Our working definition: a spatial AI agent is a software system whose perception, memory, decisions, or actions are meaningfully organized by relationships in a physical or virtual environment—and whose situated behavior can be understood and redirected by people.
The first half connects spatial agency to embodied AI. AI Habitat describes embodied AI in terms of intelligent machines with a physical or virtual embodiment and studies tasks such as navigation, instruction following, and question answering. The second half adds an interface requirement. A capable system is not automatically a good collaborator. People need cues about where it is, what it can perceive, why it moved, and how to intervene.
“Spatial” therefore means more than rendering an avatar beside a chat box. Position must have consequences. Distance might change what is audible or relevant. Orientation might determine what is perceived. A room might hold persistent context. Movement might be an action, a signal, or both.