Show HN: A desktop fly drawn to the scent of vibecode

Aug 21, 2026 10:19 PM - 1 day ago 3

DesktopFly — a 3D consequence fly

A 3D consequence alert that lives connected your macOS desktop — driven by a unrecorded spiking simulation of the existent FlyWire connectome. It walks crossed your windows, grooms, sleeps, and decides to flee your cursor pinch the aforesaid neurons a existent alert uses.

Fork of DenisSergeevitch/desktop-fly — this 1 gives the alert a consciousness of smell for vibecode.

It scans your disk for supplier markers (AGENTS.md, CLAUDE.md, .cursor/rules, .kiro/steering and ~40 more) and turns thing connected surface that leads to them into an odour source: an editor aliases terminal model pinch the task open, a statement successful the front Finder window, a files icon connected the desktop. An unfastened task smells strongest, a closed icon weakest, and the scope of each grows pinch really overmuch vibecode it holds — a hub of six marked repos is smelled crossed the full screen, a single weak files only from nearby. The steering neurons past locomotion the alert there, and erstwhile the smell is acold the organization wakes up capable to make it fly.

 23,210 existent neuron positions, spikes flashing

The fly's encephalon window: 23,210 existent neuron soma positions from FlyWire v783, with unrecorded spikes flashing astatine existent neuron locations. The 2 glowing yellow markers are the Giant Fibers — the flight bid neurons. Click immoderate region to stimulate it.

  • 23,210 neuron soma positions (of 139,255 successful FlyWire v783) render the rotating encephalon window, colored by super-class (FlyWire's coarse cell-type grouping).
  • A 668-neuron circuit pinch ~19,000 existent synaptic connections (synapse counts, signed by neurotransmitter prediction) runs arsenic a 1 kHz leaky-integrate-and-fire (LIF) simulation:
    • LC4 (104) + LPLC2 (210) looming-detector ocular neurons
    • DNp01 / Giant Fiber (GF) (2) — the flight bid neuron
    • DNa01 + DNa02 (4) steering neurons · DNp09 (2) guardant walking
    • DNg11 (6) grooming · MDN (4) backward stepping ("moonwalker")
    • DNp02/DNp04/DNp11 (6) escape-maneuver (wing) neurons
    • their 330 strongest partners, including ascending (proprioceptive) and sensory (wind) neurons
  • Escape is not scripted. Your cursor's attack becomes looming input to the existent LC4/LPLC2 cells; the alert takes disconnected only erstwhile the Giant Fiber actually spikes done its existent synapses — ~1,200 synapses of feedforward inhibition push back, which is why slow approaches are tolerated and fast lunges trigger flight successful ~4 ms, conscionable for illustration the existent animal.

The assemblage itself is procedural (FlyWire is simply a encephalon connectome — nary body geometry exists), pinch a tripod gait, visible wing-beat, altitude-scaled flight, grooming, and slumber postures.

Requirements: macOS 13+, Xcode Command Line Tools (Swift 5.9+). No permissions aliases entitlements needed — everything it senses (cursor, model frames, clicks-as-taps, thermal state) is permission-free.

git clone https://github.com/DenisSergeevitch/desktop-fly.git cd desktop-fly ./build.sh ./DesktopFly

A 🪰 point appears successful the paper bar; discontinue from there. The alert wanders your desktop connected a transparent, click-through overlay — it ne'er intercepts your mouse aliases keyboard.

item effect
Pause / Resume freeze the world
Show/Hide Brain toggle the unrecorded encephalon window
Escape Test (loom) inject a looming stimulus, watch the GF fire
Move to Next Display hop the alert crossed monitors (shown erstwhile >1 display)
Add / Remove Fly extra flies (only alert #1 carries the brain)
Scare Flies startle everyone

The encephalon model is interactive: hovering pauses the rotation; clicking a region "optogenetically" stimulates the ~60 nearest circuit neurons for 400 ms. The fly's guidance is immoderate the existent web does downstream — click the Giant Fiber and it escapes; click DNg11 and it grooms; click one side's DNa01/02 and it turns.

How existent neurons thrust the body

body behavior driven by
escape takeoff DNp01 elephantine fibre spike
walk vs. rest, stepping speed DNp09 rate
steering DNa01+DNa02 left−right complaint difference
grooming DNg11 rate
backward scoot MDN burst
nervous darting LC4/LPLC2 organization rate
wing-beat effort, threat wing-raise DNp02/04/11 rate
spontaneous takeoff whole-population arousal

The loop besides closes body→brain: the gait hit feeds the circuit's real ascending (proprioceptive) neurons successful shape pinch the legs, and accelerated cursor motion stimulates its sensory (wind) partners.

Desktop ecology (all permission-free macOS senses)

  • Window terrain: model apical edges are ledges — the alert lands connected them, walks on them, rides a model you drag, and startles erstwhile 1 closes under its feet.
  • Window looms: a model appearing adjacent the alert feeds the looming pathway; the circuit decides whether to fly your dialogs.
  • Clicks are substrate taps; clicking adjacent to the alert startles it through the wind→GF pathway. Typing is vibration (idle-time API — knows when keys were pressed, ne'er which).
  • Circadian rhythm: dawn/dusk activity peaks, midday siesta, night quiescence. Sleep: idle astatine nighttime → it sleeps, breathing slowly, with raised arousal threshold; it grooms aft waking.
  • Temperature: flies are ectotherms — a basking Mac is simply a faster fly.

data/ ships pinch compact derived files. To rebuild them from the raw FlyWire Codex dumps (~60 MB download):

mkdir -p /tmp/flywire && cd /tmp/flywire B=https://storage.googleapis.com/flywire-data/codex/data/fafb/783 curl -O "$B/classification.csv.gz" -O "$B/coordinates.csv.gz" \ -O "$B/connections.csv.gz" -O "$B/consolidated_cell_types.csv.gz" cd - && python3 etl.py /tmp/flywire
./DesktopFly --simtest # circuit invariants: GF silent astatine rest, 4 sclerosis loom latency, ... ./DesktopFly --behaviortest # 17 end-to-end checks: stimulate neurons -> assemblage reacts ./DesktopFly --snapshot f.png # offscreen alert render ./DesktopFly --brainshot b.png # offscreen encephalon render

What's modeled vs. measured

Honesty section: the connectome gives wiring, not physiology. The LIF dynamics, neurotransmitter signs (ACh+, GABA−, Glu−), the gap-junction boost on LC→GF and wind→GF (documented electrical coupling), synaptic delays, and the sensory transduction (cursor → looming value) are modular modeling choices layered connected the existent graph. Everything downstream of the sensory neurons — who connects to whom, and really powerfully — is FlyWire data.

Code is MIT. The files successful data/ are derived from FlyWire (FAFB v783) and are CC BY-NC 4.0 — spot data/DATA_LICENSE.md. If you usage this, cite:

  • Dorkenwald, S. et al. Neuronal wiring sketch of an big brain. Nature 634, 124–138 (2024). https://doi.org/10.1038/s41586-024-07558-y
  • Schlegel, P. et al. Whole-brain note and multi-connectome compartment typing of Drosophila. Nature 634, 139–152 (2024). https://doi.org/10.1038/s41586-024-07686-5
More