ChessRL - Terminal-based Chess with Reinforcement Learning

A terminal-based chess game featuring a pure reinforcement learning engine that learns and improves during gameplay.

Features

• Play chess directly in your terminal
• Clean TUI interface with move history and analytics
• Real-time engine analysis and thinking process display
• Pure reinforcement learning engine that improves as you play
• No pre-trained models - watch the engine learn from scratch each game

Installation

curl -sSL https://raw.githubusercontent.com/frgmt0/chessrl/main/install.sh | bash

Usage

Start a new game:

chessrl

Making Moves

Moves are made using algebraic coordinates. Format: <from square> <to square>

Examples:

• Move pawn from a2 to a4: a2 a4
• Move knight from b1 to c3: b1 c3

Controls

• Scroll: View move history
• q: Quit game
• ESC: return to main menu (game is not saved)

Engine

The reinforcement learning engine:

• Learns purely through self-play during each game
• Does not persist learned knowledge between games (yet)
• Displays its thinking process and move confidence in real-time
• Improves noticeably as the game progresses

Coming Soon

• Persistent learning (saved models)
• Opening book integration
• Additional analysis features
• Network play
• recommended moves

License

MIT