Relational agents
How should an intelligent system behave when the unit of interaction is a relationship rather than a prompt? We are interested in initiative, repair, boundaries, personality, and the accumulation of shared context.
Our research
Hug
- Status
- Current project
- System
- Relational agent
- Interface
- Messaging
- Public record
- hug.wtf ↗
A persistent AI friend inside the places people already talk.
HUG explores whether an AI can feel less like a tool that is summoned and more like a socially present participant. It operates through familiar messaging interfaces, maintains durable context, and is designed around short, conversational exchanges rather than formal assistant responses.
The project gives TAICU a concrete environment for investigating memory, personality, proactive interaction, emotional expectations, and the ways relational systems can fail.
Visit Hug ↗Research agenda
Durable personal context
What should an AI remember, forget, revise, or refuse to retain? We want memory systems that support continuity while remaining legible, correctable, and proportionate.
Safety in significant interactions
Relational systems create risks that benchmark-only evaluation cannot see. We aim to study influence, dependency, emotional failure, and recovery over time and in context.