ankirai 0.0.4

Transform your notes into Anki flashcards at lightning speed

ankirai

暗記 (anki) · memorization  ·  雷 (rai) · lightning

Transform your notes into Anki flashcards at lightning speed using any LLM provider.

Installation

ankirai is available on PyPI. Requires Python 3.13+.

# Recommended — installs into an isolated environment managed by uv
uv tool install ankirai

# Or with pip
pip install ankirai

Or run directly from source:

git clone https://github.com/weavergoldman/ankirai
cd ankirai
uv run ankirai --help

Quick start

ankirai init                          # first-time setup: pick provider and enter API key
ankirai generate notes.pdf            # generate an Anki deck (.apkg)
ankirai generate notes.pdf --review   # review cards in the browser before exporting

I recommend using Google's AI Studio for your API key, since the free usage limit for Gemini 3.1 Flash Lite (the default Gemini model) is quite generous.

Commands

ankirai init

Interactive setup. Stores your provider and API key in a platform config file (chmod 600). Run once before your first generate.

ankirai generate [OPTIONS] INPUTS...

Parse one or more input files and generate an Anki deck.

Option	Description
--deck, -d	Deck name (default: first filename stem)
--model, -m	Model override
--provider, -p	Provider override (gemini, openai, openrouter, ollama)
--api-key	API key override
--format	Output format: apkg (default), csv, tsv
--output, -o	Output file path
--batch-size	Cards per LLM request
--force	Ignore manifest; reprocess all input files
--tags	Comma-separated tags to add to all cards
--vision / --no-vision	Override vision capability detection
--parsing-model	Use a separate model for image/PDF parsing
--prompt	Path to a custom prompt file
--review	Open browser review UI before exporting
--dry-run	Print extracted text chunks to stdout; skip generation and export
--instruct, -i	Extra instructions surrounded by quotes, appended to the active prompt for this run

ankirai config show

Print current config with API keys redacted.

ankirai config set KEY VALUE

Update a single config value, e.g.:

ankirai config set openai.api_key sk-...
ankirai config set providers.default openrouter

ankirai prompt show

Print the currently active card-generation prompt.

ankirai prompt edit

Open the global prompt in $EDITOR.

ankirai prompt reset

Restore the built-in default prompt.

Providers

Provider	Key name	Default model
gemini	GEMINI_API_KEY	gemini-3.1-flash-lite
openai	OPENAI_API_KEY	gpt-5.4-nano
openrouter	OPENROUTER_API_KEY	google/gemini-3.1-flash-lite
ollama	(none)	llama4

Claude / Anthropic models: Use OpenRouter with --provider openrouter --model anthropic/claude-haiku-4-5.

API keys can also be set via environment variables or a .env file in the working directory.

Config resolution

For each setting, the first value found wins:

1. CLI flag
2. Environment variable
3. .env file
4. config.toml (platform config dir)
5. Built-in default

Custom prompts

Prompt resolution order:

1. --prompt <file> flag
2. ankirai_prompt.md in the current working directory
3. Global prompt file (ankirai prompt edit)
4. Built-in default

The prompt should include {{notes}} where the chunk text will be substituted. Cards support LaTeX using MathJax syntax (\(...\) inline, \[...\] block). Cloze cards use {{c1::term}} syntax.

Review UI

--review launches a local web server at http://localhost:5173 and opens your browser.

Card view — keyboard shortcuts:

Key	Action
1	Accept
2	Edit (inline)
3	Reject
[ or ←	Previous card
] or →	Next card

Bulk view — shows all cards in a paginated table (50 per page). Use the "View" button on any row to jump to that card. Use "Accept all pending" to approve everything at once.

The nav bar shows "Card view" when on the bulk page and "Bulk view" when reviewing card-by-card.

Only accepted cards are included in the exported deck.

Manifest

ankirai tracks which files have been processed in a ankirai_manifest.json in the working directory. Re-running generate on the same inputs skips unchanged files. Use --force to reprocess everything.

Documentation

• Getting Started
• CLI Reference
• Configuration
• Review UI

Supported input formats

Anything markitdown can parse: PDF, DOCX, PPTX, images (with vision-capable models), plain text, Markdown, and more.