Robots are learning to act: folding laundry, carrying a cup, opening a door. The harder question arrives later, once a machine stays. What should it do now, for this person, given everything that has happened before?

That is the question behind Liriel, the architecture we are building at Kislon. On January 21 we will present it in the open, and then hand the floor to you.

Why memory is not enough

A system can remember that someone takes their coffee without sugar. Living with people for years asks for more: knowing when to help and when to ask, when conversation is welcome and when to stay quiet, how a request from a child differs from the same words from an adult.

We call a system built for this a Persistent Cognitive Instance: an AI that carries memory, relationships, motivation and identity forward into each decision, while remaining recognizably itself.

What we will show

  • How Liriel represents the people, things and events in its life, and what each of them means to it.

  • How simulated affect and motivation give experience weight and shape what it pursues next.

  • How it models itself and others, and how those models are revised over time.

All of this at the level of the high-level architecture, which is open. Some more specific elements built on top of it will remain closed.

Designing for autonomy

The same knowledge that lets a system help can make its influence more precise. So some requirements are built in from the first line: no cultivation of guilt or exclusivity, no promises it cannot keep, room for the people who can be there in ways it cannot, and an exit by design, so a person can always step back.

Then, the conversation

This is a working session, not a keynote. We want questions and objections from researchers, engineers, psychologists, philosophers, people building home and social robots, and anyone who may one day live beside these systems. If something in the architecture is wrong, we would rather hear it now.

Practical details

  • When: Thursday, January 21, 2027, 16:00–18:00 UTC (8:00 San Francisco, 11:00 New York, 13:00 São Paulo, 16:00 London, 17:00 Berlin)

  • Where: online, free, in English

  • How: register at kislon.ai/#session. The access link and the recording are sent by email, so registering is worth it even if you cannot attend live.

— Kislon