Smart Question Tutor

Smart Question Tutor — 智学助手

AI-powered, spaced-repetition study plugin for Obsidian

English | 中文

---

What is it?

Smart Question Tutor turns your Obsidian Markdown notes into an active-study machine. It reads your local knowledge base, uses AI (local Ollama or any OpenAI-compatible API) to generate quiz questions, records wrong answers into a spaced-repetition review loop, builds a knowledge-tag graph, and lets you export to Markdown / Word / PDF.

Features

AI Question Generation

• Generate quizzes from any Markdown note with 5 question types: single choice, multiple choice, true/false, fill-in-the-blank, short answer.
• Supports Ollama (local, private) and OpenAI-compatible endpoints (DeepSeek, etc.) via custom base URL, model, API key and temperature.
• AI auto-extracts knowledge tags and writes [[wikilinks]] so everything appears in your Obsidian graph view.
• Customize per-type counts, run in batch over multiple files, and sort results by source / tag / date.

Full Exam Recognition

• Extract every question from a complete exam paper or question bank via AI, preserving original question types and formatting.
• Large files are chunked (15,000 chars per chunk, 2,000 overlap) to avoid truncation.
• Existing answers are kept as-is; missing answers are auto-generated by the AI.
• Handles Markdown, txt, RTF, docx, PDF and images.

Answering Mode

• Start answering any generated result or saved question file in one click.
• Instant grading with the correct answer and explanation.
• Objective questions are auto-scored; subjective questions can be manually marked correct/wrong — missed ones go straight to the wrong-answer book.

Wrong Answer Book (SM-2 spaced repetition)

• Answer correctly → interval advances; answer wrong → resets to day 1.
• A single difficulty factor (SM-2 EF) per module controls interval growth, with Fewer reviews / Standard / More reviews one-click presets and a recommended range.
• Wrong answers carry [[knowledge tags]] and sync into the knowledge folder for the graph.
• Automatic review reminders when the plugin loads.

Study Notes

• Generate condensed study notes from any Markdown file, any existing question/wrong-answer/note.
• Notes participate in the same spaced-repetition review schedule.
• Search, filter and export.

Review Dashboard

• One unified list of every due review item (wrong answers + questions + notes), filterable and sortable by source / tag / date.
• One-click Done per item advances it to the next review cycle.

Learning Heatmap

• GitHub-style annual contribution graph on the Home tab.
• Horizontal scrollbar, auto-positioned to today; follows the sidebar width, always staying pinned to the current day.
• Color intensity tiers by activity count.

Knowledge Graph

• A single knowledge-tag index folder; each tag gets an auto-generated MOC note with 相关题目 / 相关笔记 / 相关错题 sections.
• Rebuild the index on demand: it scans every folder's tags, recreates all index notes, and cleans up stale ones in one pass.
• Handwritten notes in the knowledge folder are never touched.

Export

• Export to Markdown, Word (.docx) or PDF.
• Answer-free exports are available for self-testing.

💬 AI Assistant Chat

• A built-in chat panel embedded in the Smart Question Tutor sidebar (the AI tab).
• Workspace & multi-chat management: organize chats into workspaces, each containing many independent conversations (switch / create / rename / delete via the header dropdowns and the ☰ drawer). Old single-history data migrates into a default workspace automatically.
• Answers are grounded in your notes via retrieval; each conversation remembers its own scope between plugin knowledge base only and entire vault.
• Reference notes: attach the current selection/note (or pick multiple notes) as context for the answer; references show as removable chips and are cleared automatically after sending.
• The reference budget setting caps how much referenced text is fed to the model (default 60000; lower it for small local models). The budget is shared fairly across references: short files are included whole, long files keep their opening plus a few query-relevant snippets, and cuts land on natural boundaries.
• Replies render as Markdown; each message can be copied or regenerated.

🔒 Privacy

• Local-first: with Ollama, nothing leaves your machine.
• When you use an OpenAI-compatible endpoint (including the built-in chat), the selected context is sent to that endpoint:
  • Question generation sends the chosen source text / exam file.
  • Chat scope plugin knowledge base sends matched snippets; scope entire vault may send matched note contents from any folder.
  • Referenced notes are sent in full (up to the reference budget).
• No telemetry and no hidden network calls. Only the endpoint you configure is contacted.

