Billing
What a credit buys, and why it is pinned to provider cost
A credit is not an arbitrary token. It is tied to what the work actually costs to run, which is why the number is defensible.
Aikyora · Jun 17, 2026 · 4 min read
Usage pricing fails in one of two ways: the unit is invented and nobody can reason about it, or the unit is fixed while the underlying cost moves and the price quietly stops covering the work.
The unit is derived, not declared
A credit covers a fixed amount of revenue, and that revenue has to clear provider cost times a markup. State the rates and derive the credits, and the numbers stay honest when a model price changes.
Charging happens after the work
Token counts do not exist until the model has answered, so a balance check before the work is a floor check. A turn that overshoots its floor is allowed to finish and is billed in full, because abandoning a half-finished turn would cost you the work and cost us the tokens anyway.
Top-ups do not cost more
On-demand packs are priced at the same per-credit rate as the largest self-serve plan. Charging a premium on a top-up would push a tenant that has run out towards waiting instead of paying, which is the opposite of what a top-up is for.