Memo AI

An Obsidian plugin that uses AI-powered chunk extraction and review with the SM2 spaced repetition algorithm to help you memorize your notes.

Features

🤖 AI-Powered Chunk Extraction

• Intelligent & Template-Based Extraction: Automatically extracts knowledge chunks from your notes using Large Language Models (LLM)
  • Create custom extraction templates with specific requirements to guide how chunks are extracted (e.g., focus on key concepts, prioritize actionable insights)
  • Select templates before extraction - your selection is remembered for convenience
  • LLM understands context and semantic meaning, not just text matching
• Incremental Updates: When you modify a note, the plugin intelligently updates existing chunks:
  • Keep: Preserves chunks that remain unchanged
  • Modify: Updates chunks with changed content (with minor/moderate/major update levels)
  • Delete: Removes chunks for content that no longer exists
  • Create: Adds new chunks for new content
• Markdown Support: Chunks support full markdown rendering including code blocks, math formulas, tables, and more

🎯 Smart Push System with SM2 Algorithm

• SM2 Spaced Repetition: Implements the proven SuperMemo 2 algorithm for optimal review scheduling
• Adaptive Scheduling: Review intervals automatically adjust based on your performance
• Importance Multiplier: High-importance chunks are reviewed more frequently
• Familiarity Tracking: Weighted average of past grades (0.0-1.0) tracks your mastery level
• Intelligent Scoring: Chunks are scored based on:
  • Importance level (low, medium, high)
  • Familiarity score (how well you know it)
  • Due date (past due chunks get priority)
• Automatic Scheduling: Top-scoring chunks are automatically scheduled for review
• Configurable Threshold: Set a minimum score threshold for push recommendations
• Auto Cleanup: Expired and completed pushes are automatically removed

💬 Interactive Review System

• AI Conversation Evaluation:
  • Engage in conversations with AI to evaluate your understanding
  • Template-Based Conversations: Use custom review templates to guide AI tutor behavior
    • Define how AI should ask questions (e.g., test deep understanding, provide detailed explanations)
    • Select templates before starting conversation
    • Templates are remembered for next conversation
  • AI generates relevant questions based on chunk content and template requirements
  • Adaptive responses and feedback
  • End conversations early to get immediate AI evaluation
  • Automatic grading (0-5 scale)
  • Markdown Support: Conversation messages support full markdown rendering
• Manual Grading:
  • Simple and intuitive 5-star rating system
  • Quick evaluation without AI interaction
• Flexible Workflow: Choose between AI conversation or manual grading for each push
• Tabbed Interface: Switch between manual and AI evaluation modes

🔧 Chunk Management

• Importance Control: Set importance level (1-3 stars) for each chunk
• Review Toggle: Enable/disable review for specific chunks
• Filter by Review Status: Toggle to hide chunks that don't need review
• Chunk Deletion: Remove unwanted chunks
• Automatic Cleanup: Chunks for deleted notes are automatically removed
• Detailed Metrics: View creation date, familiarity score, review interval, repetition count, and chunk score
• Template Selection: Choose extraction templates before extracting chunks

Installation

Manual Installation

1. Download the latest release from the Releases page
2. Extract the plugin folder to your Obsidian vault's .obsidian/plugins/ directory
3. Open Obsidian and go to Settings → Community Plugins
4. Enable the Memo AI plugin

Development Installation

1. Clone or copy this repository to .obsidian/plugins/ai_notebook_plugin/
2. Open a terminal in the plugin directory
3. Run npm install to install dependencies
4. Run npm run build to build the plugin
5. Enable the plugin in Obsidian settings

Usage

Getting Started

1. Configure LLM Settings: Go to Settings → Memo AI → LLM Settings and enter your API key

   • Supports OpenAI API and compatible APIs (e.g., Alibaba Cloud DashScope)
   • Configure API base URL, model name, and timeout

