FlyWire Neuro is where I’m exploring how a model of a fruit fly’s neural wiring can drive a body in a small 3D world. The local simulation uses a 138,639-neuron leaky integrate-and-fire model based on FlyWire connectivity. In the browser, I can change stimuli and watch the fly alongside a brain activity map.
The same body stays in the scene as it moves between walking, grooming, feeding and flight. Descending-neuron activity selects the motor behaviour, with measured walking and flight clips and cached poses providing the movement. Runs are recorded so I can go back through the neural, stimulus and body events afterwards.
One of the more useful additions was separating wall time, simulated brain time and body playback time. The full neuron model runs much more slowly than the animated body on my machine. I can either watch smooth motion using the latest neural state or make the body wait for the brain to advance, then inspect the timing in the recorded run.
I’ve shared the code, setup and research notes on GitHub, where it had picked up six stars by October 2026. There is still plenty to work out, especially the sensory loop and the mapping from neurons to behaviour. I treat it as a research experiment, with several parts of the model still based on approximations.
