An intelligent knowledge management platform for browsing, annotating, and mastering HTML learning documents. Powered by local LLM integration for AI-generated quizzes, document chat, concept extraction, and study plans, with rich interactive visualizations and science-backed spaced repetition.
InsightHub integrates with any OpenAI-compatible local LLM to bring intelligence directly into your reading workflow:
- AI Quiz Generation — Generate quiz sessions with 5 question types via SSE streaming: multiple-choice, true/false, short-answer, fill-in-the-blank, and code completion. Configurable difficulty and question count.
- AI Document Chat — Multi-turn conversation about document content with streaming responses.
- AI Document Summarization — Structured summary with core takeaways, key concepts, and content outline.
- AI Evaluation — Assess document accuracy, completeness, and depth with AI-generated ratings.
- AI Inception (Progressive Summary) — Multi-level distillation from full content to increasingly concise abstracts.
- AI Concept Extraction — Auto-extract key concepts from documents and generate spaced repetition flashcards.
- AI Concept Challenges — Devil's advocate multi-round debates that test deep understanding with scoring.
- AI Code Editor — Built-in CodeMirror editor with AI coaching, syntax highlighting for 15+ languages, and code execution.
- AI Whiteboard — Drawing canvas with AI vision analysis to interpret diagrams and handwritten notes.
- AI Study Plan — Paste a job description or learning goal, and the AI matches relevant documents to create a personalized plan.
- AI Bubble — Hover over any concept to get an instant AI explanation popover.
- Token Usage Tracking — Monitor AI token consumption with cost estimation for commercial LLMs.
- Privacy-First — All AI features run against a local model. No data leaves your machine.
- Knowledge Graph — D3-force network graph with pan, zoom, and drag. Similarity edges via TF-IDF cosine similarity.
- Personal Knowledge Map — Force-directed graph centered on "You" with engagement-based node sizing and mastery color coding.
- Knowledge Tree — Collapsible tree: Category → Document → Concept.
- GitHub-Style Reading Heatmap — Daily reading activity over time.
- Category Radar Chart — Reading distribution across categories.
- Quiz Performance Dashboard — Circular gauge, difficulty distribution, and score trend.
- Reading Habits Analysis — Hourly, weekday patterns, and streak tracking.
- Tag Cloud — Dynamic word cloud sized by usage frequency.
- Learning Path — Timeline of milestones with AI-generated study recommendations.
- Concept cards auto-generated by AI from document content.
- SM-2 algorithm (SuperMemo 2) scheduling — the same algorithm behind Anki.
- Intervals: 1 day → 6 days → 17 days → 49 days → ... based on recall performance.
- 3D flip animation with keyboard shortcuts (Space to flip, 0-5 to grade, S to skip).
- Highlight text in 6 colors with persistent overlays.
- Comment on any selection with threaded replies.
- Click-to-view — Click highlighted passages to see annotation details.
- Touch support — Full annotation on iPad and touch devices.
- XPath serialization with fuzzy text matching restore.
- LAN sync across clients.
44 achievements across 5 categories:
- Reading (11) — First Read → Million Words
- Quiz (10) — First Quiz → Question Crusher
- Annotation (9) — First Highlight → Conversation King
- Streak & Explore (9) — Three-Day Streak → Search Savant
- Special (8) — Speed Reader, AI Scholar, and more
| Category | Features |
|---|---|
| Document Management | Category browsing, full-text search (FlexSearch), tag filtering, read-later, document import (HTML/URL), cross-document navigation, hide & restore, trash |
| AI Integration | Quiz (5 types, SSE streaming), chat, summarization, evaluation, inception, concept extraction, challenges, code editor with coaching, whiteboard with vision, study plan, AI bubble, token tracking |
| Annotations | 6-color highlights, threaded comments, click-to-view popup, touch support, XPath persistence with fuzzy restore |
| Spaced Repetition | AI concept cards, SM-2 scheduling, 3D flip cards, keyboard shortcuts |
| Visualizations | Knowledge graph, personal map, knowledge tree, heatmap, radar, quiz dashboard, reading habits, tag cloud, learning path |
| Gamification | 44 achievements across 5 categories, toast notifications |
| Data & Sync | localStorage persistence, LAN sync via REST API, workspace isolation, backup/export/import |
| UI/UX | Light/dark theme, sidebar, keyboard shortcuts, iframe reader, feature toggles, multi-profile AI config |
Home Dashboard — reading stats, category overview, and recent activity
Category Browsing — filter documents by workspace and category
Document Reader — iframe-embedded documents with a full annotation and AI toolbar
AI Quiz — auto-generated quiz sessions with 5 question types, difficulty tags, and score tracking
AI Chat — multi-turn conversation about document content with streaming responses
Code Editor — built-in CodeMirror editor with AI coaching panel and multi-language support
Whiteboard — freehand drawing canvas for visual note-taking alongside documents
AI Summary — one-click structured summary with key takeaways, concepts, and outline
AI Inception — progressive multi-level summarization from full content to abstract
AI Challenge — devil's advocate multi-round debates to test concept understanding
Concept Extraction — AI-generated concept cards for spaced repetition review
Knowledge Graph — interactive D3-force network of documents, categories, and tags
Learning Path — knowledge tree with milestones and AI study plan
Learning Analytics — reading heatmap, quiz performance, category radar, and habits
Achievements — 44 unlockable milestones across 5 categories
Spaced Repetition — SM-2 algorithm flashcard review with 3D flip animation
Full-Text Search — instant FlexSearch across all documents with keyboard shortcut
Token Stats — AI token usage tracking with cost estimation per model
All Notes — unified view of all comments and annotations across documents
| Layer | Technology | Purpose |
|---|---|---|
| Framework | React 19 + TypeScript | UI and type safety |
| Build | Vite 8 | Dev server and bundling with code splitting |
| State | Zustand 5 | Lightweight reactive state (7 stores) |
| Search | FlexSearch | Client-side full-text search |
| Editor | CodeMirror 6 | Code editing with syntax highlighting |
| Charts | Recharts | Statistical charts (radar, heatmap, line, bar) |
| Graph | D3-force | Force-directed graph layouts |
| Icons | Lucide React | Consistent icon system |
| Styling | Pure CSS + Custom Properties | Theming without dependencies |
| Fonts | Inter Variable, JetBrains Mono | UI and code fonts |
| AI | OpenAI-compatible SSE | Local LLM integration |
Zero UI framework dependencies — no Material UI, no Tailwind, no Bootstrap. Every component is hand-crafted with pure CSS.
