Under review.
Everyday AI assistance depends on the places we revisit, yet users often have to describe their surroundings again with each interaction. Ivy connects activity-oriented agents with persistent Place Memory, preserving a place's contents and spatial organization independently of individual agents and conversations. Agents combine this environmental knowledge with activity experience and personal preferences, so observations made at home can inform assistance later at a grocery store. Through phone and smart-glasses interfaces, users capture their surroundings, converse with agents, and inspect and revise memories as their environments change. A four-day in-the-wild deployment with eight participants explored how remembered resources and spatial constraints shaped everyday recommendations, highlighting both the value of reusable place knowledge and the need for clearer feedback about memory updates and use.