2. Extract Chunks from a Note:

   • Open a note you want to review
   • Open "Note Chunks" view
   • (Optional) Select an extraction template from the dropdown to guide chunk extraction
   • Click "Extract Chunks" button
   • The plugin will use LLM to intelligently extract knowledge chunks based on your template requirements
   • Your template selection is remembered for next extraction

3. Review Chunks:

   • Open "Push Center" view (automatically opens on plugin load)
   • Click "Refresh Pushes" to schedule new pushes
   • Select a push to review
   • For AI conversation: (Optional) Select a review template before starting
   • Choose between AI conversation or manual grading

Note Chunks View

The Note Chunks view shows all chunks extracted from the current note:

• Template Selection: Choose an extraction template from the dropdown before extracting
  • Templates define custom requirements for chunk extraction
  • Your selection is remembered for next extraction
  • Select "No requirements" for default extraction behavior
• Hide No-Review Toggle: Toggle switch to hide chunks that don't need review
• Extract Chunks: Extract or update chunks from the current note (purple button)
• Importance Rating: Click stars to set importance (1-3 stars)
• Review Toggle: Toggle whether a chunk needs review
• Delete: Remove a chunk
• View Details: See familiarity score, review interval, repetition count, and chunk score
• Markdown Rendering: Chunk content supports full markdown including code blocks, math formulas, tables, etc.

Push Center

The Push Center is your main review interface:

• Push List: Sidebar showing all active pushes (collapsible)
• Push Details:
  • Chunk content and metadata (with markdown rendering)
  • Due time and creation time
  • Open note or delete push buttons
• Template Selection: For AI conversation, select a review template before starting
  • Templates guide how the AI tutor asks questions and evaluates responses
  • Your selection is remembered for next conversation
  • Template cannot be changed after conversation starts
• Evaluation Options:
  • Manual Grading: 5-star rating system with submit button
  • AI Conversation: Interactive Q&A with AI tutor (supports markdown in messages)
• Refresh Pushes: Automatically schedules new pushes based on chunk scores

AI Conversation Flow

1. (Optional) Select a review template from the dropdown to guide AI behavior
2. Click "Start" to begin the conversation
3. AI generates a question based on the chunk and template requirements
4. Type your answer in the input box (supports markdown)
5. AI evaluates your response and provides feedback
6. Continue the conversation or click "End Conversation" to get immediate evaluation
7. SM2 algorithm updates based on your performance
8. All conversation messages support markdown rendering

Manual Grading Flow

1. Select a star rating (1-5 stars)
2. Click "Submit Grade" to apply
3. SM2 algorithm updates based on your grade

Settings

LLM Settings

• LLM API Key: Your OpenAI API key or compatible API key
• LLM API Base URL: API endpoint (default: https://api.openai.com/v1)
  • For Alibaba Cloud DashScope: https://dashscope.aliyuncs.com/compatible-mode/v1
• LLM Model: Model name (e.g., gpt-3.5-turbo, gpt-4)
• LLM Request Timeout: Maximum time to wait for LLM responses (10-300 seconds, default: 60)

Chunk Extraction Templates

Create custom templates to guide how chunks are extracted from your notes:

• Add Template: Click "+ Add Template" to create a new template
• Template Name: Give your template a descriptive name (e.g., "Focus on key concepts")
• Requirements: Describe your requirements for chunk extraction
  • These requirements will be sent to the LLM when extracting chunks
  • Example: "Focus on extracting key concepts and definitions. Prioritize actionable insights."
• Edit Template: Click "Edit" to modify an existing template
• Delete Template: Click "Delete" to remove a template
• Template Management: Templates are saved and persist across sessions
• Last Selected: Your last selected template is remembered for convenience

Push Review Templates

Create custom templates to guide AI conversation review behavior:

• Add Template: Click "+ Add Template" to create a new template
• Template Name: Give your template a descriptive name (e.g., "Deep understanding focus")
• Requirements: Describe how the AI tutor should ask questions and evaluate responses
  • These requirements guide the AI's teaching style
  • Example: "Ask questions that test deep understanding. Provide detailed explanations when the learner struggles."
• Edit Template: Click "Edit" to modify an existing template
• Delete Template: Click "Delete" to remove a template
• Template Management: Templates are saved and persist across sessions
• Last Selected: Your last selected template is remembered for convenience

