Wolfcha is an AI-native game project that addresses the challenge of playing the social deduction game Werewolf without needing a human group. It targets players who enjoy the strategic and deductive elements of Werewolf but struggle with coordination, as well as AI enthusiasts interested in observing LLM capabilities. The project offers a unique way to experience Werewolf anytime, anywhere, by pitting multiple AI models against each other in a simulated game environment.

How It Works

The core innovation is a Dual-Layer AI Roleplay system. First, an AI acts as a "virtual player" with a defined personality and background. This virtual player then adopts a specific Werewolf role (e.g., Seer, Werewolf) and engages in real-time conversation, bluffing, and logical deduction based on the game's state. This approach leverages LLMs' growing context windows and instruction-following abilities for dynamic and unpredictable gameplay. Additionally, the project features an AI Model Arena, integrating multiple top-tier AI models (like DeepSeek, Qwen, Gemini) that compete as players, allowing users to observe and compare their performance, effectively acting as a "Turing test" for AI reasoning.

Quick Start & Requirements

To run locally:

1. Clone the repository: git clone https://github.com/oil-oil/wolfcha.git
2. Navigate to the directory: cd wolfcha
3. Install dependencies using pnpm (recommended) or npm: pnpm install or npm install
4. Start the development server: pnpm dev

Full functionality requires configuring API keys for services like ZenMux and Minimax, detailed in the .env.example file.

Highlighted Details

• Immersive Retro Design Style with clean layouts and vintage color palettes.
• Dynamic UI/UX features include eye-blink transitions for day/night cycles, character lip-sync animations, and unique character portraits for special roles.
• Multimodal experience powered by real-time voice generation from Minimax, giving each AI character a distinct voice.
• AI Model Arena allows direct comparison of leading LLMs competing within the game.

Maintenance & Community

The project is supported by sponsors including ZenMux, Dashscope, Minimax, and OpenCreator, who power core features like AI model integration, voice synthesis, and character portrait generation. Specific community channels (e.g., Discord, Slack) or direct contributor information are not detailed in the README.

Licensing & Compatibility

The project is released under the MIT license. This license is highly permissive, generally allowing for commercial use, modification, and distribution, including within closed-source applications.

Limitations & Caveats

The current implementation focuses on AI-vs-AI gameplay; multiplayer modes with human players are listed as a future roadmap item. Mobile optimization, flexible player counts, and advanced game mechanics like "time rewind" and "AI insight" are still under development. Achieving full functionality necessitates obtaining and configuring API keys for several third-party AI services.