MYCEL

What is Mycel?

Mycel is a live particle simulation that turns typed words into bioluminescent fungal networks. It is built on a simplified version of the Physarum transport algorithm: each agent senses a chemical trail ahead of it and steers toward higher concentrations, then deposits its own trail as it moves. This simple local rule produces global network topology — the agents collectively span all food sources with near-minimal paths, without any top-down instruction.

The word you type determines where food sources appear — each letter’s position in the alphabet modulates its distance from centre, so every word produces a unique source constellation. The network that grows to connect them is different every time, yet deterministic: clear the canvas, type the same word, and the same topology emerges.

The Biology Behind It

In 2010, a team of Japanese and British researchers fed oat flakes to Physarum polycephalum at positions matching Tokyo’s major satellite cities. Within 26 hours, the mould had grown a network that closely mirrored the actual Tokaidō Shinkansen line — efficient, resilient, and fault-tolerant. The organism solves optimisation problems with no brain, no central coordination, only local chemical signalling.

Mycel captures the essence of that behaviour in a browser. The glow colours shift from deep forest-green at low trail density to bright lime-amber at high density — the same qualitative transition you see in bioluminescent organisms under UV light.

How to Use It

1. Open slayerblade.site/q/mycel/ — a demo word loads automatically.
2. Type any word of up to 10 letters and press Enter (or Return). The canvas clears and the network begins growing.
3. Watch the amber food-source markers pulse at each letter’s position. Within a few seconds, green trails connect them.
4. Type a different word to grow a new network. The topology is unique to each word.
5. The URL fragment updates to your word — copy and share the link to show anyone the same network.

Technical Notes

The simulation runs at 60 fps entirely in the browser. 1,500 agents are stored as three parallel Float32Array buffers (x, y, angle) for cache-friendly iteration. The trail map runs at 36% of canvas resolution and is rendered to an offscreen canvas, then scaled up with browser bilinear interpolation — giving smooth glows without per-pixel work at full resolution. Box diffusion (3×3 kernel, decay factor 0.962 per frame) keeps trails coherent without expensive Gaussian blur.

No frameworks, no build step, no backend. The URL fragment encodes the word — share the link to share the exact network. The page is a single static HTML file.