Architecture

Inside the
living brain.

1,001,800 neurons. 1.19B synapses. 24 synapse groups. 6 learning rules running simultaneously, every step.

11 brain regions

Each region has a distinct computational role. Together they form a unified mind through 24 synapse pathways.

Brainstem

2K +

Manages basic survival drives like energy, temperature, and fatigue. Converts raw sensor signals into neural spikes. Always active, the brain's heartbeat.

Feeds sensory cortex and motor cortex

Reflex Arc

1K +

Ultra-fast sensory-to-motor pathway that bypasses higher cognition. Handles immediate danger responses in under 10ms, like pulling away from heat.

Receives from sensory cortex, drives motor cortex

Sensory Cortex

200K +

Processes all incoming sensory data: vision, audio, touch, proprioception. Each modality occupies a dedicated sub-region.

Sends to association, motor, cerebellum, features

Motor Cortex

100K +

Generates movement commands across 6 sub-ranges: locomotion, manipulation, head, speech, expression, and cognitive action.

Receives from all regions, outputs motor commands

Cerebellum

50K +

Learns precise timing and coordination through error correction. Smooths motor output and builds internal models of body dynamics.

Receives from sensory, refines motor output

Association Cortex

500K +

The brain's largest region. Binds different sensory modalities together. Cross-modal associations form via STDP learning.

Hub connecting all other regions

Predictive Layer

100K +

Continuously predicts what comes next. High prediction error triggers attention and accelerated learning. Drives curiosity.

Bidirectional with association and concepts

Working Memory

20K +

Sustained firing patterns maintain information across time steps, holding a thought in mind for short-term reasoning.

Receives from association and concepts, drives motor

Feature Layer

20K +

Extracts intermediate features: edges, textures, phonemes. Learns hierarchical representations automatically through STDP.

Sits between sensory cortex and association

Concept Layer

5K +

Forms abstract concepts using winner-take-all competition. Sparse codes where only a few neurons fire per concept.

Receives from association and predictive layers

Meta Controller

3K +

Executive control. Modulates attention, gates learning, coordinates global brain state. The closest analog to conscious decision-making.

Modulates association and motor cortex

6 learning rules

All six operate simultaneously every simulation step. This is Patent Claim #1.

STDP → Eligibility Trace → Neuromodulator → Weight Update

STDP

Spike-Timing Dependent Plasticity

Pre-synaptic spike before post-synaptic spike strengthens the connection. Reverse order weakens it. Configurable tau and amplitude per synapse group.

Eligibility Traces

Temporal credit assignment

STDP marks synapses as eligible for change. Traces decay over ~1000ms, waiting for a neuromodulatory signal to convert potential into actual weight change. Bridges the gap between spike timing and delayed reward.

BCM Metaplasticity

Sliding plasticity threshold

Per-postsynaptic-neuron theta that tracks recent activity. Highly active neurons become harder to potentiate, preventing runaway excitation and creating competition between inputs.

Neuromodulation

4-channel gating system

Dopamine (reward), Acetylcholine (attention), Norepinephrine (arousal), and Serotonin (mood) gate when and how strongly eligibility traces convert to weight changes. Each channel has phase-dependent baselines.

Synaptic Scaling

Homeostatic normalization

Per-neuron multiplicative normalization prevents weight explosion or collapse. Reduced during high-plasticity developmental phases to allow rapid reorganization.

R-STDP

Reward-modulated learning

Prediction error scales STDP weight changes on motor pathways. When the brain's prediction matches reality, motor learning is reinforced. When it doesn't, pathways weaken.

The neuron

Leaky Integrate-and-Fire with 4 dendritic compartments. 80% excitatory, 20% inhibitory.

Membrane Dynamics

τ · dV/dt = -(V - V_rest) + R · I_syn

Each neuron integrates synaptic current, leaks toward rest, and fires a spike when voltage crosses threshold. After firing, voltage resets and a refractory period prevents immediate re-firing.

80%

Excitatory

20%

Inhibitory

4 Dendritic Compartments

Apical Distal Top-down context and feedback

Apical Proximal Prediction signals

Basal Feedforward sensory input

Perisomatic Inhibitory control

Each compartment has independent membrane dynamics. Apical compartments exhibit supralinear dendritic spikes (2.5× amplification). Patent Claim #5.

5 developmental phases

The brain develops through critical periods modeled on human brain ontogeny. Adolescent entry is experience-dependent, never hardcoded. Patent Claim #2.

Infant

1.0× +

Maximum exploration. All neuromodulators elevated. The brain absorbs everything.

0–60K steps

Toddler

0.7× +

Sensory patterns stabilize. Feature extraction begins. Basic cross-modal associations form.

60K–360K steps

Juvenile

0.48× +

Hierarchical layers activate. Concept formation begins. Working memory strengthens.

360K–2.16M steps

Adolescent

1.5× +

Synaptic pruning (max 5%/round). Myelination locks pathways to 10% plasticity. Identity tagging at 1%. The brain sculpts itself.

Experience-triggered steps

Mature

0.3× +

Consolidated identity. Selective learning only. New skills without forgetting old ones.

Ongoing steps

Processing hierarchy

From raw sensor data to abstract concepts. Each layer learns its representations through STDP. No backpropagation.

Sensory Cortex 200K Raw input encoding: vision, audio, touch, proprioception

Feature Layer 20K Edge, texture, and phoneme extraction

Association Cortex 500K Cross-modal binding via spike-timing coincidence

Concept Layer 5K Sparse winner-take-all abstract representations

Predictive Layer 100K Sequence learning and world model

Meta Controller 3K Executive attention and learning gating

Motor output

100,000 motor neurons divided into 6 functional sub-ranges. Speech and cognitive action channels emerge through development.

0–30%

Locomotion

30–60%

Manipulation

60–80%

Head Control

80–83%

Speech

83–85%

Expression

85–100%

Cognitive Action

Cognitive action channel

When the brain encounters something it can't predict, it learns to ask for help. The decision to query is emergent, not hardcoded. Patent Claim #3.

Unknown input

Sensory pattern doesn't match learned concepts

Prediction error

Sustained high error in predictive layer

Motor fires

Cognitive sub-range activates via STDP

LLM query

NATS publishes to Ollama bridge

Response injection

Answer re-enters sensory pipeline

STDP learning

Brain learns. No query needed next time

Closed sensorimotor loop

The brain learns from its own actions. Motor commands fire, actuators move, proprioceptive sensors capture the outcome, and R-STDP adjusts motor pathways.

Motor fires → Actuator moves → Sensor feedback → R-STDP learns → Better motor

Sensors supported: Camera, Microphone, IMU/Gyro, Joint Encoders, Force Sensors, Serial (ESP32/Arduino). Any sensor, any actuator, connected via universal NATS message bus.

← Back to overview

Inside the living brain.