gnat
Linux/Hyprland desktop port of DesktopFly with a brain window and stimulation controls. Uses a 668-neuron circuit subset.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.Documented LIF constants - membrane decay, threshold, refractory period, weight scale, noise and inhibition floor.
https://github.com/lubabs770/gnat/blob/173a0be8158181bb675a80c9e5df366c9d214a6e/crates/gnat-sim/src/lif.rs#L14-L30Synapse weights are assigned only while constructing the CSR adjacency from the connectome file.
https://github.com/lubabs770/gnat/blob/173a0be8158181bb675a80c9e5df366c9d214a6e/crates/gnat-sim/src/lif.rs#L251-L258
A Rust/Wayland port of DesktopFly that carries the neuron model over intact: the same 668-neuron FlyWire circuit as a leaky integrate-and-fire sim with a 0.9512 per-ms decay (20 ms tau), unit threshold, 2 ms refractory, an inhibition floor and a delayed-inhibition window that the comments explain is what lets the giant fibre fire before feedforward inhibition arrives. It ships an invariant suite run against the real data (GF silent over 4 s at rest, firing 4 ms after an abrupt loom). L1. Not L2: the weight array is filled once when the CSR graph is built from circuit.json and never touched again, and a grep across every .rs file for plasticity, STDP, dopamine, training or a learning rate returns nothing. Nothing in this project learns; the behaviours the README lists all fall out of fixed circuit dynamics.
Record
Nearby
All projects →A macOS desktop fly combining FlyWire spiking circuits with a MaleCNS brain-to-leg extract, modeled senses, and articulated behavior. Includes an Electron port.
Python/GTK port of DesktopFly for GNOME on X11 and Wayland, retaining the upstream FlyWire circuit and documenting platform limitations.
Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.