Agents heat 1 · Quiet★ 1heat incomplete (no X)

jeval

A framework for decision-model harnesses: your domain offers short next-step menus, the model (including typesafe/jev-latest via pi-ai) picks one until it can answer typed questions, with full traces.

What it is

Look around with typed menus before the final answer.

What it does: Runs a decide→observe loop for any domain you define.

Why it belongs in the showcase: General harness with a real Jev-backed domain.

Caveats: Library-from-git.

Evidence

Public MIT repo, created 2026-10-25, about 1 star as of 2026-10-26. Ships a legal-judge domain and CLI (ingest/run/score/rejudge/compare). Decider example uses piAiDecider("typesafe/jev-latest", { apiKey: process.env.TYPESAFE_API_KEY }). Distinct from CMaintz/jev-eval (already listed).

Caveats

  • Install pins a GitHub commit rather than a stable npm release
  • legal-judge needs Poppler/LibreOffice/Netpbm for some documents
  • Name is close to jev-eval; different product

Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.