Usage, Billing, and Quota
Weaver meters billable operations per organization and project. North Ledger is the source of truth for active SKU prices, immutable usage events, settlement, and billing summaries; Weaver stores access policy, quota reservations, and a durable delivery outbox.
What is metered
Token usage is separated into four model-specific kinds:
| Usage kind | Meaning |
|---|---|
| Training | Tokens processed by training forward/backward operations |
| Prefill | Uncached sampling input tokens |
| Cached prefill | Sampling input tokens served from cache when reported by the router |
| Output | Generated sampling tokens |
Prices are displayed in USD per one million tokens. Weaver snapshots the active SKU and price when an operation is admitted, so a later price change never rewrites accepted work or historical usage.
Usage is attributed to the organization, project, user, Session, Training Run, and operation that produced it. Team membership only grants access; it does not become a billing boundary or duplicate usage.
Quota balance
Every user's quota account inside an organization starts at USD 0. The requesting user needs an approved grant in that organization before their billable work can be admitted. The Console always displays the current user's balance, including precise sub-cent values and negative balances after an actual operation exceeds its reservation.
Before billable execution Weaver reserves estimated cost. It then settles exactly once using completed token counts, or releases the reservation for failed, cancelled, or expired work. Missing SKU/price attribution fails closed before execution; concurrent reservations cannot spend the same balance below zero.
The primary error contracts are:
| Status | Meaning |
|---|---|
402 quota_exceeded | Available quota cannot cover the estimated operation |
429 rate_limited | Retry after the returned Retry-After interval |
503 metering_unavailable | Price, SKU, attribution, or Ledger dependency is unavailable |
Rejected requests do not create billable usage.
Requesting quota
From Usage & Billing, submit an amount in USD and an optional reason. The request belongs to the current user and organization. A Weaver system administrator can approve or reject it from System Management; approval creates one immutable grant for that user account and cannot be applied twice.
Quota approval is currently an operator workflow. Weaver does not yet connect a payment gateway, automatic top-up, invoice collection, or credit-card checkout. An approved quota balance is permission to consume metered service, not proof of an external payment.
Reading usage
The Console reads North Ledger-backed summaries grouped by day, model, and operation. A project selection adds the project as the billing instance filter; choosing all projects aggregates the organization. Ranges longer than Ledger's maximum query window are split and merged without rounding through floating point.
For reconciliation, compare one isolated Training Run across:
- completed operation token counts;
- Weaver quota reservations and settlements;
- the durable Ledger delivery outbox;
- North Ledger usage grouped by SKU;
- the resulting quota balance.
Training, prefill, cached-prefill, and output quantities must match their exact integer token totals for that run. Charts and metric downsampling are observability views and never determine billable totals.
Price changes
Only a Weaver system administrator can edit prices. A price update creates and activates a new North Ledger SKU version, then switches Weaver's active binding. The old SKU remains active and immutable so delayed events and existing reservations settle against their original snapshot. Every price mutation is written to the control-plane audit log.