Credits & cost

Live engine runs are the dominant cost of running Lokrix. They are metered in credits against your workspace balance — one credit is roughly one engine response. This page explains how a scan spends credits so nothing surprises you.

How seats and usage credits meter live runsPlan seats and usage credits feed a metered live-run step (cost equals engines times prompts times K samples), which writes to a per-tenant metering ledger.Plan seatstiered subscriptionUsage credits1 credit ≈ 1 responseLive runs meteredengines × prompts × KMetering ledgerper-tenant balance

Cost is multiplicative

The credit cost of a live scan is the product of four factors:

  • Prompts — how many questions from the prompt universe are run.
  • Engines — how many answer engines each prompt is sent to.
  • K samples — how many times each prompt is repeated to build the confidence interval.
  • Cadence — how often the scan repeats (one-off, daily, weekly).

Because these multiply, small increases compound quickly: doubling prompts and engines quadruples the run. Lokrix shows the estimated credit cost before a live scan starts.

Keeping cost down

  • Default to predictive. The calibrated model scores without credits; escalate to live runs deliberately. See Live vs predictive.
  • Cache. Recent live results are reused where valid instead of re-running, and labelled as cached.
  • Focus. Run high K only on the prompts that matter; run the long tail predictively.

For plans, top-ups and metering details, see Credits & metering and Budgets & spend caps.