Why prepaid per-token pricing fits security work

Security work comes in bursts. An engagement runs hot for a week, a bug bounty target holds your attention for a weekend, a malware case lands and then clears. A monthly subscription charges the quiet weeks the same as the busy ones. NotHoly does not sell a subscription. You add credit and pay per token for what you run.

#Prepaid, per token, no subscription

You put an amount in your balance, by card or in crypto, and it becomes credit in USD when the payment clears. Each request is metered per token at the rate published for the model it used, and the charge comes off the balance. Money is kept as integers in billionths of a dollar, and every change to a balance is a new ledger entry that only adds a row, so your spend history is exact and nothing is edited after the fact. On a quiet month the bill is a quiet month, and a balance you topped up for a busy week carries over when the week ends. There is no plan, no seat and no monthly minimum. A solo hunter and a team of ten start the same way: a top-up on pricing, then usage. The minimum first top-up is shown on the billing page.

#The credit does not expire, and you cap the spend

Credit you buy does not expire, so a balance you set aside for an engagement is still there next quarter. When you want the balance to refill on its own, auto-refill charges a saved card once when the balance drops below a threshold you set, at most a few times a day, with a monthly cap you control.

Each API key carries its own limits: a total and a daily budget in USD, a list of models it may use, an expiry, and an allowed set of IP addresses. When a key's remaining budget cannot cover a request, the output limit shrinks to fit, and below a small floor the request is refused with key_budget_exceeded. A runaway agent loop hits the budget, not your whole balance. For a team, a workspace can hold its own spending cap, and each person creates their own key, so a manager watches usage per workspace instead of handing out one shared secret. Revoking a key takes effect on the next request. When a promotion makes a model free for your account, it carries a free tag with the list price struck through, and that price returns the moment the daily allowance runs out.

#What you are billed for, and what is free

Tokens are the meter. Input, output and cached input each have a published rate per model in the catalog, and every usage record keeps the exact price and cost version that was applied, so a receipt reconciles to the run. Prices change only through a new price version, never quietly inside a component.

Not everything costs tokens. The research routes, the CVE record, the advisory query, the knowledge base search and the page reader, are rate-limited per organization and are not billed per token. A CVE lookup or a knowledge base search does not burn model tokens.

notholy credits
notholy models

The same balance and the same rates show in the console under Activity and Billing, and the CLI reads them from the API.

#Where it does not fit

Prepaid means you pay before you run, so there is no invoice at month end for a team that needs net terms, and no free tier beyond the occasional promotion. Refunds are handled case by case and never exceed your unused balance or the amount paid, and a card dispute pulls the disputed amount back and suspends the organization's keys until it is resolved. An account needs a working e-mail, so the balance is not anonymous. Per-model rates sit in the catalog and change through price versions, so quote from the pricing page or GET /v1/models rather than a number you memorized.

Pay for the week you are busy, not the three you are not. A small first top-up on pricing is enough to run real work and watch the meter move.