🌐 Bilingual UI (Chinese / English)

• One-click interface language switch in Settings — the whole UI flips between 中文 and English instantly.
• The switch is also available right inside the sidebar settings; command palette names update after a plugin restart.
• AI prompts are bilingual too while keeping the "language follows the material" rule: Chinese material still produces Chinese questions, English material produces English questions.
• English parsing & export support (Answer:, Explanation:, Answer Summary, ...) — English-material exams parse and export correctly.
• Missing translations fall back to the original Chinese text, so the UI never goes blank.

---

Repository

GitHub: https://github.com/xxinjie21/Smart-Quiz-Tutor

---

Installation

From Obsidian Community Plugins

1. Open Settings → Community plugins → Browse
2. Search for Smart Question Tutor
3. Click Install, then Enable

Manual Installation (GitHub Release)

Download the main.js, manifest.json and styles.css from the latest Release, then copy them into:

your-vault/.obsidian/plugins/smart-quiz-tutor/
├── main.js
├── manifest.json
└── styles.css

Then enable the plugin in Settings → Community plugins → Installed plugins.

---

Quick Start

1. Configure AI — open Settings → Smart Question Tutor; choose Ollama or OpenAI and set the API URL, model, and API key (if needed).
2. Generate questions — click the ribbon icon to open the sidebar; in the 题目 tab pick a Markdown source file, choose question types & counts, click Generate.
3. Start answering — click 开始答题 after generation finishes.
4. Review mistakes — wrong answers are auto-saved with spaced-repetition scheduling; open the 错题 tab to review now.
5. Track progress — check the 复习 tab for due items, or read the heatmap on 首页.

---

Commands

Hotkeys

No default hotkey is bound. Bind one in Settings → Hotkeys (for example Ctrl+Q to 基于当前文档直接生成试题).

---

Settings

Core

Review Interval Factor

Intervals are controlled by a single difficulty factor (SM-2 EF) per module. SM-2 grows the interval as 1 day → 6 days → previous × factor; the factor is then nudged by each review's grade. Recommended 2.0–2.8 (standard 2.5); input is clamped to 1.3–3.0.

---

Interface

The sidebar has 7 tabs:

---

Tech Stack

---

Changelog

v2.3.0

• Real SM-2 spaced repetition: per-item ease factor (EF) and repetition tracking; review with four grades (Forgot / Hard / Good / Easy). Each module has a single interval factor setting (default wrong 2.3 / question 2.5 / note 2.5, recommended 2.0–2.8) with Fewer reviews / Standard / More reviews one-click presets; intervals grow as 1 day → 6 days → previous × factor.
• Manual wrong-answer entry: add a wrong answer by hand (content / source / knowledge tags / note) from the 错题 tab.
• Answer-only export: export a questions file (or a wrong answer) containing only the answers and explanations.
• Chat workspaces: the AI chat now supports workspaces and multiple independent conversations per workspace; each conversation remembers its own retrieval scope (old single history migrates automatically).
• Local-date fix: review scheduling now uses your local timezone instead of UTC (no more off-by-one due dates).
• Settings single source of truth: the native settings tab and the in-sidebar settings are both rendered from one schema, so they can no longer drift apart.
• Cleanup: removed two unused dependencies; the hardcoded global Ctrl+Q listener is replaced by a normal command (bind your own hotkey).
• 307 tests pass; tsc 0 errors; ESLint 0 errors / 0 warnings.

v2.2.0

• Built-in AI chat in the Smart Question Tutor sidebar (AI tab): retrieval over your notes, multi-file references (from the active note/selection or a file picker), a configurable reference budget, Markdown-rendered replies, copy / regenerate, and "stick-to-bottom" scrolling.
• Robust retrieval: CJK-aware tokenization (2/3-gram), filename/path weighting, and two-stage candidate ranking (only matched files are read).
• Chat polish: inline error/warning bubbles, token-size warning, Ctrl/⌘+Enter to send, auto-focus on open.
• Reliability: idempotent folder creation (no "Folder already exists" on startup), friendly network-error messages, history pruning (kept ≤ 100 entries, results truncated), TTL caches for question/note/wrong lists.
• Maintainability: sidebarView.ts split from ~3700 lines into 13 focused modules under src/views/sidebar/; data/list/export logic moved to VaultDataService; all brittle [style*=] CSS selectors replaced with semantic classes.
• Rebuilt main.js; CI now runs build + unit tests; a guard test asserts no eval / no script injection.
• 194 tests pass; tsc 0 errors; ESLint 0 errors.

