Budgets & spend caps

Because live-run cost is multiplicative — prompts × engines × samples × cadence — a single over-eager schedule can burn through credits fast. Budgets let you cap what a workspace can spend on live engine runs, so measurement stays predictable and you are never surprised by a bill.

Estimate before you launch

Every scan shows a credit estimate before it runs, derived from the same four factors that set the cost: prompts × engines × samples (K) × cadence. Reviewing that estimate is the first control — it tells you the ceiling of a one-off run and the recurring draw of a scheduled one. See scan cost for how the estimate is built.

Workspace spend caps

A spend cap sets a maximum number of credits a workspace may consume in live runs over a period. When a scan would push consumption past the cap — or past the remaining balance — Lokrix stops before overspending rather than continuing silently. The cap is a hard boundary, not a warning after the fact.

  • Per-workspace — the cap and the credit balance are both held at the workspace level, shared by every property inside it.
  • Pre-flight — the check happens against the scan estimate before the run starts, and again as responses are metered during the run.
  • Fail-safe — hitting the cap pauses further live runs; predictive scores and cached data keep working.

Levers that lower spend

Because the factors multiply, trimming any one of them has an outsized effect on the total:

  • Default to predictive. The calibrated predictive model returns everyday scores and what-if simulation without spending credits; escalate to live runs only when you need ground-truth evidence. See live vs. predictive.
  • Right-size the prompt universe. Focus live runs on the prompts that matter most to your buyers rather than every generated variant.
  • Tune sampling (K). More samples tighten the confidence interval but cost more; pick the smallest K that gives a usable interval. See sampling.
  • Choose cadence deliberately. Daily live scans multiply cost roughly 30× over monthly; reserve high cadence for the properties and engines where freshness matters.
  • Scope engines. Five engines run live by default; enabling AI Overviews and AI Mode requires a search-provider key and adds live responses per prompt.

Related