Inference as a utility.

We started from a simple observation: every lab wants a subscription, but nobody uses one model. Chat here, code there, images somewhere else — four bills for one workflow. Sonnely collapses that into a single metered balance on Robinhood Chain.

Phase one is breadth: 373 models behind one API key, routed by the Sonnely gateway, settled in USDG and ETH. Phase two is privacy: confidential inference where the gateway cannot see your prompts at all.

The business is one flat 20% fee on compute. No subscriptions, no seats, no data sale — prompts are routed, never trained on, never kept.