4thfever/cultivation-world-simulator

What it solves

This project provides a simulation of a "cultivation world" (Xianxia) where characters are not controlled by pre-set scripts or simple state machines, but are instead driven by Large Language Models (LLMs). It solves the problem of static, predictable NPC behavior in simulation games by allowing characters to have independent personalities, memories, and the ability to make autonomous decisions based on their environment.

How it works

The simulator combines a rigorous rule-based system with LLM-driven agents. To prevent AI hallucinations and maintain world consistency, the simulation is built on a foundation of specific cultivation elements such as spiritual roots, cultivation realms, techniques, sects, and lifespan. Each NPC is an independent agent with its own personality and memory. The user acts as the "Heavenly Dao" (the world's governing force), observing the emergent narrative and occasionally intervening by triggering events like heavenly tribulations or altering character minds.

Who it’s for

• Players who enjoy emergent storytelling and world-simulation games.
• Developers interested in LLM-driven NPC agents and autonomous agent frameworks.
• Creators who want to build or modify a cultivation-themed AI world.

Highlights

• LLM-Driven NPCs: Every NPC has independent personality, memory, and interpersonal relationships, allowing for emergent plots like sect wars and power struggles.
• Rule-Based Foundation: Integrates a strict system of cultivation levels, spiritual roots, and alchemy to keep AI behavior grounded and believable.
• Heavenly Dao Perspective: The user observes the world as a god-like entity rather than playing as a single character.
• Extensible Architecture: Supports various LLM backends (DeepSeek, MiniMax, Ollama) and provides an external control API for automation and agent integration.