FlashyGen

Generate Anki flashcards from Notion pages using AI.

Automatically creates diverse flashcards with proper syntax highlighting for code examples, supporting both local (Ollama) and cloud (Claude) AI providers.

Features

• AI-powered flashcard generation (Ollama or Claude)
• Multiple question types: recall, conceptual, application, comparison, command, troubleshoot
• Cloze cards over code blocks, generated alongside Q/A cards
• Asset-aware cards: figures from the note are embedded in the deck, code blocks land byte-exact on cards (the model cites [CODE n]/[FIGURE n] tokens; it never retypes code)
• Explainer section below each answer — context beyond the recalled fact
• Quality gate + LLM validation pass: thin, duplicated, ungrounded, or hallucinated cards are dropped before export
• Coverage manifest next to every deck; check and update commands diff a deck against the current note and regenerate only what changed
• Resumable generation: per-section checkpoints in decks/.work/ — interrupted runs don't repeat LLM calls
• Deterministic deck/note IDs — re-importing an updated deck merges in Anki and preserves review history
• Equations rendered via Anki MathJax; tables carried through as markdown
• Works with private Notion pages; clean, styled decks ready to import

Installation

git clone https://github.com/yourusername/flashygen.git
cd flashygen
uv sync  # or: pip install -r requirements.txt

Quick Setup

1. Choose AI Provider

   • Ollama (free, local): Install from ollama.ai, run ollama pull phi3
   • Claude (paid): Get API key from console.anthropic.com

2. Create Notion Integration

   • Go to notion.so/my-integrations
   • Create new integration, copy the token

3. Configure

   cp .env.example .env
   # Edit .env and add your tokens

4. Share Notion Pages

   • Open your Notion page → "..." menu → "Add connections" → Select your integration

For detailed setup help, run:

python main.py setup

Usage

Generate a deck:

python main.py generate https://www.notion.so/your-page-url

Writes decks/<title>.apkg plus decks/<title>.manifest.json (card provenance), and prints a coverage summary of any note sections that produced no cards.

Check a deck against the current note (no LLM calls):

python main.py check <url>

Reports new sections, changed sections (stale cards), sections with no cards, and figures/code blocks no card cites. Exits 1 when gaps exist, so it can run on a schedule.

Update a deck incrementally:

python main.py update <url>

Regenerates cards only for new/changed/uncovered sections, keeps everything else from the manifest, and re-exports. Re-import in Anki: existing cards update in place, review history intact.

Options (generate/update):

--provider, -p               # AI provider: "ollama" (default) or "claude"
--output, -o                 # Output file path (default: decks/<page_title>.apkg)
--cards, -c                  # Cards per concept (default: 3, generate only)
--deck-name, -d              # Custom deck name (default: page title)
--validate / --no-validate   # LLM grounding pass on generated cards (default: on)
--manifest, -m               # Manifest path (check/update; default: decks/<title>.manifest.json)

Examples:

# Use Claude instead of Ollama
python main.py generate <url> -p claude

# Faster generation without the validation pass
python main.py generate <url> --no-validate

# All options
python main.py generate <url> -p claude -o deck.apkg -c 4 -d "Python Study"

Troubleshooting

"Configuration error: ANTHROPIC_API_KEY not found"

• Create .env file and add your API key without quotes

"Error fetching page: Could not find page"

• Share the Notion page with your integration (see setup step 4)

"No flashcards were generated"

• Page may be empty or have insufficient content

Generated .apkg file not working

• Install Anki, then double-click .apkg or import via File → Import

Duplicate decks after upgrading to 0.2.0

• Deck IDs became deterministic in 0.2.0; delete decks imported with older versions from Anki once, then re-import — subsequent updates merge cleanly

A run was interrupted mid-generation

• Just re-run the same command: sections already generated are resumed from decks/.work/<page_id>/ checkpoints without new LLM calls

License

MIT License - see LICENSE file for details