Play Go against a computer on a real board.
A webcam watches your physical Go board, detects your stones, and KataGo plays back. AI analysis, puzzles, voice interaction, and LLM coaching — Touch real Stone Rocks!
Capabilities
A webcam pointed at a physical Go board detects every stone in real time. Calibrate four corners, and OpenCV handles perspective correction, temporal smoothing, and confidence scoring — so only deliberate placements register as moves.
Full rule enforcement on 9×9, 13×13, and 19×19 boards. Five difficulty levels from 20k Novice to 1k Advanced using KataGo's human-like play profiles. Set up any position and play from there.
KataGo evaluates every move: win rate, score lead, top suggestions, territory ownership. Each move is graded — blunder, inaccuracy, good, or great — with color-coded indicators on the board.
Capture, Defend, Life & Death, and Tesuji decks with hints and multiple correct solutions. Track your progress per puzzle. Place positions on a real board and solve them with the camera.
Speak to the app hands-free during play. Gemini-powered speech-to-text captures your words, and text-to-speech reads responses back — so you never have to look away from the board.
An in-game chat overlay powered by Claude and Gemini. A coaching mode grounds advice in your current position. An opponent persona mode lets the AI analyze as your sitting opponent.
Getting Started
touchstone.rocks is a C++ desktop application built with CMake. You need a C++17 compiler, Raylib, OpenCV, and KataGo installed.
# Clone and build
git clone https://github.com/RaiderSoft/touchstone.rocks.git
cd touchstone.rocks
mkdir build && cd build
cmake ..
make -j$(nproc)
# Run
./touchstone