Quota forecasts you can trust

Estimate time to exhaustion from reported balances without inventing provider caps.

v0.3.0 · Updated November 2, 2026
On this page

Start with reported balances

Conflux stores reported remaining-quota observations per provider, key, model and metric. Remaining requests, tokens and Cloudflare Neurons are distinct units. If the provider exposes no balance, the console shows unknown instead of a fabricated bar or cap.

When a forecast appears

  • At least two decreasing balance observations span one minute or longer.
  • The latest observation is no more than five minutes old.
  • The observations belong to the same reset window.
  • The expected reset does not occur before projected exhaustion.

Observed burn is balance decrease divided by elapsed time. Remaining balance divided by that rate gives the estimated time to exhaustion. An amber warning appears below approximately 60 minutes.

Why a provider has no estimate

Flat, stale, replenished, insufficient or incompatible observations suppress the forecast. A sample can describe one key/model/metric rather than the total account. Estimates assume recent burn continues and cannot predict a sudden workload change.