DooDooLamb News
Show HN: jevals – replacing LLM judges with typed Jev decisions
Brief published September 22, 2026 · Original source published September 20, 2026
Original reporting by gbayomi at github.com.
Automated brief. Verify important details at the original source.
What happened
Openlayer AI released jevals, an open-source library on GitHub that replaces LLM-as-judge evaluation with what it calls Jev-style typed decision models. Rather than sending each evaluation to a large language model separately, jevals batches all evals for a given agent trace into a single request. The project claims this approach costs a few thousandths of a cent per request and returns results quickly, positioning it as a cheaper and faster alternative for agent evals and guardrails.
Why it matters
LLM judges are a common but expensive and slow bottleneck in agent evaluation pipelines. If jevals delivers on its claimed cost and latency profile, teams running high-volume agent traces could reduce eval overhead substantially. The typed decision model approach also suggests more deterministic, auditable outputs compared to freeform LLM scoring.
What to watch
Adoption will depend on how well Jev-style decision models generalize across diverse agent tasks, whether the library integrates with existing eval frameworks, and how the project handles community contributions given its early stage with minimal public engagement so far.