Vox Deorum (2025-Present)
Links
In 2025, I designed and developed Vox Deorum, a hybrid LLM architecture for 4X/Grand Strategy game AI in Civilization V. My responsibilities included:
- Designed a novel hybrid LLM+X architecture that separates macro-strategic reasoning (handled by LLMs) from tactical execution (handled by specialized subsystems)
- Built a multi-component technical system connecting Civilization V with AI agents through Bridge Service, MCP Server, and Vox Agents
- Conducted extensive testing with 2,327 complete game simulations demonstrating competitive end-to-end gameplay with unique play styles
Vox Deorum represents a significant advancement in integrating Generative AI models into complex strategy games, addressing challenges of latency, cost-effectiveness, and multi-agent collaboration while maintaining engaging gameplay through natural human-AI interactions.
Key Features
- LLM-Enhanced AI Opponents: AI civilizations powered by Generative AI models for strategic decision-making
- Hybrid Architecture: Layered technical design that delegates strategic reasoning to LLMs while specialized systems handle tactical execution
Design Philosophy
Vox Deorum explores how LLMs can handle macro-strategic reasoning in complex strategy games while maintaining computational efficiency. Unlike general-purpose game AI, this hybrid approach leverages LLMs’ strengths in contextual understanding and natural language while delegating precision tasks to specialized subsystems, creating more natural and engaging human-AI gameplay interactions.
Impact
Through extensive testing across 2,327 complete game simulations, the system demonstrated that LLMs can achieve competitive gameplay performance with distinct strategic personalities. This work opens new opportunities for game design and agentic AI research by establishing a viable architecture for integrating LLMs in commercial strategy games.
Related Publications
- Chen, J., et al. (2025). Vox Deorum: A Hybrid LLM Architecture for 4X / Grand Strategy Game AI – Lessons from Civilization V. arXiv preprint arXiv:2512.18564.