LLM Chess ️🤖

Play chess against local LLMs, or watch two AIs battle each other — all in a beautiful dark-themed GUI.

LLM Chess is a desktop application that lets you play international chess against AI opponents powered by local or cloud large language models. Supports Ollama, llama.cpp, LM Studio, OpenAI GPT, DeepSeek, and any OpenAI-compatible API.

✨ Features

🎮 Game Modes

• Human vs AI — Classic mode. Play as White or Black against an LLM.
• AI vs AI — Watch two LLMs play each other with configurable speed control (pause/resume/step).
• Each side can use a different model, different backend, different temperature.

💭 AI Thinking Display

• See the AI's reasoning process for every move
• Color-coded by side (blue for White, pink for Black)
• Parses Reasoning: and Move: format with fallback handling

🎭 AI Personas

Choose the AI's "personality" — affects its playing style and reasoning tone:

• Default — Calm, professional engine
• Aggressive — Loves attacks and sacrifices
• Defensive — Values king safety and solid positions
• Creative — Unusual openings and tactical surprises
• Teacher — Explains reasoning clearly, great for learning

🎨 UI Highlights

• Dark Catppuccin-themed PyQt6 interface
• Click-to-move with legal-move highlighting
• Last-move markers (yellow) and check indicators (red)
• Full move history in SAN notation
• Real-time FEN display
• Promotion dialog (queen/rook/bishop/knight)
• Undo moves

📦 Installation

Option 1: pip (recommended)

pip install llmchess
llmchess

Or run as a module:

python -m llmchess

Option 2: npm (Node.js)

npm install -g llmchess
llmchess

The npm wrapper will auto-install the Python llmchess package on first run.

Option 3: DEB package (Ubuntu/Debian)

sudo dpkg -i llmchess_1.5.5_all.deb

The launcher will auto-install missing Python dependencies (PyQt6, python-chess, httpx).

Option 4: From source

git clone https://github.com/oemoem12/LLMChess.git
cd LLMChess
pip install -e .
python main.py

🔧 Setup LLM Backend

LLM Chess is backend-agnostic — it speaks OpenAI-compatible HTTP API. Pick any one:

Ollama (easiest)

# Install from https://ollama.com
ollama pull qwen2.5:7b
ollama serve  # default: http://localhost:11434

llama.cpp

./llama-server -m model.gguf --port 8080

LM Studio

Open LM Studio → Developer tab → Start Local Server (default: http://localhost:1234)

🚀 Quick Start

1. Start your LLM server (Ollama/llama.cpp/LM Studio)
2. Launch LLM Chess: llmchess
3. Open Settings → select your backend → click Test Connection
4. Choose a model in the connection settings
5. Pick a game mode (Human vs AI / AI vs AI) and a persona
6. Click a piece → click target square to move
7. Watch the AI Thinking panel to see your opponent's reasoning

🎬 Screenshots

┌─────────────────────┬──────────────────┐
│  Game Mode          │  Move History    │
│  [Human vs AI ▼]    │  1. e4 e5        │
│  [White (first) ▼]  │  2. Nf3 Nc6      │
│                     │  3. Bb5 a6       │
│  ♔ ♕ ♖ ♗ ♘ ♙        │  ...             │
│  ─────────────────  │                  │
│  Chess Board        │  AI Thinking     │
│  (8×8)              │  ━━━ White AI ━━ │
│  ─────────────────  │  Move: e2e4      │
│  [New Game] [⚙ Set] │  Reasoning:      │
│                     │  Classical king  │
└─────────────────────┴──────────────────┘

🛠️ Tech Stack

• PyQt6 — Cross-platform GUI
• python-chess — Chess rules, FEN/PGN handling
• httpx — OpenAI-compatible HTTP client
• setuptools + Trusted Publisher — Zero-token PyPI release

📐 Architecture

chess_app/
├── __init__.py            # Package entry, version, main()
├── __main__.py            # python -m chess_app support
├── main.py                # GUI entry point
├── board_widget.py        # Chess board renderer (PyQt6)
├── game_controller.py     # Main window + game flow + AI vs AI logic
├── llm_connector.py       # OpenAI-compatible LLM client + persona prompts
└── settings_dialog.py     # Tabbed config UI (White/Black sides)

🤝 Contributing

PRs welcome! Some ideas:

• [ ] Save/load PGN files
• [ ] Tournament mode (round-robin between N models)
• [ ] Stockfish-LLM hybrid (use Stockfish for blunders, LLM for variety)
• [ ] Post-game analysis with LLM commentary
• [ ] Online multiplayer via WebSocket

📜 License

MIT — do whatever you want, just don't blame me if the AI hangs your king.

🔗 Links

• PyPI: https://pypi.org/project/llmchess/
• GitHub: https://github.com/oemoem12/LLMChess
• Issues: https://github.com/oemoem12/LLMChess/issues
• 中文 README: README_zh.md