AttendingAI - AI-Powered Answer Assessment for Anki

Version: 1.00
Size: ~28MB (lite) | ~163MB (with neural embeddings)
Anki Version: 25.02.6+ (Python 3.9)

Requires a Local or Cloud-Based LLM (Recommended: Ollama)

What is AttendingAI?

AttendingAI transforms Anki into an intelligent study companion by providing:

• ✅ AI-Powered Grading - Get instant, detailed feedback on your answers
• ✅ Semantic RAG - Retrieves relevant sections from your uploaded materials
• ✅ Tutor Mode - Interactive learning assistant that explains concepts
• ✅ Smart Analytics - Track weak concepts and learning patterns
• ✅ Batch Assessment - Process multiple cards efficiently

Quick Start

1. Install: Tools → Add-ons → Install from file
2. Restart: Close and reopen Anki
3. Start Using: Press W during review or right-click → Assess Answer

That's it! Semantic search (TF-IDF) works immediately with zero setup.

Key Features

🎯 Intelligent Grading

• Understands partial credit and medical terminology
• Provides detailed feedback with rubric scores
• Adjusts Anki intervals based on understanding

📚 Semantic Material Retrieval (RAG)

• Upload PDFs and DOCX files for your decks
• Automatically retrieves relevant sections during grading
• TF-IDF included - Works out-of-the-box, no dependencies
• Optional: Neural embeddings - 10-15% better for power users

🧑‍🏫 Tutor Mode

• Ask follow-up questions about your answers
• Get explanations tailored to your course materials
• Track conversation context for personalized learning

📊 Advanced Analytics

• Identify weak concepts across your deck
• Generate review decks targeting weak areas
• Track progress over time

Installation Options

Default (Recommended)

Size: ~28MB | Setup: 1 minute | Quality: 85-90%

Just install the add-on - everything works immediately!

With Neural Embeddings (Optional)

Size: ~163MB total | Setup: 5 minutes | Quality: 95-100%

For power users who want maximum semantic understanding:

# Windows
py -3.9 -m pip install --target "%APPDATA%\Anki2\addons21\attending_ai\vendor" fastembed==0.4.2 onnxruntime==1.19.2

# Mac
python3.9 -m pip install --target "~/Library/Application Support/Anki2/addons21/attending_ai/vendor" fastembed==0.4.2 onnxruntime==1.19.2

# Linux
python3.9 -m pip install --target "~/.local/share/Anki2/addons21/attending_ai/vendor" fastembed==0.4.2 onnxruntime==1.19.2

See INSTALLATION.md for complete instructions.

How It Works

During Review

1. Answer a card (write your response)
2. Press W or right-click → Assess Answer
3. AttendingAI:
   • Retrieves relevant sections from your uploaded materials
   • Grades your answer with detailed feedback
   • Provides a rubric breakdown
   • Suggests an Anki interval
4. Accept or modify the suggested grade

Semantic Search (RAG)

• TF-IDF (default): Keyword-based retrieval, instant, zero setup
• Neural (optional): Context-aware understanding, better for varied terminology
• Auto-fallback: Switches to TF-IDF if neural unavailable

Documentation

• INSTALLATION.md - Complete installation guide including neural embeddings and multimodal support
• Field Map Script.md - Field mapping for multimodal cards

Requirements

• Anki: 25.02.6 or higher (Python 3.9)
• Platform: Windows, macOS, Linux
• Optional: Python 3.9 for neural embeddings installation

What's Included?

