Research prototype · four-layer architecture

NeuroChemical AI

A browser-based sandbox for exploring fast electrical signalling, typed chemical messages, local synaptic state, and global neuromodulation. It is an engineering analogy—not a claim that the system is conscious.

● simulation ready

01 · FAST PATH

Electrical layer

Rapid spikes transmit immediate events through a recurrent neural network.

Spike rate64%

gain

02 · TYPED MESSAGE

Chemical channel

Signals carry identity, intensity, persistence, and receptor-dependent effects.

Transmitterglutamate

glutamateGABAdopamineserotoninacetylcholine

03 · LOCAL MEMORY

Synaptic state

Each connection tracks short-term dynamics and eligibility for later learning.

Eligibility trace0.31

04 · GLOBAL STATE

Neuromodulation

A slow controller adjusts learning, exploration, gain, and resource allocation.

Modebalanced

balancedexplorationfocusrecovery

Signal flow

Fast events are interpreted by typed chemical channels, stored in local synaptic traces, and regulated by a slower global state.

SPIKE INPUTevent + timing

→

CHEMICAL ROUTERidentity + receptor

→

STATEFUL SYNAPSEhistory + plasticity

→

MODULATORgain + learning rate

→

OUTPUTadapted response

Live system state

Illustrative variables for an online-learning experiment.

Fast activity 0.64

Chemical persistence 0.48

Synaptic eligibility 0.31

Resource budget 0.82

auto-step

Event log

Use this log to connect mechanisms to measurable outcomes.

[ready] Four layers initialized.

[info] No claim of consciousness is made.

Suggested evaluation: compare this dynamic architecture with a static-weight control on continual learning, adaptation speed, catastrophic forgetting, robustness, and computational cost.