AnkiThis

Convert any educational document into a high-quality, study-ready Anki flashcard deck.

AnkiThis uses a multi-pass AI pipeline to read your documents, identify the most important concepts, and generate cloze-deletion and Q&A flashcards optimized for long-term retention. Upload a PDF, DOCX, TXT, or Markdown file and get back a polished .apkg deck you can import directly into Anki.

How It Works

AnkiThis runs a six-stage pipeline designed to produce cards of pedagogical value that you'd actually want to study:

1. Parse & Structure — Extracts text from your document, detects section boundaries, and splits content into manageable chunks.
2. Concept Extraction — An LLM identifies key concepts, definitions, mechanisms, and relationships in each chunk.
3. Concept Merge — Duplicate and overlapping concepts are merged across sections. Concepts are ranked by importance.
4. Card Planning — The system decides which concepts deserve cards, what card type to use (cloze vs. Q&A), and how many cards to generate — scaled proportionally to document length.
5. Card Generation — High-quality flashcards are generated with precise cloze deletions (1–4 word blanks) and clear Q&A pairs.
6. Quality Control — A critique pass rewrites weak cards, deduplicates near-identical ones, and suppresses anything that doesn't meet the quality bar.

The result is a compact, curated deck — not a bloated dump of every sentence in your textbook.

Features

• Multi-format support — PDF, DOCX, TXT, and Markdown
• Configurable generation — Choose your study goal, card style, and deck size
• Two card types — Cloze deletions for terminology and definitions, Q&A for mechanisms and comparisons
• Quality pipeline — AI-powered critique + deterministic filters catch bad cards before they reach your deck
• Review before export — Preview cards, remove ones you don't want, regenerate with different settings
• Anki-ready export — Download as .apkg (native Anki package) or CSV

Configuration Options

Option	Values	Description
Study Goal	Free text (optional)	e.g. "Prepare for organic chemistry midterm"
Card Style	cloze_heavy, qa_heavy, balanced	Preferred card format
Deck Size	small, medium, large	Card density — scales with document length

Deck size uses adaptive density rather than fixed card counts. A short 5-page PDF on "Fewer" produces ~8 cards, while a 30-page chapter on "Balanced" produces ~108. Floor: 5 cards, ceiling: 300.

Tech Stack

• Backend: Python, FastAPI, Celery, SQLAlchemy
• Frontend: Next.js, TypeScript, Tailwind CSS
• Database: PostgreSQL
• Queue: Redis + Celery
• LLM: Anthropic Claude API
• Export: genanki (.apkg), CSV

Getting Started

Prerequisites

• Docker and Docker Compose
• An Anthropic API key

Setup

# Clone the repo
git clone https://github.com/logannye/ankithis.git
cd ankithis

# Create your environment file
cp .env.example .env
# Edit .env and add your ANKITHIS_ANTHROPIC_API_KEY

# Start all services
make up

# Run database migrations
make migrate

The API will be available at http://localhost:8000 and the web UI at http://localhost:3000.

API Endpoints

Method	Endpoint	Description
POST	/api/auth/register	Create an account
POST	/api/auth/login	Sign in and get a token
POST	/api/upload	Upload a document
POST	/api/documents/{id}/generate	Start card generation
GET	/api/jobs/{id}	Poll generation progress
GET	/api/documents/{id}/review	Review generated cards
POST	/api/cards/{id}/remove	Remove a card from the deck
POST	/api/documents/{id}/regenerate	Regenerate with new settings
GET	/api/documents/{id}/export/apkg	Download Anki package
GET	/api/documents/{id}/export/csv	Download CSV

All endpoints except auth require a Bearer token in the Authorization header.

Interactive API docs are available at http://localhost:8000/docs.

Development

# Run backend tests
make test-api

# Lint backend code
make lint

# View logs
make logs

# Open a database shell
make shell-db

Project Structure

ankithis/
├── apps/
│   ├── api/                    # FastAPI backend
│   │   ├── src/ankithis_api/
│   │   │   ├── app.py          # FastAPI application
│   │   │   ├── config.py       # Settings (env-based)
│   │   │   ├── db.py           # Database engine
│   │   │   ├── worker.py       # Celery worker
│   │   │   ├── models/         # SQLAlchemy models
│   │   │   ├── routers/        # API route handlers
│   │   │   ├── schemas/        # Pydantic request/response schemas
│   │   │   ├── services/       # Business logic + pipeline stages
│   │   │   └── llm/            # LLM client + prompt templates
│   │   ├── alembic/            # Database migrations
│   │   └── tests/
│   └── web/                    # Next.js frontend
│       └── src/app/
├── docker-compose.yml
├── Makefile
└── .env.example

Best Results

AnkiThis works best with:

• Digital textbook chapters (10–80 pages)
• Typed lecture notes and study guides
• Review articles and explanatory writing
• Reasonably structured documents with headings

It is not designed for handwritten notes, image-heavy slides, or scanned documents with poor OCR quality.

License

MIT