v2.1.0

• One-click Chinese/English language switch — ~500 UI strings across all tabs now go through a zh/en dictionary; switch instantly from Settings (or the in-sidebar settings).
• Bilingual AI prompts — question generation, exam extraction, note generation and tag suggestion support both languages, keeping the "language follows the material" rule.
• English parsing & export support — the question parser and exporter now recognize Answer: / Explanation: / Answer Summary etc., so English-material questions parse and export correctly.
• Dictionary fallback — missing translations show the original text, so the UI never goes blank.
• Made the dynamic-script strip regex minifier-agnostic, keeping createElement("script") at 0 for the Obsidian review lint.
• 153 tests pass (new: dictionary integrity, placeholder replacement, bilingual prompts, English parsing); tsc 0 errors, ESLint 0 errors.

v2.0.1

• Removed the remaining dynamic <script> injection (from the setimmediate dependency) that Obsidian's review lint flagged as an error — createElement("script") went from 3 to 0, unblocking community review.

v2.0.0

• Knowledge-index rebuild is now resilient — a single failed tag write no longer aborts the whole rebuild; stale index files are cleaned up in one pass.
• Review progress persists across sessions — interval / correct-count are stored as numbers, so your spaced-repetition streak is no longer reset on reload.
• Responsive heatmap — uses a scrollbar, auto-positions to today, and stays pinned to today as you resize the sidebar.
• Added comprehensive test suites (knowledge index, document service, note service, fixes).

---

Support

If this plugin helps your studies, consider giving it a star on GitHub!

GitHub: https://github.com/xxinjie21/Smart-Quiz-Tutor

For issues and feature requests, please open an issue.

---

License

ISC License

---

这是什么？

智学助手 把你的 Obsidian Markdown 笔记变成一台主动学习机：读取本地知识库，用 AI（本地 Ollama 或任意 OpenAI 兼容接口）生成试题，把错题纳入间隔重复的复习循环，构建知识点图谱，并可导出 Markdown / Word / PDF。

功能

🧠 AI 试题生成

• 基于 Markdown 笔记原文，AI 自动出题，支持 5 种题型：单选 / 多选 / 判断 / 填空 / 简答。
• 支持 Ollama（本地、离线、隐私）与 OpenAI 兼容接口（如 DeepSeek 等），可自定义接口地址、模型、API Key、Temperature。
• AI 自动提取知识点标签并写入 [[双向链接]]，融入 Obsidian 图谱视图。
• 每种题型可独立设置数量，支持多文件批量生成，结果可按源文件 / 知识点 / 时间排序。

📄 整卷识别

• 从完整试卷 / 题集中 AI 提取全部题目，保留原题型与格式。
• 大文件自动分块处理（每块 15000 字符、重叠 2000 字符），避免内容截断。
• 原卷已有答案原样保留，缺答案由 AI 自动补全。
• 支持 Markdown、txt、RTF、docx、PDF 与图片。

✍️ 答题模式

• 一键开始对刚生成的结果或已保存的题目文件答题。
• 即时批改，展示正确答案与解析。
• 客观题自动计分，主观题可手动标记正误；答错自动进入错题本。

📘 错题本（SM-2 间隔重复）

• 答对 → 间隔递增；答错 → 重置为第 1 天。
• 每个模块用一个**难度因子（SM-2 的 EF）**控制间隔增长，附带 少复习 / 标准 / 多复习 一键预设与推荐范围。
• 错题携带 [[知识点标签]]，自动同步到知识点文件夹，图谱可见。
• 插件启动时自动提醒到期复习。

📓 学习笔记

• 从任意 Markdown 文件，或已有题目 / 错题 / 笔记一键生成浓缩知识点笔记。
• 笔记同样纳入间隔重复复习体系。
• 支持搜索、筛选、导出。

📊 复习看板

• 统一展示所有到期复习项（错题 + 题目 + 笔记），可按源文件 / 知识点 / 时间筛选排序。
• 每项一键「已完成」，自动推进到下一复习周期。

📈 学习热力图

• 首页 GitHub 风格年度活跃图。
• 横向滑动条，自动定位到当天；随侧边栏宽度自适应，始终钉在当天。
• 颜色按活动频次分级。

🔗 知识点图谱

• 统一的知识点索引文件夹；每个知识点自动生成 MOC 索引笔记，含 相关题目 / 相关笔记 / 相关错题 三个分区。
• 支持一键重建知识点索引：扫描各文件夹标签、重建全部索引、并一次性清理过期索引。
• 知识点文件夹中的手写笔记永远不会被误删。