- Node.js >= 18
- A local LLM server (optional, for AI features)
cd insighthub
npm install
npm run devOpen http://localhost:5600. Add your document workspaces from the Settings page or by editing data/.insighthub-workspaces.json.
Without a local LLM, the app works fully — browsing, search, annotations, and all visualizations function normally. Only AI features (quiz, chat, concept extraction, etc.) require an LLM server.
npm run build # Copies documents into public/docs/, type-checks, and bundles
npm run preview # Serve the production build locallyinsighthub/
├── src/
│ ├── components/
│ │ ├── DocReader/ # 19 components for document reader
│ │ │ ├── AnnotationBar # Highlight/comment toolbar
│ │ │ ├── AnnotationPanel # Annotation list sidebar
│ │ │ ├── AnnotationPopup # Click-to-view annotation details
│ │ │ ├── CommentDialog # Comment editing dialog
│ │ │ ├── ChatPanel # AI multi-turn document chat
│ │ │ ├── SummaryPanel # AI document summarization
│ │ │ ├── EvaluationPanel # AI document evaluation
│ │ │ ├── InceptionPanel # AI progressive summarization
│ │ │ ├── ChallengePanel # AI concept challenges (5 rounds)
│ │ │ ├── ScriptPanel # AI presentation script
│ │ │ ├── QuizPanel # AI quiz session
│ │ │ ├── ConceptCardsPanel# AI-extracted concept cards
│ │ │ ├── SimilarDocsPanel # Document similarity sidebar
│ │ │ ├── AIBubble # Hover concept explanation
│ │ │ ├── CodeEditorPanel # CodeMirror editor with AI coaching
│ │ │ ├── WhiteboardPanel # Drawing canvas with AI vision
│ │ │ ├── ShadowTypingPanel# Shadow typing practice
│ │ │ ├── RatingDialog # Document rating dialog
│ │ │ └── WikiLinkRenderer # Wiki-style bidirectional links
│ │ ├── Layout/ # App shell, sidebar, navbar, file tree
│ │ ├── visualization/ # 12 interactive visualization components
│ │ ├── stats/ # Chart containers and stat components
│ │ ├── shared/ # Reusable UI (DocCard, DocGrid, FilterBar, etc.)
│ │ ├── Import/ # Import, move, and URL import dialogs
│ │ └── search/ # Search dialog
│ ├── pages/ # 16 page components
│ ├── services/ # 13 service modules
│ ├── stores/ # 7 Zustand stores
│ ├── hooks/ # 6 custom hooks
│ ├── utils/ # XPath, graph builders, aggregators, exporters
│ ├── types/ # Centralized TypeScript type definitions
│ └── styles/ # 7 CSS files (globals, layout, components, etc.)
├── vite-plugins/ # Custom Vite plugin (document discovery + 36 API endpoints)
├── scripts/
│ └── copy-docs.ts # Build-time document copy script
└── vite.config.ts
Configure from the Settings page — supports multiple AI model profiles with full CRUD. Compatible with any OpenAI-compatible server: llama.cpp, Ollama, vLLM, LM Studio, etc.
Workspaces are configured in data/.insighthub-workspaces.json or added via Settings. Each workspace maps to a local directory with its own ID prefix. Categories are discovered dynamically from the folder structure.
[
{
"id": "mydocs",
"label": "MyDocuments",
"icon": "BookOpen",
"path": "../MyDocuments",
"prefix": "my",
"subtitle": "My Knowledge Base"
}
]Toggle between workspaces from the navbar. Each workspace independently filters documents, annotations, flashcards, tags, and sidebar navigation.
Individual AI features can be enabled/disabled from Settings. Disabling a feature only hides its button — existing data is preserved.
Full data backup and restore: export all data (localStorage + server endpoints) as a single JSON file, and import to restore on any machine.
Configure difficulty (easy/medium/hard), question count, and enabled question types (multiple-choice, true/false, short-answer, fill-in-the-blank, code completion).