wtf is pekingman?

Pekingman is a research project that builds an agent system meant to give virtual characters, like NPCs in games or digital humans, more lifelike and consistent behavior over time. It ties together four capability areas: perception, memory, reasoning, and action, so a character can notice changes in its environment, remember past events, decide what to do based on that history and its current situation, and then act in a way that stays consistent with how it has behaved before. The project describes a persistent memory system that lets a character store, retrieve, and revisit earlier interactions, rather than resetting its understanding each time. The authors say this is meant for scenarios like virtual NPCs, digital human projects, metaverse platforms, and simulation games, particularly larger or more complex worlds where simple trigger based behavior is not enough to feel natural. According to the README's news section, the full code was released in August 2026, alongside a GitHub Pages demo page showing a project video. To run it, you need Python 3.11 and a Conda environment, set up through a provided environment file, plus optional pieces like GPU acceleration for more advanced features, a 3D simulation engine, and separate optional services for storage, model access, and speech processing that are not required just to view the documentation or demo page. There is also a static demo page you can run locally with Python's built in web server. The README notes this is an academic project associated with researchers at MIT, USC, and Université Bourgogne Europe, and includes a citation entry for anyone using it as a research artifact. It also warns that several advanced capabilities are still experimental and may not work if their optional services are not set up, and that any public release should avoid including local credentials or private data. The README does not state a software license.