oz0 compare-savings

Usage and flags for oz0 compare-savings.

For: Operators and developers looking up a command.

A comparison runs an agent as defined (the baseline) and with one change (the variant: another model, lazy or eager tool loading, a narrowed tool set, the prompt cache off, large tool results offloaded or not) on the same cases, several runs of each, as shadow runs: writes answer from a stub and no approval is asked for, so nothing acts on the world. The gateway meters every run.

A saving is stated only when the variant's pass rate is equal to the baseline's within a margin (non-inferiority, 5 points at 95% by default) and the interval of its billed cost per successful run is below the baseline's; otherwise the result says there is no measured saving, and why. A comparison spends the tenant's money and budget: start --dry-run shows the estimate and the power first. It proves nothing about other cases, other times or another tenant's work.

Usage

oz0 compare-savings [command]

Subcommands

Inherited flags

FlagTypeDefaultDescription
--contextstring$OZ0_CONTEXT or <config dir>/orchestrator-zero/context.jsonCLI context file written by server init
--log-levelstringinfolog level: debug, info, warn or error

See all oz0 commands and the global flags.

Copyright © 2026