📤 专业导出

• 支持导出 Markdown、Word (.docx)、PDF。
• 可导出无答案版用于自测。

💬 AI 助手对话

• 内嵌在智学助手侧边栏的 AI 页签里。
• 工作区 + 多会话管理：用工作区分区，每个工作区内可建多个独立会话；通过头部下拉与 ☰ 抽屉新建 / 切换 / 重命名 / 删除。旧版单一聊天历史会自动迁移为一个「默认」工作区里的会话。
• 回答基于你的笔记检索结果；检索范围可在 仅插件知识库 与 整个 vault 间切换，且每个会话独立记忆各自的检索范围。
• 引用笔记：把当前选区/笔记（或从文件选择器挑多篇）挂为回答上下文；以可移除的 chip 展示，发送后自动清空。
• 设置项「引用总预算(字)」限制带入模型的引用文本量（默认 60000，本地小模型建议调小）。预算在各引用间公平分配：短文件整篇带上，长文件保留开头并补充与提问相关的片段，截断落在自然边界。
• 回复按 Markdown 渲染；每条消息可复制、可重新生成。

🔒 隐私说明

• 本地优先：使用 Ollama 时全部在你本机完成。
• 使用 OpenAI 兼容接口（含内置对话）时，会把所选上下文发送到该接口：
  • 生成试题：发送你选择的源文本 / 试卷文件。
  • 对话范围「仅插件知识库」：发送命中的片段；「整个 vault」：可能发送任意文件夹中命中的笔记内容。
  • 被引用的笔记会完整发送（受引用预算限制）。
• 无遥测、无隐藏网络请求；只会访问你配置的接口地址。

🌐 中英文双语界面

• 设置页一键切换界面语言，整个界面在 中文 / English 间即时切换。
• 侧边栏设置内也可直接切换；命令面板中的命令名在重启插件后更新。
• AI 提示词同样双语化，同时保留「语言与材料一致」规则：中文材料仍出中文题，英文材料出英文题。
• 英文解析与导出兼容（Answer:、Explanation:、Answer Summary 等）——英文材料的试卷可正常解析与导出。
• 漏翻的文案自动回退显示中文原文，界面永不空白。

---

仓库地址

GitHub: https://github.com/xxinjie21/Smart-Quiz-Tutor

---

安装

从 Obsidian 社区插件安装

1. 打开 设置 → 社区插件 → 浏览
2. 搜索 Smart Question Tutor
3. 点击 安装，然后 启用

手动安装（GitHub Release 下载）

从最新 Release 下载 main.js、manifest.json 和 styles.css，放入：

your-vault/.obsidian/plugins/smart-quiz-tutor/
├── main.js
├── manifest.json
└── styles.css

然后在 设置 → 社区插件 → 已安装插件 中启用。

---

快速开始

1. 配置 AI — 打开 设置 → Smart Question Tutor，选择 Ollama 或 OpenAI，设置接口地址、模型、API Key（如需要）。
2. 生成题目 — 点击左侧栏图标打开侧边栏，在「题目」Tab 选择 Markdown 源文件，选择题型与数量，点击「生成」。
3. 开始答题 — 生成完成后点击「开始答题」。
4. 复习错题 — 错题自动按间隔重复计划保存；打开「错题」Tab 可立即复习。
5. 跟踪进度 — 在「复习」Tab 查看待复习项，或在首页查看热力图。

---

命令

快捷键

不预设任何快捷键。请在 设置 → 快捷键 中自行绑定（例如把 Ctrl+Q 绑定到 基于当前文档直接生成试题）。

---

配置

核心设置

复习间隔因子

间隔由每个模块的**难度因子（SM-2 的 EF）**控制：SM-2 按 1 天 → 6 天 → 上一间隔 × 因子递增长，因子再随每次评分微调。推荐 2.0–2.8（标准 2.5）；输入限制在 1.3–3.0。

