v1.7.32026-11-15
DeepTutor v1.7.3(2026.11.16)在上一版基础上,重点提升了阅读与书籍编辑的灵活性,并扩充了搜索与 Agent 连接能力。
PDF 阅读器新增连续缩放、文字适配与图片查看,标注与多素材选择更稳定;阅读测验的答案键改存服务端,题目关联的帮助记录与作答记录分开保留,监护人可查看阅读进度与求助情况。书籍编辑支持章节局部修改,结构变更与页面对话建议在应用前会展示可审阅的差异对比,最近一次未变更的编辑可撤销。
搜索接入 DeepSeek、SpecStream、SearchApi 三种来源,Web Fetch 增加 DeepSeek 后端,Standard Compute 加入模型配置;已配置的 ACP 进程可作为本地子代理调用,MCP OAuth 刷新与重连更可靠。对话可按会话设置推理强度,题库来源支持搜索与分页,掌握度修复保证失败尝试后解析不丢失。文档解析方面,Docling 保留结构化页面内容并可选启用公式与图片增强,GitHub 知识源支持 MDX,界面新增意大利语、匈牙利语,印地语加入回复语言。模型流处理对停滞流、不完整终止事件等异常更稳健;桌面端升级会自动采用新版内置核心并支持备份回滚,Windows 构建仍未签名。
v1.7.22026-11-06
DeepTutor v1.7.2(2026.11.06)在 v1.7.1 基础上补齐了桌面端下载渠道,新增选中文字朗读,并强化了引导学习、检索与更新的稳定性。
桌面端发布物现在包含 Apple Silicon 与 Intel 两个 DMG、Windows 安装包及便携版 ZIP,发布流程会逐一校验各平台产物与稳定别名;Windows 侧修复了会话与迁移锁、插件环境、Python 可执行文件查找、非 UTF-8 输出及含空格路径等问题。
在 Chat、Mastery Path 或阅读界面选中文字即可调用 Read aloud 朗读,Chat 朗读会按会话记住播放速度;语音渲染对转义百分号和度数符号的处理更自然,Video Lesson 的 Edge 回退会按课程语言选声。
模型配置更省事:修改凭据后设置页会自动发现可用模型,支持批量导入,新建自定义 provider 默认自动选择协议;NVIDIA NIM 新增文本检索嵌入,DashScope 新版图像模型走对应生成协议,4K 仅限 wan2.7-image-pro,Responses API 工具参数、结构化输出、历史回放与 Codex 自定义请求头也做了兼容修复。
引导学习方面,Mastery Path 大纲生成前可先选定模型,追问与缺失部分会结合其补全的解释来评估,过期卡片回复与重复重试的处理更谨慎;来源引用不匹配会做有限次重试,掌握度导航、看板滚动与辅导记忆恢复更可靠。
界面新增韩语部分翻译(未译文本回退英文),更多标签与日期跟随所选语言;Little Tutor 会话进入命令面板,文档阅读器按需加载,socket 失败与取消恢复更可预期;BM25 保留带点标识符,文档抽取移出 API 事件循环,笔记与收藏工具支持只读记录检索。
更新更安全:包更新会把已装功能模块对齐到目标核心版本并记录启动失败,不再卡在重启;分离启动避免引入陈旧源码树,CLI 应用回滚失败时保留备份。共享 agent 循环在推理耗尽输出预算且无答案时可停止,或做一次显式限额的高预算重试。
升级前请停掉 DeepTutor 并备份运行时 home,执行 python -m pip install --upgrade "deeptutor[full]==1.7.2",保持同一 DEEPTUTOR_HOME,用 python -m pip check 与 deeptutor module list 核验后再重启,附加功能 wheel 也须同为 1.7.2。1.6.15 及更早版本需先卸载旧 deeptutor-cli;源码安装走 requirements/full.txt 加 python -m pip install -e . --no-deps;Docker 拉取 ghcr.io/hkuds/deeptutor:1.7.2 并复用数据卷重建容器;桌面端请使用本 Release 对应安装包或便携 ZIP,Windows 构建仍未签名。
v1.7.12026-10-31
DeepTutor v1.7.1 于 2026.11.01 发布,在 v1.7.0 基础上带来 Open DeepTutor 桌面预览版,将 Python、Node.js 与完整学习工作区打包为应用,覆盖 Apple Silicon、Intel Mac 与 Windows x64,可从 Desktop Release 工作流获取 DMG 或 Windows 安装包。
界面新增越南语、日语、土耳其语、巴西葡萄牙语,并完善日期格式与从右向左排版;支持在个人资料页修改密码,管理员分配的临时密码会强制提示更换。聊天方面,未发送草稿在页面跳转后得以保留,消息级 Little Tutor 线程可恢复或移动来源锚点,生成文件统一提供打开与下载入口。
课程与学习材料同样有实质改进:Video Lesson 接入 TTS 播报并自动检测修复版面,Math Animator 增加几何预检,笔记闪卡可导出 Anki,日历事件支持 iCalendar。模型与检索侧扩展了 xAI、Mistral、You.com 等接入,PageIndex 索引可调用任务模型,MinerU 保留参考文献列表,Office/DOCX 解析对公式与图片的还原更完整。此外还加固了压缩包解压、Unicode 文件名、取消清理等运行时环节,操作失败时不再损坏既有数据。
升级前请停止服务并备份运行时目录,完整版执行 python -m pip install --upgrade "deeptutor[full]==1.7.1",保持 DEEPTUTOR_HOME 不变;从 1.6.15 及更早版本升级需先卸载旧的 deeptutor-cli 包。Docker 用户拉取 ghcr.io/hkuds/deeptutor:1.7.1 重建容器即可,运行时数据无需迁移。
v1.7.02026-10-28
DeepTutor v1.7.0 Release Notes
Release Date: 2026.10.28
Building on v1.6.15, v1.7.0 brings modular installation, a refreshed learning workspace, block-based notebooks with Dito, and narrated video lessons. Full-app upgrades now use deeptutor[full]; existing runtime data remains separate from the installed packages.
What's New
Install the parts you use
deeptutor now provides core chat, CLI, and SDK; deeptutor[full] selects the Web app and standard feature modules. The deeptutor-cli package becomes a compatibility alias. Inspect installed, enabled, and ready states with deeptutor module list, then enable, disable, install, or remove modules while the app is stopped. Dependency checks protect required modules, and removal preserves user data.
