Several hundred thousand particles drift across a wrap-around canvas. Each one carries a position and a 2D velocity, and on every step it deposits that velocity into a field painted across the whole canvas — so the canvas itself becomes a flowing current that every particle both writes into and reads back from.
Each particle steers with a tiny neural network — its brain. It feels the current ahead-left and ahead-right through two sensors, runs those readings through a small sine-activated net, and the outputs become a forward force and a sideways nudge in its own frame. Nothing about flocking or trails is written by hand; the streams, filaments, and shifting textures all emerge from this feedback loop. Several species can share the canvas, each with its own brain and body.
species — each species keeps its tint. flow — hue follows the direction of the local current. brain — hue is pulled from the brain's own decision, so behaviour reads directly as colour.
Breed brains: fill a grid with random variations, find one you like, mutate around it to explore nearby relatives, then promote the winner to the full canvas. Lock a body attribute to hold it fixed while the rest re-rolls.
murk grows out of Fluoddity by aphid91 and Sage Jenson's physarum transport models, which pioneered the velocity-field-as-substrate and sensor-driven agents. murk is its own take — a sine-MLP brain instead of a Fourier sum, multiple species, and enforced chiral symmetry so no brain has a built-in left/right bias.
Top-left runs the sim; the pills on the right edit each species; the gear (bottom-left) holds visual settings; open grid view (top) breeds brains. There's no goal — drift, tune, breed, and watch what the field decides to become.