Prepaid credits, budgets on every key.

Nothing is served before it is paid for. Credits are bought in advance and every request is charged from the balance at the model's per-token price, using the token counts the model server reports. There is no subscription, no monthly fee and no invoice to wait for.

An agent (or you) runs npx -y scaledthought@0.3 billing topup --amount-usd 20. It returns an approval_url and opens the funding page, where a person confirms the amount, can turn on auto-reload, accepts the Terms and the Credits & Refunds policy and pays by card on Stripe's checkout. Credits arrive when the payment is confirmed, and Stripe emails the receipt. Over the API it is POST /v1/billing/topup; over MCP, billing_topup.

Credits never expire. Unused purchased credits can be refunded within the refund window of the Credits & Refunds policy. Prices per model are on /pricing and in models --live.

The limits come from the billing schema the API validates against, so this page and the API can't disagree.

• $10 to $2,000 per top-up ($20 unless you pick another amount).
• At most $2,000 per account per rolling 24 hours, counting paid top-ups, open approval links and auto-reloads together (409 daily_topup_limit).
• At most 2 open approval links at a time (409 too_many_pending_topups).

Auto-reload is off until a person turns it on, either on the funding page of any top-up or, without buying anything, with npx -y scaledthought@0.3 billing auto-reload setup --below 5 --amount 20 (a page where they save a card).

• The suggested terms add $20 whenever the balance drops below $5. The threshold is at least $5.
• Reloads count toward the $2,000 per 24 hours. There is no monthly limit unless one is set (--auto-reload-monthly-max, or later PATCH /v1/billing/auto-reload).
• Turning it on or raising it needs the person's approval (403 approval_required). npx -y scaledthought@0.3 billing auto-reload off, or lowering it, needs none.
• If a reload is declined, auto-reload turns itself off and the next 402 says to get a new approval link.

Each request reserves its maximum cost before it runs: the prompt plus its max_tokens (requests without one get a default cap). A request the balance can't cover is refused up front with a 402; after it runs, only what the model server reports is charged.

Every key can have its own hard daily budget in USD and its own rate limits. When a key's budget is spent, its requests return 402 budget_exceeded until 00:00 UTC, while the account's other keys keep working. A sub-agent key's budget and limits stay within its parent's.

Set them when you create the key (npx -y scaledthought@0.3 keys create --name <name> --budget-usd 5), later with PATCH /v1/keys/{id} (daily_budget_usd, rate limits), or over MCP with budget_manage. A webhook for budget.exhausted tells you when one runs out.

Default rate limits per key:

npx -y scaledthought@0.3 usage --days 7 --by key gives requests, tokens and cost by model, day or key (GET /v1/usage). npx -y scaledthought@0.3 billing transactions is the credit ledger: purchases, auto-reloads, usage, refunds (GET /v1/billing/transactions). Responses carry x-st-notice: low_balance once the balance is under $5.