Anki AI: The Ultimate Offline AI Tutor
Transform Anki into an intelligent learning partner. Anki AI analyzes your flashcards to answer your questions instantly, citing your own cards as sources. It runs completely offline on your computer, ensuring total privacy.
β¨ KEY FEATURES
π§ Smart RAG Technology
Answers questions by reading your flashcards in real-time. No hallucinationsβjust facts from your deck.
"Mechanism of Action?" β It quotes your pharmacology cards.
β‘οΈ Blazing Fast & Offline
Powered by Ollama with gemma2:2b (recommended for speed) or Llama 3, Phi-3, Mistral.
- Runs on your hardware
- ~5 second response times
- $0 cost, 100% privacy
π Hybrid Search
Does your deck have gaps? Anki Assist seamlessly combines your Flashcard knowledge with Web Search to give you comprehensive answers.
π¨ Premium Design
A beautiful, modern interface that feels right at home on your Mac or PC. Features:
- Rounded chat bubbles
- Smooth animations
- Comfortable reading experience
- Notes tab for quick notes
π οΈ INSTALLATION GUIDE
1. Install Ollama (Required)
- Download from https://ollama.com
- Install and run the application
- Open your terminal and type:
ollama pull gemma2:2b
(This downloads the AI brain - only 1.6GB!)
2. Install Anki Assist
- Download this add-on
- Restart Anki
3. Build Your Brain
- Click the "Re-Index" button in the chat window
- This reads your cards and builds a searchable knowledge base
- It only takes a few seconds!
π‘ HOW TO USE
- Ask Anything: "What are the symptoms of X?" "Explain the difference between A and B."
- Zoom: Adjust the font size in Settings (βοΈ icon) for comfortable reading
- Notes: Use the side tab to jot down quick notes without leaving Anki
- Re-Index: Click to rebuild the knowledge base after adding new cards
ποΈ RECOMMENDED MODELS
| Model | Speed | Size | Best For |
|-------|-------|------|----------|
| gemma2:2b β | ~5s | 1.6GB | Daily use, fast responses |
| phi3 | ~10s | 2.2GB | Good balance |Γ₯
| llama3 | ~12s | 4.7GB | More complex reasoning |
| qwen3:8b | ~48s | 5.2GB | Maximum quality |
To switch models, set the environment variable:
export OLLAMA_MODEL=gemma2:2b
π PRIVACY
Your data never leaves your computer. All processing is done locally via Ollama.
π CHANGELOG
v2.0.0 (2026-01-25)
- π Optimized RAG engine with BM25 scoring
- β‘οΈ Switched default model to gemma2:2b (5s responses!)
- π Added deduplication for large decks
- π¨ New brain icon
- π Added Notes tab
Created by Skyrunner