YiGo (弈境)

English | 简体中文

Introduction

YiGo is a cross-platform Go (Weiqi/Baduk) AI analysis and review software built entirely with native Qt C++. Inspired by Katrain, it abandons Python dependencies for faster startup, lower memory usage and more consistent cross-platform experience. Designed for Go players, learners and professionals, providing stable, lightweight and high-performance AI analysis and game record management.

Current status (v0.5.0): M1–M6 milestones complete — playable human-vs-engine games, live analysis overlay, batch review with winrate chart and blunder detection, SGF round-trip, settings persistence, and a UOS debian package.

Core Features

• Pure Qt Native: Built with C++17 / Qt 5.11, lightweight and script-free
• Human vs Engine Play: Choose board size (9/13/19), komi, handicap and sides (human/AI); engine moves via GTP genmove; two-pass game end with local scoring
• AI Engine Integration: Native GTP protocol support for KataGo / Leela Zero, with live candidate-move overlay and real-time winrate in the status bar
• Professional Review: One-click batch analysis of the main line, winrate chart with colored segments (blue = black gains / red = black loses), blunder markers (default 5% threshold), click chart to jump to any move
• Full SGF Support: Open, save, variations, handicap stones (AB/AW), compressed point ranges, multi-encoding (UTF-8 / GB18030 detection), round-trip safe
• Engine Crash Isolation: Engine runs in a separate QProcess; a crash never takes down the app — offline play and review continue
• Settings Persistence: Engine path/args, last game setup and window geometry restored across restarts
• High DPI Aware: AA_EnableHighDpiScaling, sharp rendering on high-resolution displays
• Handicap & Komi: Standard star-point handicap placement (9/13/19 boards), Chinese/Japanese scoring, two-pass endgame with local scoring

Tech Stack

• Framework: Qt 5.11 (Core / Widgets / Gui — system Qt on Linux)
• Language: C++17
• Build System: CMake ≥ 3.16 + Ninja
• AI Protocol: GTP (Go Text Protocol)
• Supported Engines: KataGo (kata-analyze / genmove), Leela Zero (lz-analyze)
• Tests: QtTest, 16 suites, fully offline (fake GTP engine; no real engine needed)

Build from Source

# dependencies (UOS V20 / Debian-based)
sudo apt install cmake ninja-build qtbase5-dev g++

# configure + build + test
cmake -B build -G Ninja -DCMAKE_BUILD_TYPE=Release
cmake --build build
ctest --test-dir build --output-on-failure

# run
./build/src/ui/yigo

Debian package (UOS V20 / deepin)

cmake --build build --target deb
sudo dpkg -i build/yigo_0.5.0_amd64.deb   # installs /usr/bin/yigo + desktop entry

Using an Engine

YiGo ships a KataGo build, so the Engine panel is pre-filled and Start works out of the box:

engines/KataGo/gtp.cfg                  # ready config
engines/KataGo/models/*.bin.gz          # weights (architecture-independent)
engines/KataGo/<arch>/katago            # one build per architecture: amd64 / arm64 / loong64 / mips64el

The architecture directory comes from YIGO_ENGINE_ARCH, detected from the host CPU (loong64 on this machine) and overridable with cmake -DYIGO_ENGINE_ARCH=arm64. cmake --build build --target deb packages only that directory plus the weights and config into /usr/share/yigo/engines/, and produces yigo_0.5.0_<arch>.deb. The config sets reportAnalysisWinratesAs = BLACK because YiGo's internal winrate convention is black's perspective.

To add another architecture, drop its katago under engines/KataGo/<arch>/ and build with that YIGO_ENGINE_ARCH.

To use another engine instead: install KataGo or Leela Zero yourself, then in the app:

1. Engine panel (right dock): pick engine type, set the executable path and arguments, press Start
2. Play against it (Game → New…, choose sides/handicap/komi), or open an SGF and press Game → Analyze Game (Ctrl+R) for batch review
3. Recommended KataGo launch: katago gtp -model <weights>.bin.gz -config gtp.cfg (set path + args in the Engine panel; the model is user-supplied)

Use Cases

• Human-vs-engine play with handicap & komi; two-pass scoring
• Game review: batch analysis, winrate chart, blunder detection (default 5% threshold)
• SGF management, teaching demonstration & variation research

License

This project is released under an open source license. Feel free to Star, Fork, submit PRs and Issues.