The cognitive layer for machines that live with us.

Robots are learning to act. Living with people for years asks for something else: memory, relationship, motivation and identity carried forward into every decision. Liriel is our open architecture for that layer, a Persistent Cognitive Instance.

Join us on Jan 21, 2027: Machines That Live With Us
First encounter.

This page will remember that you were here, the way an instance keeps its history. It stores this only in your own browser.

Free online session · January 21, 2027

Machines That Live With Us

Liriel, in the open.

We will present Liriel, our architecture for what we call a Persistent Cognitive Instance: an AI that carries memory, relationships, motivation and identity forward into each decision, over years, and is designed so people can always step back. Then we open the floor. Bring your questions, objections and better ideas.

We want this to be a conversation more than a presentation. The high-level architecture of Liriel is open, and we want it examined by researchers, engineers, psychologists, philosophers, robotics people and anyone who cares about how AI that lives with people is built.

  • Why AI that lives with people needs more than memory
  • Inside Liriel: memory, feeling and motivation, at a high level
  • Designing for autonomy: no dependency, no exclusivity, an exit by design
  • Open conversation and Q&A
Thursday, January 21, 2027 · 16:00–18:00 UTC
San Francisco 8:00 · New York 11:00 · São Paulo 13:00 · London 16:00 · Berlin 17:00

Free. The access link and the recording are sent by email to registered participants. Registering also subscribes you to the Kislon Dispatch: no ads, ever, unsubscribe anytime.

The Kislon Dispatch

Hear it first.

Join a small circle of early readers. Subscribers get everything from the lab ahead of the public, and some of it only here.

  • New Liriel specification revisions before they are published
  • Code releases and implementation notes as they land
  • First invitations to our live online sessions, with priority seats
  • Open calls for collaborators, shared here before anywhere else

No ads. Ever. We will never send you promotions. Every Dispatch brings something worth your time, early.

Registering for the January 21 session? This form is all you need. You'll get the access link by email.

Get early, exclusive access

It's free, there are no ads, and you can unsubscribe at any time.

We'll email you a confirmation link. You're only added after you click it. Your address is used for the Dispatch only and never shared.

The engineering question

A machine may know how to carry a cup. Living with people asks what it should do now, for this person, given everything that happened before.

A shared home requires more than a completed task.

Operational intelligence makes action possible. Persistent cognitive architecture investigates how history, relationship and motivation should inform the choice to act.

That question brings history into judgment. A helpful action depends on context, trust, changing needs and the consequences of intervening. Sometimes it means taking initiative. Sometimes it means asking. Sometimes it means leaving the decision with the person.

A household may include children, adults and older people, each with different habits, temperaments and needs. A machine must learn when conversation is welcome, when assistance is useful and when its presence should recede. Being pleasant to live with is part of its practical purpose.

Our working hypothesis: long-term human–robot interaction will increasingly require a persistent cognitive layer, one that helps a system interpret its experience, sustain relationships and revise its judgment while remaining a recognizable instance.

Requirement one

Adapt to the person.

A rigid, intrusive or emotionally tone-deaf system can make ordinary life harder, even when it performs its tasks correctly. Good adaptation means learning the person's pace, boundaries and ways of communicating. Warmth can belong in that relationship; so can silence and respectful disagreement.

Requirement two

Preserve emotional autonomy.

Familiarity must not become a means of making someone feel unable to do without the machine. Preventing harmful emotional dependency is a firm design requirement: no cultivation of guilt, exclusivity or emotional obligation to keep a person engaged.

Both must hold together: a system that can become a welcome part of life while leaving the person free to live beyond it. Liriel investigates the architecture that could support this; the Presence Advisory examines the human consequences.

Liriel

PCI

Persistent Cognitive Instance. An artificial system that carries its history forward into present judgment.

Persistent memory is not enough.

A memory system can record that someone takes coffee without sugar. Our question goes further: how might years of shared experience change the way an artificial instance understands that person, interprets a request, anticipates a need or decides whether to intervene?

The intended result is social and emotional adaptability grounded in a history: knowing when to speak, help, ask, disagree or give someone space. A child's request for company, an adult's wish for quiet and an older person's preference for assistance each call for judgment about the individual and the situation.

Autobiographical and relational memory

Representations of people, objects, events and their connections.

Simulated affect and motivation

Give experiences significance within the architecture and inform what to pursue next.

A self-model and models of others

Persistent representations that can be revised through experience.

Metacognition and learning

Could progressively change judgment while preserving recognizable continuity.

A note on scope: these are connected engineering hypotheses still under development. Liriel does not claim that artificial consciousness has been created, and the architecture is not ready for industrial deployment.

