Token FinOps: Make AI Spend Allocable and Stoppable
A self-owned rate card turns tokens into real cost, tags land spend on cost centers, and daily / monthly cost quotas turn budgets into request-blocking gates.
What the community actually complains about
Vendor list price is not your real negotiated cost — discounted and private prices have nowhere to be recorded, so cost reports are fiction; when a model drops its price mid-cycle, historical months get polluted by the re-priced history.
Chargeback is a manual CSV exercise, and overspend is only discovered after the fact.
How GateLLM does it
GateLLM gives you a self-owned rate card: a hand-entered private price (pinned) becomes authoritative and system refreshes never overwrite it. Each price is a snapshot that takes effect by time window — backfill history or pre-schedule a price cut, without retroactively rewriting cost.
Tags roll keys / key groups into cost centers, and daily / monthly cost quotas inherit key → group → global, judged independently on UTC calendar day and month, blocking the moment either trips.
5 levers that actually land
- Self-owned rate card: pinned private price shadows auto, refreshes never overwrite
- Price snapshots cover [own time, next time), backfill history / pre-schedule future
- Three billing modes: per-token, per-call, and per-second (transcription, video generation)
- Tag keys / key groups and aggregate usage & cost by tag
- Daily / monthly cost quotas: three-level inheritance + max fold, 429 quota_exceeded on trip