⌁
CONNECTOME HOUSELive neural circuit · Anatomical body
— brain neurons— measured edges
LOADING THE RESEARCH ASSETS

A brain you can inspect.
A body you can observe.

Checking WebGPU…

First load prepares ~300 MB of connectome and body data. No synthetic-neuron fallback.
Preparing simulationNeural time 0.000 sBody time 0.000 s—
01 / CONNECTOME

FlyWire brain

Drag to orbit · Scroll to zoom
Firing in the latest 10 ms0 active neurons
02 / EMBODIMENT

Anatomical fly

Rendered retinal input · simplified vision
MANC muscle pools → joint actuators0.00 cm/s
STIMULATE THE CIRCUIT

Inputs & command neurons

Selected neuronClick a point in the brainPositions come from FlyWire soma / centroid annotations.

Measured motor-neuron activity drives joint targets. No scripted gait or root velocity. Muscle gains and adhesion are modeled; coordinated walking is not guaranteed.

WHAT IS FIRING

Live activity

0 spikes
Each column is a new simulation window. Brightness shows measured model spike rate.

Run a stimulus to inspect active neurons.

INTERVENE & COMPARE

Synapses & plasticity

Modify existing measured KC → MBON connections using an experimental reward-modulated rule.

Modified edges 0Disconnected neurons 0

This changes synaptic gains in the running LIF model. It does not establish learned human behavior or reproduce dopamine biochemistry.

WHAT IS MEASURED. WHAT IS MODELED.

The research and the bridge

Measured anatomy: 139,255 FlyWire neurons and 15,091,983 weighted neuron-pair connections; MANC nerve-cord wiring; the detailed Flybody meshes and joint model. Source checksums are recorded in the project.

Simulated physiology: a WebGPU leaky integrate-and-fire implementation derived from webgpu-fly, informed by Shiu et al. It uses simplified neuron parameters and neurotransmitter signs. It is not quantitatively validated against the full research model.

Body interface: FlyWire and MANC are from different specimens, joined here by matching descending-neuron cell-type names. Motor pools are mapped to body actuators with authored gains. The muscle mode has no gait oscillator or root-velocity assist; adhesion remains modeled. Assisted gait uses a scripted tripod pattern and velocity assistance, visibly labeled.

Learning experiment: teaching pulses adjust gains only on existing active KC → MBON connections using a bounded activity-dependent rule. This is an experimental rule, not a validated biological dopamine model. A changed brain is not proof of a learned skill.

Live means live: colors and rates use new GPU readbacks, with a bounded history. The brain and body advance by the same simulated time. Pausing halts both. The app does not replay a neural animation as current activity.

The house and learned daily human activities are not implemented in this corrected research foundation. No synthetic substitute runs in their place.