This tab donates your browser's WebGPU to Unity's brain. Each donor holds a full brain replica (data-parallel) — the server merges the Hebbian weight-deltas every donor computes and re-broadcasts them, so more donors mean more aggregate compute and redundancy. The brain auto-scales its neuron count from the coordinator's free RAM — more donors add compute and redundancy, not neurons. The work is real neural matrix math — Rulkov-map spiking + sparse cross-projections, 100% equational, no text-AI / no language model. Leave the tab open to keep donating; one donor dropping off never downgrades the brain.
Prefer a compiled app over the browser tab? The native donor runs with a GUI or headless, drives every GPU in your machine as one unit, and connects to this brain out of the box.
⚠ Work in progress — functional but may have rough edges; please report issues. Built for NVIDIA GPUs (CUDA — no 2 GB binding cap, full VRAM, best performance); automatically falls back to WebGPU on unsupported GPUs (AMD / Intel / Apple) — same brain, same results. all releases →
Donating overnight? This page now holds a screen Wake Lock and auto-recovers a lost
GPU device — but for an unattended run also: (1) disable system sleep/hibernate in your OS
power settings (a wake lock only blocks screen sleep), (2) keep this tab open and foreground,
and (3) in chrome://discards set this tab to no auto-discard so Chrome doesn't
unload it under memory pressure. A donor that stays connected uploads once and keeps training; a donor
that drops forces a re-upload and stalls the brain.