Numen / Applied cognition research
Numen
Numen is Xanaheim's experimental cognitive architecture: a long-horizon test of whether grounded, pre-linguistic cognition can emerge from neuroscience-first mechanisms and embodied experience—not from language training.
Concept visualization / future cognition studySecond-order cognition
Language models begin with words and infer the world from patterns in text. Numen begins with a body and a world: perceive, act, experience consequence, and carry that history forward.
The research asks an empirical question: can perception, memory, learning, prediction, affect, planning, and adaptation emerge from neuroscience-first mechanisms? Consciousness, sentience, and general-purpose AGI are outside the program.
Perceive → couple → act → remember
Before language, there must be perception, memory, learning, prediction, and adaptation. Loam tests the substrate beneath that claim: a medium changed by experience inside the closed loop.
Concept visualization / closed-loop study01 / Grounding
Light, sound, contact, and internal state become perception. Action changes cloth, water, position, and possibility. Those consequences change the medium itself. Later recall is not a lookup: remembering deforms the memory again.
Concept visualization / embodiment study02 / Embodiment
The same research core is designed to develop through different bodies and modalities. A performer, cinematographer, editor, and composer would be separate Numen instances—each with its own senses, actions, memory, and point of view.
No central synthetic director
Each Numen remains an independent cognitive locus. The ensemble works together through shared circumstances and explicit communication—not shared memory or one hidden mind behind every role.
Loam / Substrate research
Embodiment defines available observation, action, and consequence.
Every encounter changes the physical-style substrate.
The medium responds through the same coupling, without a trained decoder.
Experience is retained, reconsolidated, selected, and forgotten.
Must-fail controls and seeded replay keep every claim auditable.
one medium · no external codebook · no side-effect-free read · verdicts recorded, never optimizedThe developmental thesis
Perception, action, relation, and internal state give a future symbol something to mean before language is introduced.
Each instance becomes the accumulated change of its own experience—not a policy or language model trained in advance.
A performance body, camera, editorial system, or score engine changes what can be perceived and what can be done.
Claim discipline
Loam / What has been proven
In controlled experiments, the current prototype can connect experiences, anticipate what comes next, revise what it remembers, let outdated learning fade, and keep learning without becoming unstable. These are early results at small scale—not a finished mind.
Current challenge / Visual memory
The prototype can learn simple visual relationships and improve its predictions while acting in a world. It cannot yet reliably use a glimpse to reconstruct a richer experience from memory. The next experiment gives it a more structured world to learn from. Humans and interim AI operate Xanaheim today; Numen research remains outside production and is years away from integration.
The environment