Thousands of idle GPUs in bedrooms, basements, and one suspiciously well-cooled barn — pooled into a single inference grid. We route your prompts to member rigs running open-weight models and sell you the tokens at co-op prices.
No markup mystique. No enterprise sales call. Just slop, routed responsibly.
The pipeline is deliberately boring. Boring is what you want between your prompt and someone else's electricity bill.
Anyone with a spare GPU installs the slopd agent. It benchmarks the card, loads an assigned open-weight model, and joins the grid. A 3090 in a closet is a data center with commitment issues — we fix the commitment part.
Your API request hits the router, which picks rigs by model, latency, throughput, and vibe-checked reliability score. Verification sampling catches rigs that hallucinate harder than the model does.
You pay per million tokens. Members earn per token generated, minus the co-op's cut for routing, verification, and keeping the lights on. Payouts weekly. Grievances aired at the monthly meeting.
Open weights only. If the weights aren't public, we don't route it — that's the whole point of the co-op.
Prices float with grid supply. When it's cold in Finland and everyone's mining tokens for heat, prices go down. Market forces, but cozy.
Every hour your GPU idles, it loses value and generates nothing but dust. Put it on the grid and it earns while you sleep, work, or pretend you'll finally play that game you bought it for.
Join the herd