GenerativeAgentsCN

Why we included this project

This fork makes the Stanford AI-town social simulation usable for a Chinese-speaking audience without fighting the original code. It rebuilds the Generative Agents engine on the cleaner wounderland codebase, rewrites every prompt so agents converse naturally in Chinese, and defaults to a local Qwen model served by Ollama, letting the whole town run on your own hardware with no OpenAI dependency. The research design survives the rewrite: 25 residents keep long-term memory, reflection, and daily planning, and their social behavior follows the original paper's approach. Practical touches such as checkpoint resume, template-based prompts, and a replay mode that exports both movement and dialogue to Markdown make long experiments easier to pick back up. If you want to study emergent agent behavior or build your own scenarios on a known reference, this is a maintainable starting point rather than a throwaway demo.