AnkiNote

AI-powered Anki card generator — vocabulary, phrases, sentences, and STEM concepts

📖 About

AnkiNote is an automated Anki flashcard generator that uses litellm to support a wide range of AI providers — Gemini, GPT, Claude, DeepSeek, and more — for generating definitions, examples, mnemonics, and images, then syncs directly with Anki through AnkiConnect.

✨ Features

• 🤖 Multi-Provider AI - Powered by litellm, supporting Gemini, GPT, Claude, DeepSeek, and more for text and image generation
• 🔊 Audio Generation - Text-to-Speech using Google Cloud TTS API
• 🖼️ AI Image Generation - Automatic image generation via Google AI (Gemini)
• 🔄 Direct Anki Sync - Seamless integration with Anki through AnkiConnect plugin
• 📝 Dual-Direction Cards - Supports both word→definition and definition→word learning modes
• 🎨 Beautiful Templates - Built-in Light/Dark mode responsive card templates
• ⚡ Batch Processing - Generate multiple cards from word lists efficiently
• 🌐 Multi-Language - Japanese, English (US), and extensible for more languages
• 🧮 STEM Concepts - Math, science, and programming concept cards with MathJax rendering

🚀 Quick Start

Prerequisites

• Python 3.14+
• uv package manager
• Anki with AnkiConnect plugin installed, or the in-process (headless) backend — see below

Installation

# Install from PyPI
uv pip install ankinote-ai

# Or install the CLI/GUI as an isolated uv tool
uv tool install ankinote-ai

AnkiNote has two front ends that share the same card-generation engine:

• Web UI (ankinote-gui) — everything, including AI provider keys, is configured from the browser. No .env file needed. Start here if you're new.
• CLI (ankinote) — scriptable, batch-friendly, configured via .env / environment variables. See CLI usage below.

🖥️ Web UI

Launching

# From a uv-managed checkout
uv run ankinote-gui

# Or, if installed as a tool / into a venv
ankinote-gui

This opens http://localhost:8080 in a browser (set ANKINOTE_SHOW=false to skip auto-open, e.g. on a headless server). The UI has six pages, reachable from the left drawer: Word, Phrases, Sentences, STEM, Card Types (note type + deck setup), and Settings.

Configuration — all in the Settings page

Nothing needs to be in a .env file for web UI use; every credential lives in the browser-saved settings file and is applied to the process at save time.

• Generation route (text) and Image route — each is a rack of named provider profiles. Pick a vendor template (OpenAI, Anthropic, Gemini, Fal, etc.) or "Custom / Other" for any OpenAI-compatible endpoint, then fill in base URL, model (with a refresh button to fetch live model IDs), and API key. Multiple profiles per vendor are supported (e.g. two OpenAI accounts), and you switch which one is active by clicking its pill.
  • For Fal image generation, use a Fal profile with base URL https://fal.run and your Fal API key, with a full endpoint id such as fal-ai/z-image/turbo.
• TTS (Google Cloud) — paste the Google Cloud TTS API key.
• Defaults — native/target language and whether image generation is on by default for new cards.
• AnkiWeb sync — shown when the in-process backend is active (see below): login/logout, sync status, manual sync, sync interval, and the upload/download choice on a required full sync.
• Backup & transfer — export all provider profiles + the TTS key as a passphrase-encrypted JSON bundle (scrypt + AES-256-GCM), and import one back in, merged by profile name. Useful for moving settings between machines or backing them up.

What still needs to be set outside the browser

A few things are process-level, not per-user settings, so they're still env vars:

The in-process (collection) backend additionally requires the ankinote-ai[headless] extra. It synchronizes with AnkiWeb at startup, after each note save or note-type setup batch, and every five minutes. A fresh collection blocks writes until its initial sync completes; an initialized collection remains writable offline. A required full sync blocks writes until you resolve the upload/download choice, in the Settings page or via ankinote anki sync.

Beside the collection file, .sync.json stores credential-free status, .credentials.json stores saved login credentials with mode 0600, .account retains an account binding after logout, and .backups/ holds recoverable collection backups made before full sync. Logout removes the saved credential and pauses synchronization without deleting the collection or media. Use a different data directory to switch accounts. Open a collection from only one process at a time.

Run the web UI with Docker

Published images (multi-arch amd64 + arm64): ghcr.io/ignity21/ankinote-ai and ignity21/ankinote-ai (Docker Hub), tags latest, <major>.<minor>, and the exact version.

Two ready-made compose stacks under deploy/:

cd deploy/standard        # GUI + an AnkiConnect you already run
# or: cd deploy/headless  # GUI with the in-process Anki backend, syncs to AnkiWeb
cp .env.example .env      # set ANKINOTE_STORAGE_SECRET (+ AnkiWeb login for headless)
docker compose up -d      # http://localhost:8080

AI provider keys are added in the web UI (Settings page), not .env. See deploy/README.md for the difference between the stacks, the AnkiConnect host setup, and building locally. The published image bundles the anki library, so both backends work without a custom build.

AnkiNote CLI - Usage Guide

Overview

The AnkiNote CLI is a scriptable, batch-friendly way to generate AI-powered Anki flashcards from the terminal. Unlike the web UI, it's configured with a .env file / environment variables rather than in-browser settings.

