary lookup to avoid batch lookup stalls.
• Cleaned up network/model capability, selection toolbar, local dictionary, and review-card loading boundaries.

1.7.0

• Added Reading Notes access from the reader toolbar and refined note editing, Ask AI, paste, and context-menu behavior.
• Improved read-aloud controls with clearer toolbar states, stop icons, a speed slider, and better floating control layout.
• Added vocabulary review priority options and tightened the vocabulary review/list architecture for more reliable daily review.

1.6.6

• Lowered the Piper runtime minimum macOS marker to macOS 12.0 and strengthened release-time bundle checks.
• Improved speech runtime status detection so missing runtimes and missing models are reported separately.
• Synced local and remote speech model manifests and clarified the Apple Silicon requirement for local TTS runtimes.

1.6.5

• Fixed the installed Piper runtime so its bundled dynamic libraries are found at launch.
• Split read-aloud error messages between missing models, Chinese text requiring Kokoro, and runtime launch failures.
• Added a release-time Piper bundle check so packages missing the required LC_RPATH fail before publishing.

1.6.4

• Added Piper as a local English read-aloud model option with bundled runtime support.
• Improved read-aloud controls with stop and settings actions in the floating player.
• Tightened TTS model download rows, Piper speed handling, and release asset publishing.

1.6.3

• Added a next-page control to the floating read-aloud player.
• Fixed PDF two-page read-aloud navigation so moving from the left page to the visible right page does not turn to the next spread.
• Added theme-matched SF Symbol icons to the bottom toolbar Shelf and Vocab buttons.

1.6.2

• Added an in-reader floating read-aloud controller with previous, pause/resume, next, and auto/manual advance modes.
• Improved read-aloud queue behavior so Next immediately stops the current sentence and starts the next one, while manual mode plays one sentence at a time.
• Kept the two most recent wav segments cached to make Previous sentence playback faster.
• Fixed layering and click handling conflicts between the floating player, PDF word highlights, AI hint bubbles, and page clicks.

1.6.1

• Reduced the bundled speech footprint by keeping only the runtime resources needed for local speech.
• Trimmed duplicate Kokoro English voices while keeping Bella, Heart, Adam, Emma, and George available.
• Kept local playback working with the smaller bundled runtime resources.

1.6.0

• Refined the reader, AI chat, settings, recent documents, and vocabulary surfaces with a shared theme palette.
• Added SF Symbol icons to the selected-text floating toolbar actions and tightened the toolbar layout.
• Improved AI chat bubble persistence, linked word handling, and source annotations across reading sessions.
• Made embedding lifecycle/status handling more resilient during backfill, rebuild, and document changes.
• Polished vocabulary review/list navigation and recent document cleanup flows.

Earlier versions are listed in GitHub Releases.

License

Leaf Reader is licensed under the Apache License 2.0.

Third-party speech models and runtimes remain copyrighted by their respective projects:

• FluidAudio Kokoro Core ML / Kokoro model: Apache License 2.0.
• Piper: MIT License; Piper voice model assets follow the metadata shipped with the upstream model package.

What's New in 1.7.10

• Added continuous PDF scrolling with reliable previous-page navigation, plus a paper-style table of contents and richer vocabulary learning progress.
• Made PDF search progressive and moved document indexing, vocabulary persistence, and visible highlight restoration off the interaction-critical path.
• Added content-addressed PDF text snapshots and prepared DOCX caches, with streaming DOCX parsing and cancellation when switching documents.
• Improved saved-word learning with lemma-and-part-of-speech grouping, related-form highlights, a visibility toggle, and direct access to full definitions.
• Fixed line-broken hyphenated PDF selections and strengthened Web Reader search, marks, and repeated-occurrence restoration.
• Added transactional user-data backup and recovery while keeping API credentials out of backup packages.
• Hardened untrusted document, archive, cache, and runtime handling, and avoided unnecessary credential access during ordinary reading.

