Fly Terrarium
Every move comes from 138,639 simulated neurons, wired exactly like a real fly’s. Drop sugar, puff air or cast a shadow — then watch which neurons answer.
The brain is overexcited: activity is going round in circles and may never die out, because the model has no neuron fatigue. While it lasts, time in the dish slows down.
No neuroscience needed. Each card explains one piece of what happens between your click and the fly’s reaction.
Scientists sliced a fly’s brain into thousands of layers, photographed each with an electron microscope and traced every neuron and every contact between them. That wiring map is the connectome: 138,639 neurons and 15 million connections for the female, 165,122 neurons for the male, whose map also covers the nerve cord.
Incoming signals pour charge in, and it slowly leaks out. When the level reaches −45 millivolts, the neuron fires a spike and empties back to −52. That is the whole neuron model — “leaky integrate-and-fire” from Shiu et al., Nature 2024 — and all of them follow it.
A spike reaches the neuron’s partners 1.8 ms later. How hard it pushes depends on how many synapses join the two cells; which way — excite or calm — on the chemical the sender uses. Nobody tuned a single connection: they come straight from the map.
The page works out what the fly senses — sugar under its mouth, air on its antennae, a shadow growing in one eye — and turns it into spikes in the matching sensory neurons. The brain runs 20 ms in 0.1 ms steps; then the page reads a handful of output neurons and moves the fly. And again.
Updating every neuron 10,000 times a second is too slow for a browser. So a neuron is only updated when a spike reaches it: an exact formula jumps it forward and predicts when — if ever — it will fire. A quiet brain costs nothing, and the result matches the Python version to a correlation of 0.9996.
Tick “Shuffle the wiring” and every neuron keeps its synapses, but they now lead to random cells. Same neurons, same numbers, a random map — and the fly no longer eats, grooms or escapes. The behaviour lives in the connections.
The wiring of every neuron, from two scanned flies: FlyWire’s female and Janelia’s male.
The neuron model and all its constants, from Shiu et al. (Nature, 2024).
Which output neurons answer a stimulus. The page never chooses that — the wiring does.
The sensory neurons each tool stimulates, from published lists and cell types.
The stroll. The model has no activity of its own, so while the brain is quiet the fly wanders by a simple rule.
How strong each stimulus is, and the eye: a formula for how fast a shadow grows, not a simulated retina.
Turning spike rates into speed, turns and jumps. For the female, “a giant fibre spike means jump” is our rule.
The dish, its walls and the body animation. The Python version of the project has a real physical body.