Cultivation World Simulator is an open-source AI agent simulation with 2,100+ stars on GitHub that uses an agentic workflow of autonomous LLM-driven characters to grow a dynamic, emergent xianxia — Chinese cultivation fantasy — world.
Instead of scripting a story, this Python project builds a living world and lets it run. AI agents inhabit the world as cultivators with their own goals and relationships; driven by large language models and an agentic workflow, they cultivate, scheme, ally and clash, and the narrative emerges from those interactions rather than from a fixed plot. It is part game, part world-simulation experiment: procedural generation supplies the setting, prompt engineering shapes the inhabitants, and the simulation ticks forward producing events no one explicitly wrote. That makes it a fascinating testbed for questions the agent community cares about — how autonomous agents behave over long horizons, how emergent social dynamics arise, and how to keep a many-agent world coherent and affordable to run. The project is actively developed, with updates through 2026, and its bilingual presentation reflects an audience split between Chinese xianxia fans and the global AI-agent hobbyist scene.
Cultivation World Simulator is free and open source. Your only cost is the LLM API usage (or local model compute) that powers the agents, which scales with how large and long-running you make your world.
AI-agent hobbyists, simulation enthusiasts and xianxia fans who want to watch an AI-populated fantasy world evolve on its own — and developers curious about multi-agent orchestration who prefer learning from a playful, complete system.