What's New in 1.7.9

• Improved PDF read-aloud by learning repeated header and footer rows and skipping them during TTS.
• Page-number and footer rows are removed as whole visual rows, so mixed footer text is not spoken.
• Added launch performance diagnostics and delayed speech setup so everyday app startup stays lighter.
• Added focused regression coverage for PDF footer filtering, low-body-text false positives, and launch timing snapshots.

What's New in 1.7.8

• Improved reader theme coverage so chrome, toolbar, search, bookshelf, vocabulary, and AI surfaces consistently follow white, eye-care, and dark themes.
• Tightened release and documentation checks: local checks now build the manual in a temporary directory to avoid noisy generated HTML diffs.
• Cleaned up speech runtime release maintenance by removing obsolete runtime download wrappers and splitting Piper worker helpers into focused support code.
• Added regression coverage for theme palette colors and kept UI theme scanning warning-free.

What's New in 1.7.7

• Added a Difficult Sentence action for selected text in Reading Notes and for selected sentences in the main reader.
• Difficult sentence analysis now asks AI for sentence structure, subject/verb/object and clauses, layered translation, common expressions, and why the sentence is written that way.
• Refactored selected-text AI action handling to share the common summary, translation, and difficult sentence request flow.

What's New in 1.7.6

• AI conversation export now supports Markdown, HTML, and PDF, and the export button follows the current theme.
• Reading Notes export now uses one save panel with scope and file type selectors, without the macOS tags field.
• HTML and PDF exports render Markdown content instead of showing raw syntax, including headings, lists, and **bold** text.
• Renamed the Reading Notes export action to Export Notes and cleaned up shared export rendering coverage.

What's New in 1.7.5

• Added richer Reading Notes export options for Markdown, HTML, and PDF, including favorite-only export scope.
• Improved PDF read-aloud segment matching so repeated text and spoken sentence highlights land on the intended location.
• Fixed PDF line-broken hyphenated vocabulary words, so selections such as Nine-\ntenths are saved and queried as Nine-tenths.
• Tightened PDF vocabulary underline bounds to avoid overly long lines when a word crosses a line break.
• Updated code indexes and focused logic coverage for the new export and vocabulary behavior.

What's New in 1.7.4

• Faster vocabulary lookup from the floating selection menu: AI requests no longer block on slow local dictionary work, and dictionary metadata is filled in asynchronously.
• Tighter ECDICT behavior: multi-word selections skip local dictionary lookup, misses no longer fall back to table scans, and lookup keys are deleted with their main entries.
• Read-aloud navigation now focuses the active spoken sentence when using previous, next, replay, or their keyboard shortcuts.
• Selection floating menu text is larger for word and sentence actions.
• Cleaned up AI bubble, vocabulary persistence, read-aloud navigation, and dictionary lookup code, including removal of temporary performance tracing.

What's New in 1.7.3

• Reading Note images now persist through save, reopen, and AI organize/polish flows.
• Markdown handling in Reading Notes is cleaner and more stable, with dedicated parsers, render policies, image protection, and asset storage.
• Read-aloud now supports replaying the current segment and can provide the spoken sentence as AI context when no other text is selected.
• Vocabulary AI answers reuse local dictionary tags, show tags more compactly, and fall back to local dictionary data more reliably.
• AI bubbles, Reading Note typography, underline colors, and related UI details were polished for consistency.

What's New in 1.7.2

• Improved Reading Notes with a save button, copy/paste shortcuts, and fixes for reopened styling and word-count refresh.
• Fixed read-aloud continuation edge cases across manual/automatic advance and next-page flows for more reliable PDF/Web playback.
• Refactored Reading Notes, read-aloud, document, AI, and embedding state boundaries to reduce maintenance risk.

What's New in 1.7.1

• Improved offline and unconfigured-model selection behavior with local ECDICT lookup, speech, copy, and model settings states.
• Made vocabulary review default to frequency-first ordering and kept dictionary lookup scoped to the current card.
• Refactored network/model capability checks, selection toolbar configuration, local dictionary lookup, and review-card loading.

