pip install cruqai. Authenticate with cruqai login, or set
CRUQAI_API_KEY and CRUQAI_BASE_URL in the environment.
--json for raw output.
Agents & runs
cruqai runs and cruqai trace print - where the provider reported no usage.
That means “not measured”, not “free” - see
Observability.
Evals
--model repeats, and takes an optional label= prefix. --scorers accepts
inline JSON or @file.json. With --watch the command polls until the
comparison finishes and exits non-zero if any cell errored, which is enough to
gate a CI step. See Evals.

