@_vMyth’s jevals runs agent evals and guardrails as one Jev decision request per trace, cheap enough to score every run.

Benchmark / library. The post says agent evals and guardrails run as one request to a Jev decision model (not an LLM judge), cheap enough to score every trace, with quick-start defaults that send traces to a hosted API and a gate if it is down, linking github.com/openlayer-ai/jevals.