Reading Your AI Bill: Credits, BYOK Fees, and What Actually Drives Cost
A plain look at how Piskes prices AI usage — credits, the reserve-then-settle model behind them, and why the BYOK fee is a starting number, not a fixed one.
If you're used to flat, per-seat software pricing, AI features can feel like a different kind of bill — because they are. Every model call is a real, variable cost, and how Piskes prices that is worth explaining plainly rather than burying in a flat tier.
How credits work
AI usage on Piskes runs on a credit system, not a flat allowance. When a request comes in, credits are reserved up front and then settled against the actual cost once the response completes — a reserve-then-settle model built specifically to stay accurate through streaming responses, where the final cost isn't known until the last token arrives. Every credit grant your workspace receives is tracked in its own ledger, so what you've been given and what you've used are both auditable, not just summarized into a single balance.
Bring your own key (BYOK)
If you connect your own API key for a provider instead of using Piskes' shared credits, you're charged a platform fee on top of what that request would have cost at the provider directly — currently set at 5%, rather than a flat markup on every call. It's important to be upfront about where that number comes from: it's explicitly logged internally as a starting rate, set to match the platform's existing internal fee structure, pending real usage data — not a number the platform has committed to holding fixed. Treat it as the current rate, not a permanent one.
Why this approach
Unlike a database query, a model call has a real, variable cost behind it every time. A credit system that reserves and settles per request — rather than papering over that with a flat "unlimited AI" claim — is our attempt to make that cost visible and honest instead of hidden in a tier you can't see into.