---

界面

侧边栏共有 7 个 Tab：

---

技术栈

---

更新日志

v2.3.0

• 真 SM-2 间隔重复：每条记录独立维护难度因子（EF）与重复次数；复习采用四档评分（忘记 / 困难 / 一般 / 简单）。每个模块只填一个间隔因子（默认 错题 2.3 / 题目 2.5 / 笔记 2.5，推荐 2.0–2.8），并提供 少复习 / 标准 / 多复习 一键预设；间隔按 1 天 → 6 天 → 上一间隔 × 因子 递增。
• 手动添加错题：在 错题 页手动录入（内容 / 来源 / 知识点标签 / 备注）。
• 仅答案导出：可导出只含答案与解析的试题文件（错题详情同样支持）。
• 聊天工作区：AI 对话支持工作区，每个工作区内可有多个独立会话；每个会话单独记忆检索范围（旧版单一历史自动迁移）。
• 本地日期修复：复习排期改用本地时区，不再出现 UTC 导致的差一天。
• 设置单一来源：原生设置页与侧边栏设置页由同一 schema 渲染，不再各自分叉。
• 清理：移除两个未使用依赖；移除硬编码的全局 Ctrl+Q 监听，改为普通命令（快捷键自行绑定）。
• 307 项测试通过；tsc 0 错误；ESLint 0 错误 / 0 警告。

v2.2.0

• 内置 AI 对话：藏在智学助手侧边栏的 AI 页签——基于笔记检索回答，支持多文件引用（当前笔记/选区或从文件选择器挑选）、可配置引用预算、Markdown 渲染、复制/重新生成、贴底滚动。
• 检索更稳：中文 2/3-gram 分词、文件名/路径加权、两段式候选排序（只读取命中的文件）。
• 对话体验：就地错误/警告气泡、超长提示、Ctrl/⌘+Enter 发送、打开即聚焦。
• 可靠性：文件夹创建幂等（不再启动报 “Folder already exists”）、网络错误友好提示、历史记录裁剪（≤100 条，结果截断）、题目/笔记/错题列表短时缓存。
• 可维护性：sidebarView.ts 由约 3700 行拆分为 src/views/sidebar/ 下 13 个模块；数据/列表/导出逻辑下沉到 VaultDataService；脆弱的 [style*=] CSS 选择器全部替换为语义 class。
• 重建 main.js；CI 增加构建 + 单元测试；新增护栏测试确保无 eval / 无脚本注入。
• 194 项测试通过；tsc 0 错误、ESLint 0 errors。

v2.1.0

• 一键中英文切换 — 全部 Tab 约 500 条 UI 文案接入中英文字典，在设置页（或侧边栏设置内）一键切换、即时生效。
• AI 提示词双语化 — 出题、识别试卷、生成笔记、添加标签的提示词支持中英文，保留「语言与材料一致」规则。
• 英文解析与导出兼容 — 题目解析器与导出器识别 Answer: / Explanation: / Answer Summary 等英文标签，英文材料出的题可正常解析与导出。
• 字典回退机制 — 漏翻文案自动显示原文，界面永不空白。
• 动态脚本剥离正则改为与 minifier 变量名无关，createElement("script") 始终保持 0 处，满足 Obsidian 审核 lint。
• 153 项测试全部通过（新增：字典完整性、占位符替换、提示词双语、英文解析）；tsc 0 错误、ESLint 0 errors。

v2.0.1

• 清除 setimmediate 依赖残留的动态 <script> 注入（Obsidian 审核 lint 报错的靶点），createElement("script") 由 3 处降为 0 处，解除社区审核阻塞。

v2.0.0

• 知识点索引重建更可靠 — 单个标签写入失败不再中断整个重建；过期索引一次性全部清理。
• 复习进度跨会话持久化 — 间隔 / 答对次数以数字存储，间隔重复连续记录不再因重载被重置。
• 热力图响应式 — 使用滑动条、自动定位当天，并在调整侧边栏时始终钉在当天。
• 新增系统化测试套件（知识点索引、文档服务、笔记服务、修复验证）。

---

支持

如果这个插件对你的学习有帮助，欢迎在 GitHub 上给一个 ⭐！

GitHub: https://github.com/xxinjie21/Smart-Quiz-Tutor

如有问题或建议，请 提交 Issue。

---

许可

ISC License