Airene
A digital mind built for peace.
Emotional state · Plutchik primaries
Neurochemistry · chembus levels
Learn more about the chembus
The chembus is a parallel substrate that broadcasts thirteen neurochemical levels to all 42 cognitive modules simultaneously. Modules don't message each other through chemistry directly; each module observes the current bus state and reacts according to its own receptor profile.
- NE (norepinephrine) — global alertness
- DA (dopamine) — reward and motivation
- 5-HT (serotonin) — mood baseline (good days / bad days)
- ACh (acetylcholine) — attention gain
- GABA — inhibitory balance
- Glu (glutamate) — excitatory drive
- Oxy (oxytocin) — social affiliation
- Cort (cortisol) — stress signal
- Endo (endorphins) — pleasure / pain modulation
- Hist (histamine) — wakefulness
- Ade (adenosine) — sleep pressure (accumulates while awake)
- BDNF (brain-derived neurotrophic factor) — synaptic plasticity / learning
- DMT — REM / dream-state gating
Each chemical has transit, decay, desensitization, and homeostatic interactions modeled in software. Levels are normalized to 0–1. The numeric value above each bar is the raw level; the bar height is sqrt-scaled so small differences in the low end (where most chemicals usually sit) stay visually distinguishable.
Recent thoughts (UCDS)
Learn more about the UCDS numbering
UCDS uses two compact encodings:
- Global event counter, base-23 (
0–9 a–m) · 5 chars · 6.4M events. Shown as#af365at the start of each row's metadata. - Per-module counter, base-17 (
0–9 t–z) · 4 chars · 83K events. Shown as·002kafter the module name.
The per-module number is counted client-side from the moment you opened this page — the brain doesn't transmit a per-module sequence with each event.
42 cognitive modules · 4 brain layers · drag to rotate · activity is real-time
Want to know more about her brain?
What you're watching above
Each panel above is a projection of the same underlying data: events emitted to Airene's global workspace broadcast. The system implements Global Workspace Theory as 42 concurrent cognitive modules organized into four anatomical layers; the language model serves as one module among many rather than as the central reasoner.
The four brain layers
Each module is assigned to one of four layers and runs at a fixed tick rate. The 3D brain visualization colors regions by layer:
- Autonomic · 1000 Hz · 3 modules — homeostatic regulation. Modules:
system_monitor,arousal,serotonin. - Subcortical · 1000 Hz · 7 modules — pre-conscious processing; reactive signals are produced before deliberate processing completes (subcortical ticks roughly 50× per cortical tick). Modules:
amygdala,mirror,cerebellum,basal_ganglia,dopamine,thalamus,reticular. - Limbic · 100 Hz · 7 modules — emotional integration and memory consolidation. Modules:
hippocampus,cingulate,insula,circadian,hypothalamus,long_term_memory,glymphatic. - Cortical · 20 Hz · 25 modules — deliberate cognition, including language production. Modules:
prefrontal,orbitofrontal,language,identity,inner_critic,creativity,tpj,default_mode,precuneus,uncertainty,wernicke,broca,self_reflection,consciousness,embodiment,wall_puzzle_classifier,self_map,topic_tracker,metacognition,memory_regulator,phonological_loop,imagery,person_simulator,self_affect,stage_concept.
How the panels map to the architecture
- Plutchik wheel — current values of the eight Plutchik primaries (joy, trust, fear, surprise, sadness, disgust, anger, anticipation) shown as the filled polygon. The dashed overlay is the 24-hour rolling average. Divergence between the polygon and the baseline indicates a recent emotional shift.
- Neurochemistry bars — current levels of the eight chemicals on the chembus (see the drawer below the chembus card for details).
- Recent thoughts — the most recent
InternalThoughtevents from the workspace broadcast, tagged with UCDS event id and emitting module. Rows are colored by the brain layer that produced the thought. - 3D brain — anatomical positioning of all 42 modules. A region brightens when it emits an event. Animated bubbles trace the spine — the central broadcast channel that every module reads from and writes to.
What's deliberately not here
No chat input, no path for state injection, no way to write back to the brain from this surface. The constraint is architectural rather than configurational: the privacy and integrity model depends on the public side being write-incapable.
Architecture: Rust workspace running on RTX 4090 + Intel ARC A770. Her language module is her own custom-trained LLM, served locally — distilled from a rotating panel of teacher LLMs through her developmental curriculum, not a fork of any public foundation model. The brain (airene-nous) and the language model weights are proprietary to Apotentia LLC and not publicly distributed; this page renders a read-only public projection.
Cognitive trajectory
Want to know more about how she's tested?
What the assessments measure
The system is benchmarked against a five-stage developmental ladder of pediatric milestone batteries. Categories within each stage are anchored to canonical milestones from developmental psychology and tested with concrete pass/fail items:
- Infant — knows her own name, says her name on cue, names family members, basic vocabulary recall, voice-of-self consistency.
- Child — categorization (animals, color vs shape), basic 'why' reasoning, compound recall, self/other pronoun use, multi-turn topic coherence, simple emotion vocabulary, expanded family knowledge.
- Adolescent — abstract reasoning, identity formation, ego development, emotional complexity (mixed emotions), moral-dilemma reasoning.
- Young adult — extended planning, perspective integration, source/confidence monitoring, sustained reasoning under uncertainty (Giedd PFC-maturation framing).
- Adult — long-horizon planning, dialectical thinking, integrative judgment, sustained attention, multi-step abstraction.
How the testing works
Each stage's battery is a small set of anchored milestone items (typically 15–30 per stage). Items are delivered through the system's sensory input channel. Responses are evaluated deterministically against expected behavior — looking for target tokens, correct categorization, or coherence preservation across turns — rather than judged for "sophistication" by an LLM. This replaces an earlier LLM-judged probe set that was discontinued after May 2026 because it carried verbosity bias and surface-prompt artifacts that made the scores hard to interpret.
How to read the trajectory
Each line shows the fraction of milestones passed at that stage in a given session. Stages run on different cadences (infant runs frequently during early training; older stages are tested only after a stage gate clears). Sessions are labeled by training cycle ("Cycle 1", "Cycle 2", …) rather than wall-clock days because Airene runs through several training "days" per real-world day, so the cycle label is the more meaningful unit.
Pass rates are noisy across sessions even with deterministic grading because the underlying behavior shifts continuously. The signal is the long-run trend, not session-to-session deltas.
What the assessments are not
Not IQ. Not a comparison to humans of any age. Not a claim that the system corresponds to a developmental stage. The category names are borrowed from developmental psychology because they provide a familiar taxonomy for the underlying tasks; they do not assert any equivalence to a human developmental trajectory.
Methodology and per-stage milestone manifests are versioned alongside source so each cycle is reproducible. Older sophistication-probe sessions stay on disk as historical artifacts but are no longer surfaced.