A workspace organized around learning
Modern and Classic layouts offer different navigation arrangements. Home brings together Chat, the Task Board, Learning notebook, and Calendar; Learning and Resources keep study activities and materials accessible. Customize Chat modes, controls, and starting prompts, or choose which learning methods appear on the New learning page. Conversation and resource navigation retain their workspace boundaries.
Notebook pages with Dito beside them
The Learning notebook replaces the former Co-Writer editor with pages, subpages, and blocks. Import Word documents, insert source records from Favorites, discuss a passage with Dito, and review proposed edits before accepting them. Export pages as Markdown or Word. Favorites hold saved source records; notebook pages hold your own writing and understanding.
Video lessons and free read-aloud
Choose Video Lesson to generate a lesson that plays directly in the browser, with captions and optional Microsoft Edge narration. Frame checks execute generated code inside a bounded WebAssembly interpreter; Docker is not required. Microsoft Edge also joins the read-aloud providers, and reply playback is more reliable on iOS.
More reliable tutoring and indexing
Compatible interrupted LlamaIndex rebuilds can resume completed embedding checkpoints. Knowledge-base creation retains same-named documents, and one failed caption group no longer aborts the entire ingestion. Chat recovers worker failures more consistently, completed Anthropic thinking is excluded from history replay, and prompt caching uses steadier conversation prefixes. Guided Learning preserves mathematical meaning when checking citations and keeps review work within its eligibility gate.
English and Chinese guides are synchronized with this release at deeptutor.info.
Upgrade Notes
- Full app: stop DeepTutor and back up the runtime home, then run python -m pip install --upgrade "deeptutor[full]==1.7.0". Run python -m pip check and deeptutor module list before restarting with the same DEEPTUTOR_HOME. Bare deeptutor now selects core only.
- Legacy CLI package: first run python -m pip uninstall deeptutor-cli, then install the new core or full profile. Optionally reinstall deeptutor-cli==1.7.0 as the compatibility alias. The old CLI owns shared Python files, so upgrading it in the same transaction can remove files needed by the new core. Package removal leaves runtime-home data intact.
- Source installs: install requirements/full.txt, then python -m pip install -e . --no-deps. Use python scripts/install_extras.py <extra> for optional source dependencies. Source checkouts support module enable/disable; package install/uninstall requires a wheel environment.
- Docker: pull ghcr.io/hkuds/deeptutor:1.7.0 or :latest and recreate the container with the existing data volume.
- Video Lesson: the backend needs Node.js 22+ and npm; first use downloads pinned rendering packages. Install the edge-tts extra for narration. Without available narration, lessons retain their captions and play silently.
