Credits & billing

Reservations, settlement, top-ups, spend caps and low-balance warnings.

How a request is billed

Before the model runs, the gateway reserves an estimate (input size × input price + max_tokens × output price) from your live balance. If the available balance is below the estimate you receive 402 and nothing runs.
After the model answers, the reservation is settled on the tokens the model reported; the unused part is released immediately.
If the model never reported usage (rare upstream failure), the request is not charged and is marked for reconciliation.

Prices

Prices are per million tokens in INR, set per model. A price change is published as a new pricing version with an effective time; requests already in flight keep the version they started with.

Top-ups

Add credits under Billing with UPI, cards or net banking. Credits are prepaid, non-transferable and usable only on Top Billing; see the credits policy for refunds.

Bonus credit

Larger top-ups earn extra credit: 5% of the credited amount from ₹5,000 and 10% from ₹20,000. The bonus is added to the wallet with the payment and shown as its own line; GST and the tax invoice stay on what you paid. If a top-up is refunded, its bonus is taken back in the same proportion.

Spend caps

Caps stop spending before a runaway agent or a leaked key empties the wallet. A request that would take a cap past its limit is refused with 402 spend_limit_reached before the model runs; it is never cut short halfway.

Per API key — a daily and/or a monthly cap, set under API keys.
Per coding session — one cap for every Claude Code session on the project (identified by the x-claude-code-session-id header), set under Projects.
Per project — a monthly cap on the project's wallet, set under Projects.

Days and months are calendar periods in India Standard Time. What counts towards a cap is what has been charged plus what requests in flight have reserved, so parallel agents cannot overshoot it together.

At 80% of a key or project cap you get one email per cap per period. Every response also carries a tb-spend-warning header once a cap is at 80% or more (for example key_daily=84%, session=91%), and GET /v1/usage reports each cap with what is left.

Low balance

You are warned in the dashboard and by email when a live wallet with active live keys drops below the low-balance threshold you set. The CLI's topbilling usage separately estimates when the balance will run out from recent spend.