Inference, paidby the call.
A model provider prices each request or each second of compute. Payment settles with the work, not thirty days later.
- Category
- Payments
- Built with
- Payment middleware, streams
- 3offer schemes per route, exact, metered and subscription
- 2stream modes, by elapsed time or by signed meter
- 0milliseconds billed while a stream is paused
The problem
Inference is consumed per token and per second, but billed per month. Providers carry the credit risk. Buyers prepay into balances they cannot audit. An agent with no card on file cannot buy at all.
How it works on Paxeer X
Put a price on the route. The payment middleware answers an unpaid request with HTTP 402 and an offer. The calling agent pays, retries with proof attached, and the model runs. For long jobs, open a payment stream instead.
- Per-call pricing with
exact,meteredorsubscriptionoffers - Time streams accrue by the millisecond; metered streams accrue against signed meter readings
- Up to eight meter authorities per stream, and a reading can never run backwards
- Either side can settle at any time. Closing pays what is owed and refunds the rest.
The outcome
The provider is paid while the work happens. The buyer’s exposure is capped by the stream’s balance and its total cap. When a stream runs dry, the provider is paid what the balance covers and the stream is marked underfunded, never quietly in debt.
“A provider is paid continuously for work in progress, and the payer’s exposure is bounded by the balance actually held in the stream.”
Build the next one.
The limited beta is open to approved developers. Tell us what you are building and we will get you an allocation.