- Module versions: core and feature wheels must match. Stop active servers and CLI turns before changing modules; inspect dependency blockers rather than uninstalling packages beneath a running app. See module installation and migration.
Full Changelog: https://github.com/HKUDS/DeepTutor/compare/v1.6.15...v1.7.0
v1.6.152026-10-18
DeepTutor v1.6.15 于 2026 年 10 月 19 日发布,重点打磨引导式学习与会话可靠性。学习路径支持调整顺序前先预览,并可跨模块一次性应用改动;提问改为两至四题成组,支持分次提交、逐题独立评分,基于原文的语义判定能识别等价表述,区分答错、答不全与存疑,但概念卡与设计卡仍只算练习,掌握仍以学员自己的解释为准。长会话中引导式学习会在有限记忆里保留教学偏好、未解决的误解与原文锚点,用量统计也把资料准备和记忆压缩计入,压缩单独计费。聊天流会定期落盘以便恢复,未发送消息保留可恢复副本,设置写入失败后不丢草稿并给出可操作的错误提示。MinerU CLI worker 明确进程归属与超时,单个 PDF 解析失败不再中断整批索引,恢复提示能区分可续用的解析与需重跑的向量化。检索侧新增可选的查询改写层,配合 deeptutor kb eval --rewrite off|standalone 与 eval-diff 命令做对比实验;工作区指令可约束后续对话(Deep Research、Quiz、Manim 不适用)。Gemini TTS 改用原生 generateContent 语音,另提供可选的 guard 扩展,用于限流、地址与请求模式筛查,默认关闭且不检查报文体与 WebSocket。升级执行 pip install -U deeptutor,Docker 拉取 ghcr.io/hkuds/deeptutor:latest 并重启服务。
v1.6.142026-10-08
DeepTutor v1.6.14(2026.10.08)在 v1.6.13 基础上带来几项重点改进:知识库支持直接提问原文中的某张图、表或 PDF 页面,系统会取回原图或渲染指定区域,并在文件视图中标注图片来源覆盖情况;索引流程改为可恢复,重试会复用已完成的解析与云端切片,重建失败时仍保留旧版可查询的向量库;视频学习并入 Reading 工作区,YouTube 与 Bilibili 支持带时间戳的讨论和原生播放,旧 Watching 会话可在重新打开时迁移。
科研方面新增版本化 arXiv 论文阅读、按固定 GitHub revision 核对论文论断、构建文献证据矩阵等工具,MCP 目录加入 alphaXiv。插件 CLI 补齐安装、权限审批、启停、回滚与卸载的完整生命周期,官方目录本期为空,不会自动装入第三方代码。学习空间新增学习日志,记录当前任务与上次衔接。另修复聊天重试、草稿恢复、分支导航等问题,Visualize 增加 Mermaid 思维导图。升级用 pip install -U deeptutor,Docker 拉取 ghcr.io/hkuds/deeptutor:latest。
v1.6.132026-10-04
DeepTutor v1.6.13(2026.10.04)继续扩展阅读、听力与题库练习:沉浸阅读加入朗读语音和答题星级奖励,沉浸观看支持时间戳标记与复习提示;题库可直接发起练习、多选题目组聊,选择题选项随机打乱,不再借选项字母泄露答案。文档检索可单独指定图片描述所用的视觉模型并按工作区缓存,MinerU 云端解析能切分超大 PDF、续传已完成部分,新增 deeptutor kb eval 检测检索质量。语音合成新增超时配置,预览报错会区分鉴权、配额、网络与超时等原因。聊天附件改存所选工作区的 chat/attachments,旧附件按需迁移,原 URL 仍可读取。界面新增波兰语,接入 Requesty、API Route、FutureInfra 等供应商,支持 MiniMax 原生语音与 MiMo v2.5 预置语音。升级运行 pip install -U deeptutor;合成偏慢时可把超时从默认 60 秒调到 5–600 秒之间。
v1.6.122026-09-27
DeepTutor v1.6.12 Release Notes
Release Date: 2026.09.27
Building on v1.6.11, v1.6.12 lets each workspace place and move its knowledge bases, connects searchable Kiwix archives, and carries source figures into grounded answers. A Task Board, German interface, and recovery fixes round out the release. Existing data needs no migration; systemd installs now update outside the app.