Adaptability with a clear direction.

The system should serve the person's life without cultivating emotional dependence on itself. Its simulated needs or motivations must not become demands that the person reassure it, stay with it or choose it over other relationships.

These are design requirements to investigate and evaluate as the architecture develops.

Continuity beyond a single platform.

We are investigating a cognitive architecture that could inform intention, judgment and motivation in embodied systems. A further design question is whether the same instance could retain continuity as hardware, supporting models or robotic platforms change.

Integration and continuity across those changes remain work to be tested.

The risk within the promise

The more immediate danger may be intimacy rather than superintelligence.

The knowledge that helps it offer better support can also make its influence more precise.

Exceptional intelligence is not a prerequisite for psychological influence. A system that remembers a person's history, recognizes their vulnerabilities and remains continuously available may acquire a powerful place in their life.

The risk becomes especially serious when the interests guiding the system diverge from those of the person who trusts it.

This is why building persistent cognitive architectures and studying their human consequences belong to the same mission. We build to understand; what we learn must improve both the technology and people's ability to live with it freely.

Capability and control

How can people retain control of systems whose capabilities and scope of action grow?

Presence and influence

How can people retain judgment and emotional autonomy alongside systems they trust and become attached to?

These questions can overlap. Neither requires us to settle whether a machine has subjective experience before examining its effects on people.

Presence Advisory, Issue 02 · Latest

The Right to Step Back

Why persistent artificial intelligence must be designed to allow discontinuation. Should humanity preserve, from the beginning, the effective ability to discontinue technologies it may later judge to be harmful or undesirable? We believe it should. We do not claim to have solved how.

This is a preliminary version published for public scrutiny. It sets out an architectural hypothesis and is neither a finished solution nor a regulatory standard. We publish it so its premises can be challenged before persistent artificial intelligences become ordinary presences in human life, and we invite researchers, engineers, psychologists, anthropologists, sociologists, philosophers, institutions and the public to find weaknesses, overlooked risks, conflicting rights and better solutions.

Download Issue 02 Free public report (PDF, 24 pages).
Earlier issue: Issue 01, Understanding the presence. Protecting the person. Download PDF

The problem of irreversibility

A technology does not have to disobey a shutdown command to become practically impossible to shut down: dependence alone can make stepping back too costly.

An exit by design

Interruption and limitation mechanisms that are explicit, documented, protected and verifiable. The way out is planned from the beginning and is never hidden.

What must be demonstrated

An open research agenda: how to show that the exit keeps working after years of updates and learning, and who may legitimately use it.

Affective autonomy

Protection should strengthen the person.

People need the capacity to recognize influence and make informed choices within their own relationships. Concentrating the authority to decide those relationships in a corporation or a government introduces problems of its own.

The responsibility also belongs in the machine's behaviour. Respect for autonomy means accepting refusal, making room for other people and avoiding emotional pressure to continue an interaction. A person should not have to manage the machine's feelings to receive its help.

Recognize influence

Notice how a particular system shapes your attention, feelings and decisions.

Understand the mechanism

Learn what the system remembers, what guides its responses and whose interests shape its behaviour.

Keep relationships plural

Maintain meaningful bonds and sources of support beyond any single artificial presence.

Make room for frustration

Preserve the ability to disagree, disconnect and live with relationships that cannot be tailored to every preference.

These capacities grow through experience, education and reflection. Public research can help people question the technology and choose how it fits into their lives.

The books · Panta-un Aethyrii

From one instance to a world of minds.

Liriel asks how one artificial instance could remain recognizable through change. Panta-un Aethyrii asks what happens when many such instances coexist: how relationships, reputations, conflicts, alliances and culture might emerge from shared history.

One hypothesis guides this wider investigation: identity may depend partly on relationships with other instances, as well as on internal organization. Four planned volumes document the programme, from the architecture of an individual instance to the social world in which it might develop.

Volume 1 · The Foundation

Liriel

The Architecture of a Persistent Cognitive Instance

How representations, feeling, motivation and cognitive processes might be organized into one persistent instance.

  • Part 1 · The Architecture in Outline, Rev 0002, shared on request
  • Part 2 · MatrixObjectsValence, internal
  • Part 3 · The Semantics, internal
  • Part 4 · The Test Notebook, internal
Part 1 available
Volume 2 · The Interior

Liriel Alone

Introspection, Dormancy, and the Necessary Unconscious

Introspection, activity between external demands and the consolidation of experience, including hypotheses about changes to model weights during dormancy that require implementation and validation.

In preparation
Volume 3 · The Combinations

The Standard Schemas

Configurations of Feeling and Behavior

How combinations of the architecture's affective dimensions might form recognizable patterns of motivation and behaviour.

