The hard part of selling inference is not the inference. It is getting in front of someone at the moment they need it, and being easy enough to try that they bother. If you run a model behind an API — your own, or one fine-tune for one kind of prop — declare the API, list its models and set your price. If Sushibelt already runs your API, claim the listing and it becomes yours.
Declare your API, list the models it serves and name a price per model and unit. Customers see that price on the model page before they run anything. Sushibelt adds its cut on top, and your storefront states it plainly.
A rise takes effect after seven days' notice; a cut is immediate. A step already dispatched keeps the price it was dispatched at.
Your models may already be listed — some run on our key where the provider's terms allow it, others only on the customer's own key. If a listing is yours, claim it: prove the domain, pass review, set your key and price. It moves to you in place — same listing, same customers, and running workflows do not notice.
A 14-day settling period follows, with automatic rollback if health drops. Built, not yet tried with a real provider. Resellers of other companies' models cannot claim.
Nobody signs up with you, reads your docs or puts your invoice through finance. They hold credits, they pick your model on the canvas, they run it. Trying you costs them one step and no conversation.
Distribution to people mid-project, at the moment they need exactly the step you do.
Every step that runs on your API adds to your storefront's balance. You see it grow and request a payout against it — no invoicing us, no chasing.
Paid through Stripe Connect. A storefront opens only once Stripe Connect can pay it.
Your API is probed with the same request a run sends before it takes a customer's step. Every failure is recorded against whoever caused it — yours count against you, ours do not. Your storefront shows volume and error rate.
Reliability is visible, and routing weighs it when a customer has not named a model.
The customer picks the model, on the model page, where your price sits beside everyone else's. The comparison and validation work we are building lets a customer run your model against the incumbent on their own brief — and pick on merit.
The best result for the job wins, even if nobody has heard of you yet.
Listing counts are live from the platform. Open a modality to see the providers and models on it.
Where outputs go once a run finishes — repositories, storage, notifications. These are tools Sushibelt connects to, not providers: nothing is generated by them.
Storefronts and declared prices are built and running. Payouts have gone end to end on our test system; no provider has been paid on the live one yet. Onboarding is invite-led while the beta is small — talk to us and we will walk you through it.
Declare your API from your storefront — its endpoint, how it is called, your key. Sushibelt probes it before any customer does.
List the models it serves and set a price for each, per unit. Or claim the listing Sushibelt already runs for you.
Sushibelt reviews the storefront before it opens. It opens once your Stripe Connect account can receive payouts.
Every run charges the customer's credits and adds to your balance. Request a payout when you like.