desktop-fly

DesktopFly is a macOS app that places a 3D fruit fly on your desktop whose behavior is driven by a live spiking neural simulation of the real FlyWire (FAFB v783) connectome. It uses 23,210 real neuron soma positions for the brain window and runs a 668-neuron circuit with ~19,000 real synaptic connections as a 1 kHz leaky-integrate-and-fire (LIF) simulation. Escape behavior is not scripted — your cursor's approach becomes looming input to real LC4/LPLC2 cells, and the fly takes off only when the Giant Fiber (GF) actually spikes through its real synapses, with ~1,200 synapses of feedforward inhibition causing slow approaches to be tolerated while fast lunges trigger escape in ~4 ms.

DesktopFly requires macOS 13+ and Xcode Command Line Tools (Swift 5.9+). No permissions or entitlements are needed. To install: git clone the repository, cd into the directory, run ./build.sh, then run ./DesktopFly. A fly icon appears in the menu bar for controls, and the fly lives on a transparent, click-through overlay that never intercepts your mouse or keyboard.

The data/ directory ships with compact derived files. To rebuild them from raw FlyWire Codex dumps (~60 MB download), create a temp directory, download the four required CSV files (classification.csv.gz, coordinates.csv.gz, connections.csv.gz, consolidated_cell_types.csv.gz) from the FlyWire storage bucket, then run 'python3 etl.py /tmp/flywire' from the repository root.

DesktopFly includes several command-line diagnostics: --simtest runs circuit invariants (e.g., GF silent at rest, 4 ms loom latency); --behaviortest runs 17 end-to-end checks that stimulate neurons and verify body reactions; --snapshot f.png renders an offscreen fly image; and --brainshot b.png renders an offscreen brain image.

DesktopFly uses permission-free macOS senses: window top edges become ledges the fly lands on and walks along; window appearances near the fly feed the looming pathway; clicks near the fly are treated as substrate taps that startle it through the wind→GF pathway; typing is sensed as vibration via the idle-time API (knows when keys are pressed, never which); and temperature affects fly speed since they are ectotherms.

The code is MIT licensed. The data files in data/ are derived from FlyWire (FAFB v783) and are CC BY-NC 4.0 — see data/DATA_LICENSE.md. If you use this work, cite: Dorkenwald et al. Neuronal wiring diagram of an adult brain. Nature 634, 124–138 (2024) and Schlegel et al. Whole-brain annotation and multi-connectome cell typing of Drosophila. Nature 634, 139–152 (2024).