What's New in 1.7.0

• Added easier access to Reading Notes, with improved note editing, Ask AI prompts, paste handling, and disabled note-editor context menus.
• Improved read-aloud controls with clearer loading/pause/stop icons, a speech-speed slider, and better floating player layout.
• Added vocabulary review priority choices and refined the review UI, scoring, and speech playback internals.

What's New in 1.6.6

• Lowered the Piper runtime minimum macOS marker to macOS 12.0 and added checks to prevent mismatched bundles.
• Improved speech model status text for missing runtime versus missing model cases.
• Synced speech model manifests and clarified local TTS compatibility.

What's New in 1.6.5

• Fixed Piper read-aloud startup in installed builds by bundling the required dynamic library search path.
• Improved read-aloud error messages for missing models, Chinese-only Kokoro requirements, and runtime launch failures.
• Added release checks that catch malformed Piper runtime bundles before publishing.

What's New in 1.6.4

• Added Piper as a local English read-aloud model option with bundled runtime support.
• Improved read-aloud controls with stop and settings actions in the floating player.
• Tightened TTS model download rows, Piper speed handling, and release asset publishing.

What's New in 1.6.3

• The floating read-aloud player now includes a next-page button.
• In PDF two-page mode, next-page read-aloud moves from the left page to the visible right page without turning the spread.
• The bottom toolbar Shelf and Vocab buttons now include SF Symbol icons.

Development

Requirements

• macOS 12.0 Monterey or later.
• Swift toolchain with Cocoa, PDFKit, WebKit, and CryptoKit frameworks.
• Sparkle for release builds.

Build From Source

Install Sparkle first:

brew install --cask sparkle

Build and run the app:

./scripts/build_app.sh
open "Leaf Reader.app"

Daily builds default to --debug --arm64 for faster iteration. Use ./scripts/build_app.sh --release --universal when you need a local release-style universal app; release packaging already does this automatically. Use ./scripts/audit_app_bundle.sh to inspect the generated app, speech runtime sizes, symlinks, and largest bundled resources.

Tests

Run lightweight logic regression tests:

./tests/run.sh

Run the full local pre-commit check:

./scripts/check.sh

Speech Model Packages

Generate speech model packages with:

./scripts/package_speech_models.sh

The packaging script also writes docs/tts/speech-models-manifest.json with each asset's file size and SHA256 digest. Publish with --with-speech-models only when the model archives change.

Project Layout

• Leaf Reader.app - generated macOS application bundle, ignored by git.
• mac-app/*.swift - native Swift source code.
• tests/ - lightweight Swift logic regression tests.
• docs/ - GitHub Pages site, manual, and Sparkle update feed.
• assets/ - README icon and screenshots.
• release/ - local release artifacts when generated.

Code Wiki

Developer notes live in docs/wiki/:

• Code Wiki index
• Code Map

Regenerate the code map after larger refactors:

./scripts/generate_code_wiki.sh

Release

Current version: 1.7.10

Git tag: v1.7.10

Latest installer:

Leaf Reader-1.7.10.pkg

Local release package path:

release/1.7.10/LeafReader-1.7.10.pkg

Build the signed release package without publishing:

./scripts/release_pkg.sh 1.7.10

Run the full publish flow from a clean working tree:

./scripts/publish_release.sh 1.7.10

The publish script runs tests, builds/signs/notarizes the pkg, smoke-tests the package payload, reports speech runtime size, commits version/appcast changes, tags the release, pushes main and the tag, creates the GitHub Release, uploads the pkg, and verifies the download URL. Pass --with-speech-models only when publishing changed speech model archives in docs/tts/. Add --push-wiki --cleanup-releases to sync GitHub Wiki and clean old ignored local release artifacts after publishing.

Notes

• Bundle identifier: com.linlu.leafreader.
• Automatic updates use Sparkle and the public EdDSA key embedded in mac-app/Info.plist.
• PDF rendering uses PDFKit.
• EPUB and DOCX rendering uses WebKit. DOCX support is optimized for readable text extraction rather than exact Word layout fidelity.
• Search selections are kept separate from AI passage selection so search navigation does not accidentally populate the assistant.
• AI requests use the model, endpoint, language, and API key configured locally in the settings panel.