The gallery
Each panel is a readout, not a render farm. The hex mosaic is the fly’s 892-column retina. The gauges are descending neurons a fly walks with. Lite mode is honest about the reduced mapping; full-brain mode drives the same UI from the real LIF connectome.
A luminous target on a dark field. The hex retina lights. DNa02 left/right asymmetry steers a fly-cursor after it.
open demo → 02 · spike avalanchePoke the visual field. Watch a cascade through the CNS volume — measured somata when the annotations are loaded.
open demo → 03 · learning theaterKC→MBON depression under dopamine. The −6% story, watchable. Novelty-as-reward is labelled as a modelling choice.
open demo → 04 · what the fly seesSplit view: page texture → 892-hex mosaic → descending-neuron gauges → cursor path. Every pane is a readout.
open demo → 05 · nose / ORNA scent plume across 53 receptor types. 2,635 ORNs flycoinrh left unused, now lighting the antennal lobe.
open demo →Motor readouts
Honesty
The connectome is CC-BY — HHMI Janelia FlyEM, the Cambridge Connectomics Group, and Google Research. Simulation approach after fruitflydev/flycoinrh, Shiu et al. 2024 and Lappalainen et al. 2024. This gallery does not own that brain.
Lite mode samples the stimulus through the same 892-hex L1/L2 math as FlyEye, then uses a
reduced spatial pooling onto descending neurons, decoded with FlyPilot’s equations.
When graph.npz is present, those gauges come from FlyBrain.run.
The mushroom-body reward signal is not real. The olfactory motor bias is demo wiring.
Private preview — not published. MIT for new code. Keep the CC-BY attribution wherever the connectome goes.