Configuration

Create a .env file with your API keys. At minimum, you need one AI provider key and the Google TTS key:

# At least one AI provider (for text and image generation)
DEEPSEEK_API_KEY=your_deepseek_key
# GEMINI_API_KEY=your_gemini_key
# OPENAI_API_KEY=sk-...
# ANTHROPIC_API_KEY=sk-ant-...
# XAI_API_KEY=xai-...          # for xAI image models

# Google Cloud TTS (for audio generation)
GOOGLE_TTS_KEY=your_tts_api_key

# AnkiConnect (defaults to http://localhost:8765)
ANKI_CONNECT_URL=http://localhost:8765

The ANKI_BACKEND, ANKI_COLLECTION_PATH, and ANKIWEB_USERNAME/ANKIWEB_PASSWORD variables described above for the web UI apply the same way to the CLI — the in-process backend and AnkiWeb sync aren't web-UI-only features.

No ankinote config/ankinote settings command yet. There's currently no CLI subcommand to set API keys — only .env / environment variables, plus ankinote anki login|logout|status|sync for AnkiWeb. A generic config subcommand has been sketched in the past but never implemented; see docs/plans/ if you pick this up — worth adding a plan doc for it before starting.

Commands

The CLI currently provides four collection entrypoints. Run ankinote <type> --help or ankinote <type> <command> --help for the complete, current option list.

Word, phrase, and sentence cards

word, phrase, and sentence all follow the same init / add / batch shape — init creates the note type and deck, add takes one item, batch takes several (as arguments or --file):

ankinote word init
ankinote word add serendipity
ankinote word batch serendipity ephemeral eloquent
ankinote word batch --file words.txt

ankinote phrase add "look after"
ankinote phrase batch --file phrases.txt

# sentence argument is in the native language; the target-language version
# is generated for the card back
ankinote sentence add "我今天起晚了。"
ankinote sentence batch --file sentences.txt

STEM cards

ankinote stem init
ankinote stem add "What is a derivative?"
ankinote stem add "State Newton's second law" --type formula
ankinote stem add "How do I invert a matrix?" --type procedure
ankinote stem add "Solve the problem in this photo" --type example --image problem.png
ankinote stem batch --file topics.txt
ankinote stem batch --file problems.txt --type example

STEM uses four independent note types: AINote STEM Concept, AINote STEM Formula, AINote STEM Procedure, and AINote STEM Example, all in the AINote::STEM deck. Each has its own fields and templates, with front first for duplicate detection and default sorting. Formula variables, solution steps, and images are stored in dedicated fields; tags use Anki's native tag store.

--type auto (the default) first classifies the request, then generates with the selected type's schema and prompt. Choosing a type skips that extra AI request. stem init initializes all four types; stem init --type concept initializes only Concept. The GUI supports the same type selection and previews every type for editing before saving, including variables, steps, tags, and diagram prompts.

The old AINote STEM type is no longer managed. Existing test notes are left untouched; this change does not migrate or delete them. Initialize the new types using ankinote stem init or the GUI's Card Types page.

Common options

Language-learning commands accept --native, --target, and --llm. Batch commands also accept --file and --rpm. STEM commands accept --llm, and can additionally configure diagram generation with --image-model and --image-size.

--llm and --image-model take any model id LiteLLM recognizes; the provider is inferred from the id and its key is read from the matching environment variable. Defaults are gpt-5.6-luna for text and gpt-image-1.5 (quality low) for images — neither default matters much in practice since any provider works, including fal.ai. Examples:

• --llm: gpt-5.6-luna, deepseek/deepseek-v4-flash, gemini/gemini-2.5-pro, claude-sonnet-4-20250514
• --image-model: gpt-image-1.5, gemini/gemini-3.1-flash-lite-image, fal_ai/fal-ai/flux/schnell, xai/grok-2-image

For example:

ankinote word add serendipity --native English --target 'Chinese(Simplified)'
ankinote word batch --file words.txt --rpm 30
ankinote stem add "State Bayes' theorem" --image-model gpt-image-1 --image-size 1024

Getting Help

# General help
ankinote --help

# Command-specific help
ankinote word --help
ankinote phrase --help
ankinote sentence --help
ankinote stem --help
ankinote word batch --help

📦 Tech Stack

• Language: Python 3.14+
• Package Manager: uv
• AI/ML: litellm (OpenAI, Gemini, Claude, DeepSeek, fal.ai, etc.)
• TTS: Google Cloud Text-to-Speech API
• Image Generation: OpenAI (gpt-image), Gemini, fal.ai
• Anki Integration: AnkiConnect
• Card Templates: HTML + CSS

🔧 API Services

AI Provider (via litellm)

• Text generation for definitions, examples, and mnemonics
• Image generation (Gemini) for card visuals
• Supports Gemini, GPT, Claude, DeepSeek, Qwen, and more

Google Cloud TTS

• High-quality audio generation
• Multiple voice options

AnkiConnect

• Direct communication with Anki
• Real-time card creation
• Deck management

📝 Usage Examples

Single Word

from ankinote import CardGenerator

generator = CardGenerator()
card = generator.generate("ephemeral")
generator.add_to_anki(card, deck="Vocabulary")

Batch Processing

words = ["ephemeral", "serendipity", "eloquent"]
for word in words:
    card = generator.generate(word)
    generator.add_to_anki(card)

🛠️ Development

# Clone the repository
git clone https://github.com/ignity21/ankinote-ai.git
cd ankinote-ai

# Install the project and development dependencies
uv sync

# Run tests
make test

# Format code
make format

# Type checking
make check

Documentation

• Note Types
• Skill — for using AnkiNote with AI coding assistants

📄 License

MIT License - see LICENSE file for details