mewamew/my_ai_town

What it solves

It creates a living, autonomous simulation of a pixel-art town where residents are not scripted NPCs but AI agents. It solves the problem of static, predictable game worlds by allowing characters to make their own decisions based on personality, memory, and environmental context.

How it works

Built with Godot 4.7, the project integrates Large Language Models (LLMs) to drive resident behavior. Agents analyze their current location, profession, relationships, and memories to decide where to go and who to talk to. A world system validates actions (like working or moving) and returns actual results, which are then recorded in a global world log and individual resident memories. Players can interact with the world by chatting with residents, issuing announcements, sending photos (which the AI converts to text), or changing the weather.

Who it’s for

Developers interested in LLM-driven agent simulations, game designers exploring emergent gameplay, and users who want to observe and influence a self-evolving virtual society.

Highlights

• Autonomous Decision Making: Residents act based on context rather than fixed schedules or dialogue trees.
• Persistent Memory: Experiences, promises, and relationships are saved and influence future interactions.
• Dynamic Social Web: Relationships evolve naturally based on actual shared experiences within the simulation.
• World-State Integration: Actions like operating equipment or completing work tasks have concrete results verified by the game system.
• Player Agency: Players can directly intervene in the town's life and watch the ripple effects spread through the resident population.