1. What Nozzle is

Nozzle is one API key in front of every way a model can be served.

You send an OpenAI-shaped request naming a model. Nozzle resolves that name to a binding — a row that says who serves it and how — and relays your request there. The caller never learns whether the answer came from OpenAI on Nozzle's credential, from a Qwen pod running on a GPU Nozzle rents, from a dedicated GPU that belongs only to you, or from your own Anthropic account.

That indistinguishability is the product. It means you can move a model between those four modes — chasing price, latency, or data residency — without touching a line of caller code.

The commercial shape: you pay a subscription for platform access. GPU time is passed through at cost, sourced from the cheapest provider Nozzle can find. Nozzle does not mark up compute; the subscription is the product.