01Established field
Predictive world models
Research systems can learn representations that support prediction, planning, and action in physical or simulated environments. This provides the analogy—not proof—for an internal world model.
Research map
Separate what is known, what we are building, and what remains a hypothesis.
This thesis connects mature research areas with a new product direction. The connections are ours. We do not present them as scientific consensus or claim that the hard problems are already solved.
02Established research
Interoception + predictive processing
Work in neuroscience and consciousness research examines how the brain integrates signals from within the body and predicts internal state. This supports the importance of embodied experience; it does not imply that software can directly know subjective experience.
03Established field
Subjective wellbeing
Decades of research distinguish life evaluation, affect, context, and individual differences. Wellbeing is multidimensional and partly subjective, which is exactly why a single external optimization target is inadequate.
04Product hypothesis
Longitudinal personal context
We hypothesize that permissioned memory, a whole-life ontology, and explicit user correction can make AI support less generic and more useful across time. Product evidence must test this rather than assume it.
Open questions
The research agenda is as much about restraint as capability.
How can a person inspect and correct a model that necessarily contains uncertainty?
Which information is genuinely useful, and which should never be collected?
How should memory expire, branch, or be deleted as a person changes?
How can support learn from outcomes without confusing correlation for causation?
How do we preserve cultural and personal pluralism instead of encoding one ideal life?
How should the system recognize moments that require qualified human support?
What measures capture durable benefit without turning fulfillment into a single score?
How can advanced personal intelligence be made accessible without subsidizing surveillance?
Architecture disclosure
Enough to make the edge legible. Not enough to publish the blueprint.
The public model is intentionally architectural, not procedural. It shows what our cognitive architecture is designed to understand and protect while keeping the methods that create differentiated performance proprietary.
The thesis and system requirements
We publish the purpose of the architecture, the kinds of human context it must account for, the permission and agency requirements, the research traditions that inform it, and the questions the work must answer.
The machinery that makes it work
We do not publish the internal ontology, memory structures, inference and confidence logic, cross-domain weighting, orchestration methods, adaptation policies, evaluation systems, or the implementation details that connect them.
The person can still challenge the model
Protecting our architecture does not require making it a black box to the person using it. People should be able to see, correct, scope, and remove what the system believes about them without receiving our technical blueprint.
Claim boundaries
What we are—and are not—saying.
We are saying: AI meant to support a human life needs a richer, more personal, more longitudinal representation than a prompt history.
We are not saying: subjective experience is fully computable, consciousness has been solved, software can read minds, or a model can define a person better than they can define themselves.
We are building: a proprietary cognitive architecture for permissioned context, Life Design reasoning, adaptive software, user correction, and outcome-informed support.
We will measure: whether that architecture helps people make clearer decisions, follow through more consistently, and report greater agency and life satisfaction—without promising medical, psychological, or social outcomes.
Contribute to the work
We welcome rigorous disagreement and serious collaboration.
We are interested in research, safety, product, clinical-boundary, privacy, measurement, and public-benefit perspectives.
Start a research conversation