Push Settings

• Max Active Pushes: Maximum number of pushes active at the same time (1-20, default: 5)
• Push Due Window: Duration of a push in hours before expiration (1-168 hours, default: 24)
• Push Score Threshold: Minimum chunk score required for push recommendation (2.0-6.0, default: 2.0)

Commands

• Extract chunks from current note: Extract or update chunks from the active note
• View chunks for current note: Open the Note Chunks view
• Open push center: Open the Push Center view

How It Works

Chunk Extraction

1. Initial Extraction: When you first extract chunks, LLM analyzes the note and creates knowledge chunks
2. Incremental Updates: When you modify a note and extract again:
   • LLM compares new content with existing chunks
   • Determines which chunks to keep, modify, or delete
   • Creates new chunks for new content
   • For modified chunks, determines update level (minor/moderate/major)
3. Update Level Impact:
   • Minor: Slight reduction in familiarity and EF
   • Moderate: Moderate reduction, one repetition removed
   • Major: Significant reduction, repetitions reset to 0

SM2 Algorithm

The plugin uses the SuperMemo 2 algorithm with the following parameters:

• E-Factor (EF): Ease factor (1.3-2.5) representing memory difficulty
• Repetitions: Number of successful reviews
• Interval: Days until next review
• Familiarity Score: Weighted average of past grades (0.0-1.0)

Review Intervals:

• First review: 1 day
• Second review: 6 days
• Subsequent reviews: Previous interval × EF (adjusted by importance)

Grade Impact:

• Grade < 3: Reset to beginning (repetitions = 0, interval = 1 day)
• Grade ≥ 3: Increase repetitions, calculate new interval

Chunk Scoring

Chunks are scored using a formula that considers:

1. Importance Weight: Higher importance = higher base score
2. Familiarity Boost: Lower familiarity = higher priority
3. Due Date Boost: Past due chunks get exponential boost
4. Future Penalty: Future due chunks get slight penalty

Only chunks with scores above the threshold are recommended for pushing.

Automatic Cleanup

• Deleted Notes: When a note is deleted, all associated chunks and pushes are automatically removed
• Expired Pushes: Pushes past their due window are removed when refreshing
• Completed Pushes: Completed pushes are removed when refreshing

Tips & Best Practices

1. Start Small: Begin with a few notes to understand the workflow
2. Use Templates: Create templates for different types of content
   • Create a "Key Concepts" template for extracting important definitions
   • Create a "Deep Understanding" template for AI conversations that test comprehension
   • Create a "Quick Review" template for faster, surface-level questions
3. Set Importance: Mark important chunks with higher importance for more frequent review
4. Regular Reviews: Use "Refresh Pushes" regularly to keep your review queue active
5. AI vs Manual: Use AI conversation for complex topics, manual grading for quick reviews
6. Template Selection: Select templates before extraction/conversation - your choice is remembered
7. Update Threshold: Adjust push score threshold based on your review capacity
8. LLM Model: Use GPT-4 for better chunk extraction quality (if available)
9. Markdown Support: Use markdown in your notes - chunks will render code blocks, math formulas, and more beautifully
10. Filter Chunks: Use the "Hide no-review" toggle to focus on chunks that need attention

Troubleshooting

Chunks Not Extracting

• Check your LLM API key is correct
• Verify API base URL is correct for your provider
• Check network connection
• Increase timeout if using slower models

Pushes Not Appearing

• Click "Refresh Pushes" to schedule new pushes
• Check push score threshold isn't too high
• Ensure chunks have needsReview enabled
• Verify chunks have valid due dates

AI Conversation Not Working

• Check LLM settings are configured correctly
• Verify API key has sufficient credits/quota
• Check timeout setting is appropriate

Development

# Install dependencies
npm install

# Build for development (with watch mode)
npm run dev

# Build for production
npm run build

License

MIT

Author

Jiawei Yang