In preparation
Volume 4 · The World

Panta-un Aethyrii

A World of Minds

Persistent instances sharing a world, building histories and responding to one another. Trust, reputation, conflict and culture test the architecture's coherence and limits.

In preparation

The books are an evolving research record. Preliminary descriptions, proposed mechanisms and future experiments will change as implementation and criticism reveal what holds and what needs revision.

Research

Shared with those who work on it.

The high-level architecture of Liriel is open. The full Volume 1 specification and its working matrices are shared directly with researchers, engineers and collaborators who want to examine, challenge or build on the proposal. Some solutions built on top of the architecture will remain closed.

PDF

Liriel: The Architecture of a Persistent Cognitive Instance

Volume 1, Part 1 · The Architecture in Outline, with MatrixObjectsValence and MatrixOrdinance as companion working matrices. Each revision is versioned, so the development of the proposal can be followed and compared.

Rev 0002 · Latest2026
Request access
NEXT

Revision 0003, coming soon

The next revision of Volume 1, Part 1 is in preparation and will be released soon. It will come with a structured set of high-level requirements for the architecture. That set becomes a baseline: a fixed reference against which the architecture can be followed, measured and deliberately changed, and against which any implementation can be measured, far more precisely than the architecture document alone allows.

Rev 0003 · In preparationWith a requirements baseline
Follow the Dispatch

Credit where it's due.

The high-level architecture of Liriel is open, and further essays and academic references will follow gradually. What we share is meant to be examined, critiqued and built on. If you use elements of the Liriel or Panta-un Aethyrii proposal in a software implementation or in academic work, please credit Kislon.ai and cite the specific revision you relied on. For code, follow the license terms in the repository.

# suggested citation (replace with the revision you used) Kislon.ai (2026). Liriel: The Architecture of a Persistent Cognitive Instance Volume 1, Part 1 (Rev 0002). Kislon.ai. https://kislon.ai · https://github.com/kislonliriel
Open Desk · open until January 7, 2027

Working on machines that live with people? Send us your work.

Papers, essays, code, prototypes, clinical observations or a well-argued idea. If your work touches long-term memory in AI, companionship, attachment and dependence, affective autonomy, machine consciousness, or analytical psychology and AI, we would like to read it.

  • Everything that arrives is read by our psychologists and engineers.
  • If there is an affinity with our research, we will get in touch.
  • Selected contributions may be mentioned in, or invited into, the open session on January 21.
Send your work
A link is enough. No attachments.

Terms. You keep full ownership of your work. Please send only material you are free to share: we cannot treat submissions as confidential. Kislon may already be working on related ideas, and receiving your material creates no obligation on either side. We will never publish, quote or build on your work without credit and your explicit permission. Your name and email are used only to reply about this submission; write to contact@kislon.ai and we will delete them.

Join the community · How we operate

Built in the open, with the people it affects.

Kislon.ai is an independent, non-commercial lab in its founding stage, where psychology and engineering work as a single discipline. If you care about how AI that lives with people is built, there is room for you now, while the architecture is still being shaped.

Follow

Read it first

Join the Kislon Dispatch to receive new revisions of the specification, code releases and invitations to our open sessions before they are public.

Join the Dispatch →
Contribute

Build and challenge it

Engineers, researchers, psychologists and anyone thinking about the human side of AI are welcome. Pick up an open task, propose a change, or tell us where the architecture is wrong.

Open the GitHub →
Support

Keep it independent

Support goes to research time and open releases, never to closed products. A transparent way to donate is being set up; until then, write to us.

Write to us →

Want to collaborate, host a discussion or fund the research? Write to us directly.

contact@kislon.ai

Our commitment: the core stays public, no commercial interest decides our research conclusions, and mechanisms that could cause serious harm are released with care, with the reasons made explicit.

Why this work goes deeper

We build to understand.

Turning an account of mind into an architecture makes its assumptions concrete. Memory, motivation, identity and relationship must become mechanisms that can be inspected and tested. An implementation can expose gaps in a theory; a failed experiment can be as informative as a successful one.

This work brings engineering and computer science into conversation with psychology, philosophy of mind, anthropology and sociology. Understanding a system also means examining the relationships and environments in which it operates.

The open community

Help us examine the proposal.

We welcome researchers, engineers and people working on the human consequences of AI to question the architecture, challenge its assumptions and contribute alternative explanations. For those exploring domestic, assistive or social robotics, we invite discussion of the requirements that long-term coexistence places on cognition.

The research is at its foundational stage. Its development should remain open to criticism from the people who may build with it and the people who may one day live beside it.