Core Files:

• All Python modules for grading, RAG, tutoring, analytics
• TF-IDF implementation (zero dependencies)
• Hybrid fallback system
• PDF and DOCX parsing
• Documentation

Vendor Dependencies (~28MB):

• PyPDF2, python-docx (document parsing)
• nltk, lxml (text processing)
• requests, certifi (HTTP client)
• coloredlogs, click (utilities)

NOT Included (Install Separately):

• FastEmbed (neural embeddings) - ~70KB + 135MB deps
• ONNX Runtime (inference engine) - ~11MB

Technical Details

Semantic Search Methods

TF-IDF (Default):

• Pure Python implementation
• Term Frequency-Inverse Document Frequency
• Cosine similarity for relevance
• 85-90% quality, instant retrieval
• Zero external dependencies

Neural (Optional):

• FastEmbed with ONNX Runtime
• Model: BAAI/bge-small-en-v1.5 (384 dimensions)
• Context-aware semantic understanding
• 95-100% quality, 2-5s first run
• Requires ~135MB installation

Database

• SQLite3 for materials and embeddings
• Efficient storage with float32 precision
• Automatic indexing for fast retrieval

License

AttendingAI is Copyright (c) 2025 Matthew Palmieri

AttendingAI is free and open-source. The add-on code that runs within Anki is released under the GNU AGPLv3 license, extended by a number of additional terms. For more information please see the LICENSE file that accompanied this program.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY.

Credits

Built with:

• FastEmbed - ONNX-based embeddings by Qdrant
• ONNX Runtime - Fast inference by Microsoft
• BAAI/bge-small-en-v1.5 - State-of-the-art retrieval model
• Anki - Spaced repetition system

Troubleshooting

Common Issues

"API Key Invalid" Error

• Verify your API key is correct in Settings → LLM Configuration
• Ensure you have an active subscription with your LLM provider
• Check that the API endpoint URL is correct

"No embeddings found" Warning

• Upload course materials first (Settings → Materials)
• Wait for embedding generation to complete
• Check storage permissions for AttendingAI user data

Grading Takes Too Long

• Reduce max_tokens in LLM settings
• Use faster LLM model (e.g., GPT-4o-mini instead of GPT-4o)
• Disable multimodal processing if not needed

Images Not Being Analyzed

• Verify "Enable Vision Grading" is checked in Settings
• Ensure images are in supported formats (PNG, JPG, WebP)
• Check that image fields are correctly configured

Materials Not Being Retrieved

• Check that materials were successfully uploaded
• Verify embeddings were generated (check database file size)
• Try increasing RAG top_k value in settings
• Ensure materials are associated with the correct deck

Getting Help

• Check the FAQ below
• Review INSTALLATION.md for setup issues
• Check docs/ folder for detailed guides
• File an issue on GitHub with error details

FAQ

Q: What LLM providers are supported? A: OpenAI (GPT-4o, GPT-4o-mini), Anthropic (Claude 3.5 Sonnet), OpenRouter (various models), and any OpenAI-compatible API.

Q: Can I use this offline? A: No, AttendingAI requires internet connection to access LLM APIs for grading.

Q: How much does it cost to use? A: AttendingAI itself is free and open source, but you need your own API key and will incur costs from your LLM provider (typically $0.001-0.01 per card assessment).

Q: Is my data private? A: Your flashcard content is sent to your chosen LLM provider's API. Review their privacy policy. Data is not stored by AttendingAI beyond local analytics in your Anki profile.

Q: Can I customize the grading rubric? A: Yes! Go to Settings → Tone & Style to adjust grading strictness and tone. You can also configure deck-specific rubrics.

Q: What file formats are supported for materials? A: PDF, DOCX, and TXT files. Materials are chunked and embedded for RAG retrieval.

Q: Do I need neural embeddings? A: No. TF-IDF (included by default) works well for most users. Neural embeddings provide 10-15% better retrieval quality but require additional setup.

Q: Can I use this with my existing Anki collection? A: Yes! AttendingAI works with any Anki deck. Just answer cards normally and press W to grade your response.

Q: Does this change my Anki scheduling? A: Only if you choose to apply the suggested grade. You always have full control over the final grade applied to your card.

Q: What languages are supported? A: AttendingAI primarily supports English. Other languages may work but accuracy depends on your LLM provider's capabilities.

Support

For issues, questions, or feedback:

• Check the FAQ above
• Check documentation in docs/ folder
• Review troubleshooting in INSTALLATION.md
• File an issue on GitHub: https://github.com/matthewpalmieri02/AttendingAI/issues

Start smart. Study smarter. With AttendingAI.