pawnbookYour chess mistakes, turned into lessons.

Self-hosted chess trainer. Play engines, get every game analysed, drill your own mistakes with spaced repetition.

♟
Human-shaped opponents
19 opponents from Maia-1100 to Stockfish-max. Maia neural networks play like real humans at each rating level — they make human mistakes, not engine mistakes.
🔍
Automatic analysis
Every game is analysed with Stockfish and Maia after you finish. Three passes extract move grades, alternative lines, and a personalised playing-strength estimate.
🗂
Spaced repetition
Mistakes become FSRS flashcard puzzles. The drill queue shows cards just before you'd forget them. An empty queue is the win state.
📖
Repertoire coach
pawnbook silently builds a personalised opening book from your own games. Once your book is established, the coach intervenes when you deviate — in the moment, before the move commits.
📈
Playing-strength tracking
Your strength is estimated from move quality (Regan-Haworth scaled error), independent of game outcomes. A rolling aggregate over your last 10 games gives a more stable reading.
🏠
Fully self-hosted
No accounts, no cloud sync, no telemetry. Your games, repertoire, and card deck stay on your machine.

How a session works

Four timescales, one loop:

Move (seconds) — play a move, see the evaluation.
Game (10–30 min) — play to a conclusion against an engine opponent. Analysis runs automatically when you finish.
Session (15–40 min) — one game plus drilling the mistakes from last time, or drill-only.
Improvement (weeks) — Elo rises, the drill queue reflects your actual weaknesses, the opening book adapts to how you really play.

bash

git clone https://github.com/JohnnyFoulds/pawnbook.git
cd pawnbook
docker compose up

See Installation for the full setup guide including engine weights and native (non-Docker) paths.