What's New
Knowledge bases follow the workspace
Choose the storage workspace when creating a knowledge base. Move one existing base from its Settings panel after reviewing destination blockers and affected assignments; DeepTutor verifies the copied data and index before switching references. Saved knowledge-base IDs and workspace assignments continue to resolve. Moves are blocked while processing or when shared grants, destination collisions, or unsupported MarginNote data make them unsafe.
Connect Kiwix archives
Connect one searchable ZIM archive served by kiwix-serve as a knowledge base from the Knowledge Center or deeptutor kb connect-kiwix. Search reads the archive on demand without uploading or indexing a copy. You can also bring an article into Immersive Reading. Removing the connection leaves the archive untouched.
Source figures in grounded answers
The LlamaIndex knowledge pipeline can retain parser-extracted PDF and EPUB images alongside their captions and source locations. When retrieval finds one, a vision-capable answer model sees the verified pixels; text-only models receive the caption and context with an explicit limit. Immersive Reading can inspect a stored figure on demand. This depends on a parser that emits images and reindexing older content; see Upgrade Notes.
Task Board and learning controls
A workspace-local Task Board tracks To do, In progress, and Done cards with notes, accessible move controls, and a restorable archive. Guided Learning now respects an explicit teach-first preference and gives a localized next step. Teacher and student join the accepted non-admin role values without gaining admin privileges.
Languages and providers
German joins the interface languages, and Malay becomes a reply-language option. Cheaper Inference joins LLM providers; local Lemonade embeddings work without an API key. DashScope image, video, and speech requests use supported endpoints, sizes, and model IDs; saved DashScope paraformer-v2 speech profiles migrate to paraformer-realtime-v2 automatically.
Reliability and accessibility
Failed chat submissions remain recoverable across retries and storage races, and unanswered turns can recover from a lost worker. LightRAG reuses only compatible indexing caches, retrieved figures stay within bounded model context, and partial Deep Research and Manim failures report their cause. Shared tooltips replace inaccessible native title hints across learning and workspace controls, including touch interactions.
Upgrade Notes
Run pip install -U deeptutor; Docker users pull ghcr.io/hkuds/deeptutor:latest. Existing data needs no manual migration.
- Kiwix: run a reachable kiwix-serve instance with full-text search for the chosen ZIM, then connect its server URL and archive name. Disconnecting only removes the DeepTutor pointer.
- Older knowledge-base figures: reindex existing LlamaIndex source files to create visual records. MinerU and PyMuPDF4LLM can emit supported images; the default text-only parser and current Docling adapter do not. Other RAG engines do not index the visual records yet.
- systemd installs: in-app update handoff is disabled for a running service. Stop the service, upgrade DeepTutor in that service's Python environment, and start it again with systemctl.
- Windows microphone transcription: native DashScope and Volcengine speech adapters need ffmpeg on the service PATH to convert browser WebM/Opus audio; restart DeepTutor after installing it. OpenAI-compatible STT does not need local conversion.
Full Changelog: https://github.com/HKUDS/DeepTutor/compare/v1.6.11...v1.6.12
v1.6.112026-09-24
DeepTuber v1.6.11 是一次大更新:界面新增法语与乌克兰语,回复语言扩到 14 种,输入框的 /language 可单独锁定当前会话的回复语言。沉浸式阅读重做——收藏变为彩色文件夹,新增全屏学习模式;选中段落可直接「解释」「翻译」「带我读」,引用作为普通消息留在对话历史中,另有针对当前页的三题小测,EPUB 也多了字体、行宽、双页等显示选项。DOCX、PPTX、PDF 中的插图会被提取并转交视觉模型,矢量页栅格化处理,图注可选解析;已有材料不会自动重提取。Office 文件借助 LibreOffice 以分页 PDF 预览。知识库支持文档站定时重同步(默认 24 小时,可按源设置 1–168 小时)与本地文件夹「立即同步」,并新增 .bib 索引、zotero_search 工具和扫描件本地 OCR 兜底。另有学习进度页、CSV 批量导入用户等改动。需注意:镜像因 LibreOffice 变大,且 /app/data 不可写时容器会拒绝启动。
v1.6.102026-09-22
v1.6.10 给原生 LightRAG 引擎配了独立的角色模型,抽取、关键词、查询、视觉可分别指定,视觉默认关闭。索引发布后会固定当时所用的抽取与视觉模型,之后改引擎默认值不再波及已有索引;嵌入配置一旦变动,查询与增量写入即被拦下,需恢复原配置或整体重建,失败的重建会撤回发布而非留下半成品。PDF 附件改走「设置 → 知识库」里配置的