Every Interview Comes With a Receipt

“What does one of these AI interviews actually cost us?” is a fair question — and for most AI features it goes unanswered, because nobody keeps count.

Here, every interview run carries an itemized cost receipt: which model, how many tokens, for what. Across every provider involved, from the language model to the assessment.

Interview #2417 14 min · DE

Speech → text 4.210 T

Conversation model 18.930 T

Scoring 6.400 T

Total 0,41 €

Billing: per interview, per model, per step
Prices: public list prices, computed at display time
The browser reports: meter readings only, never prices

A Ledger for Tokens

Under the hood, this is a token ledger: every model call writes its usage figures into a journal attached to the interview. The client reports nothing but meter readings — which models sit behind them, and what they cost, only the server knows.

Pricing happens only at display time, using the providers’ stored list prices. If a provider changes its prices, the receipt still holds up: the usage data is stored raw, and the valuation is a calculation, not a snapshot. Models without a stored price are flagged visibly instead of being silently priced at zero.

That turns a vague “AI just costs money” into a solid number per business transaction — the basis for honest decisions about model choice and scaling.