1,204,318events counted this cycle1 client retry dropped
Meter it Every call becomes one event, and a client retry is counted once.
Stop building billing.Metering, quotas, per-user balances and top-ups in one API.
Customer balance
7,400credits
- Allowed
- 0
- Denied
- 0
- On hold
- 0
- authorize req_8841 allow -312
- authorize req_8840 allow -96
- authorize req_8839 allow -188
Simulated in this page. No network calls.
Metering every event once
Contracts every customer's price
Balances per customer and per member
Quotas checked before the model runs
Retries safe to send twice
Top-ups one crossing and one charge
Statements final means final
Portal self-serve and scoped
Budgets company, pool and member
Errors a code for every failure
You set out to ship a model.
Then a customer asked for a quota.
Then finance asked for an invoice.
None of that is your product.
So we run it.
Skip to the summaryEventsexactly once Rate cardcontract overrides Balancesper customer, per member POST /v1/authorizebefore the model runs Budgetscompany, pool, member Auto top-upone crossing, one charge StatementOctober Portal tokenscoped to one customer
What you stop buildingEach job below leaves your backlog.
- Event pipeline and dedupe
- Per-contract pricing
- Per-user balance ledger
- Real-time quota checks
- Budgets and spend limits
- Auto top-up and retries
- Monthly statements
- Customer portal
0 of 8 off your backlog
video.generate640 seconds× 210 credits
Kestrel Labsrate card
134,400credits
Halden18% contract discount
110,208credits
Same usage.Different contract.No code change.
Price it Each customer pays their contract rate, so a negotiated deal needs no code.
Kestrel Labs total1,975
req_3112 yuki215 creditsprepaid 215
- Inesown seat600
- Tomasown seat300
- Yukiown seat0
- Team poolshared0
- Prepaidshared1,075
Track it A live balance for every customer and every member, drawn down in a fixed order.
Stopped
- req_1042180 creditsallowed
- req_1043240 creditsallowed
- req_1044200 creditsinsufficient_balance
balance left 80
Gate it The check runs before your model does. At the limit the request stops, and nothing is spent.
Tomas stopped.
member_spend_limit_reached
Ines and Yuki keep working.
- 2,4001,500Company
- 1,8001,500Pool
- 900300Inesallowed
- 900900Tomasdenied
- no limit300Yukiallowed
Cap it Limits for the company, the pool and each member. Only the person at the limit stops.
4requests crossed
1top-up opened
6,720credits in the balance
Recharge to 8,000Threshold 3,0003 saw it open5,750 credits charged
Refill it When the balance crosses your threshold, one top-up runs, however many requests cross it together.
Kestrel LabsOct 1 to Oct 31
- Standard seats, 8 at $30.00$240.00
- Seat added Oct 17, 15 of 31 days$14.52
- Overage, 4,180 credits at $0.01$41.80
Total$296.32
FinalizedCorrections go on a credit note.
Bill it Seats, credits and overage close into one statement. Finalized means final.
Self-servePricing stays out of reach.
Token for Kestrel Labsbalance:readbudgets:writetop_up:write
- Reload thresholdwritable
- Member limitswritable
- Contract discount403 portal_scope_violation
Hand it over Customers manage their own budgets and top-ups through scoped tokens.
Two calls at runtime.Set up customers and contracts once.
Waffo runs
- Event pipeline and dedupe
- Per-contract pricing
- Per-user balance ledger
- Real-time quota checks
- Budgets and spend limits
- Auto top-up and retries
- Monthly statements
- Customer portal
You keep
- Your model
- Your product
- Your prices
- Your customers
// Call 1. Ask before the model runs.
const decision = await billing.authorize({
request_id: job.id,
customer_id: "cus_kestrel",
seat_id: "ines",
product_id: "video.generate",
estimated_quantity: "12",
});// Branch on the answer.
if (!decision.allowed) {
return reject(decision.reason);
}// Run your model.
const used = await runModel(job);// Call 2. Report the actual quantity.
await billing.ingest({
events: [{
transaction_id: job.id,
customer_id: "cus_kestrel",
seat_id: "ines",
event_type: "video.generate",
timestamp: new Date().toISOString(),
properties: { seconds: String(used) },
hold_id: decision.hold_id,
}],
});Try the rules yourself.
Each demo below runs the same billing model that drives this page.
Every customer gets their own price
Rates, discounts and fixed prices live on the contract, so a negotiated deal needs no code.
One rate card
Set list prices once, by product and model.
Overrides on the contract
Add a percentage discount or a fixed price. The two customers above share a rate card and get different bills.
Exact arithmetic
Usage is rated once, with one rounding rule. Money is a decimal string or an integer, and floats are rejected.
One request, four pockets
Each request draws from seat credit, then the team pool, then prepaid balance, then overage.
Seats carry credit
A seat tier sets the included credit each member gets every month. Tier credit can be zero.
A shared pool
When a seat runs out, the request moves to the pool the whole team shares.
Overage is opt-in
Overage is off by default. Turn it on per contract.
Checked before it runs, settled after
Authorize holds the estimate. Reporting usage swaps the hold for the actual cost.
Holds stop races
Concurrent requests count held amounts, so two jobs cannot spend the same credit.
Failed tasks cost nothing
Release the hold and every pocket gets its credit back.
Retries are safe
Authorize twice with one request id and you get the first decision back. Try it with the Retry button.
Top up before anyone notices
When the balance drops below your threshold, one payment refills it. The liquid below is the balance.
One crossing, one charge
Any number of requests can cross the threshold together. Only one top-up opens. Press Race to see it.
Failed cards
A failed payment leaves the balance alone and tries again on the next check, so nobody is charged twice.
Frequency limits
Cap top-ups per window, so a runaway job cannot empty a card.
Limits at every level
Set a monthly budget for the company, a quota for the pool and a limit for each member.
Only one person stops
A member at their limit is blocked. Everyone else keeps working.
Takes effect at once
Lower a budget below current spend and the next request stops.
Hard or soft
Stop requests at the limit, or keep serving and send an alert.
A statement that explains itself
Every month closes into one statement, and each line opens to the usage behind it.
Finalized means final
A closed statement never changes. Corrections are credit notes.
Late events still count
Usage that arrives after close moves to the next statement, labeled with the day it happened.
Collect your way
Statements go to your payment provider or your own invoicing.
Give customers the controls
Your customers manage their own budgets and top-ups through scoped portal tokens.
Scoped tokens
A portal token covers one customer and a short list of fields. Pricing stays out of reach.
API first
Everything the portal can do is an API call, so it fits inside your product.
Built for the bad day
Every error has a code, an HTTP status and a retry answer, and says what was charged.
Denials are answers
A denied request returns 200 with a reason, so your code can branch on it.
Every write is idempotent
Send the same Idempotency-Key again and the first result comes back.
Fails closed
If the money path is down, included credit still works and prepaid spend stops.
Start with the guide
See how the pieces fit, then read the contract for each endpoint.