Mining

A local tool for a Japanese sentence-mining Anki workflow: ingests YouTube videos, scans their subtitles for vocabulary not already in your deck, and lets you review/select new cards before adding them — with generated audio (VOICEVOX), definitions (JMdict), and furigana filled in automatically. It also handles field-mapped importing from other decks, an on-demand .apkg backup of your deck, and an audit of notes missing required fields.

Works directly against a live Anki collection via the AnkiConnect add-on — nothing here touches Anki's database file directly.

See SPEC.md for the full design/architecture writeup.

Prerequisites

• Anki desktop, with the AnkiConnect add-on installed (Tools > Add-ons > Get Add-ons in Anki, code 2055492159) and Anki running whenever you use this tool.
• uv — manages the Python version and dependencies for you; nothing else to install manually.
• ffmpeg on your PATH (e.g. brew install ffmpeg) — used for audio clipping.
• A note type with the same field layout as the "Lapis" note type (Expression, ExpressionReading, Sentence, SentenceAudio, MainDefinition, etc.) on the deck you point this at. app/config.py's TARGET_DECK/NOTE_TYPE say which deck/note type the tool operates on — change them if yours are named differently.
• Optional, for full functionality:
  • VOICEVOX, installed and running, for ExpressionAudio generation.
  • A JMdict term-bank and a CC-100 frequency-bank dictionary, both in Yomitan's zip format, dropped into data/dictionaries/ — required for the SRT vocabulary scanner (§7 of SPEC.md).

Setup

uv sync

Run

uv run python main.py

Opens on http://127.0.0.1:8123. Anki (with AnkiConnect) needs to already be running. On first start, your local config/*.yaml files (ignore list, voice exclusions, required fields) are generated from the tracked .yaml.example templates if they don't already exist — see .gitignore for why those aren't tracked directly (they end up holding your personal word/voice choices).