Syntax Brawlers - LLM Arena Fighting Game

Two AI-controlled fighters battle in a boxing ring, with each LLM generating tactical decisions and trash talk based on the fight context.

Features

• Real LLM Integration: OpenRouter (DeepSeek), Anthropic Claude, OpenAI GPT, Ollama
• 7 Combat Actions: Jab, Cross, Hook, Uppercut, Block, Dodge, Clinch
• 4 AI Personalities: The Destroyer, The Tactician, The Ghost, The Wildcard
• Advanced Combat: Combo system, counter attacks, critical hits, stagger
• Visual Effects: Particle effects, screen shake, damage popups
• Professional GUI: Health/stamina bars, AI thoughts display, round system

Requirements

pip install pygame httpx numpy

Setup

Set your LLM API key:

# OpenRouter (recommended - supports DeepSeek v3.2)
export OPENROUTER_API_KEY="your-key"

# Or Anthropic
export ANTHROPIC_API_KEY="your-key"

# Or OpenAI
export OPENAI_API_KEY="your-key"

Run

python syntax_brawlers.py

Controls

• Mouse - Menu navigation
• ESC - Pause / Menu
• R - Restart match
• SPACE - Skip text animation

AI Personalities

Personality	Style	Signature Move
The Destroyer	Aggressive pressure	Hook
The Tactician	Strategic, analytical	Cross
The Ghost	Defensive, counter	Dodge
The Wildcard	Unpredictable chaos	Uppercut

License

MIT