jevbooks

← All projects

jev-cli

Nasrallah-AL/jev-cli · Homepage

Typed AI judgments from the command line, powered by TypeSafe's Jev.

Verification & Guardrails64%Runner-up: Evaluation & Benchmarking

What it is

A TypeScript CLI (installed as jevctl, command jev) that wraps TypeSafe's Jev model into pipeable, exit-code-gated shell commands for verifying claims, screening text, classifying, extracting, matching, routing, finding, reranking, compacting transcripts, and batching over JSONL.

How it uses Jev

Each command sends content plus a fixed-answer question to Jev and uses the returned per-answer probabilities. Thresholds and --fail-on policy live in code, not prompts; exit code 2 gates CI. Works over TypeSafe, OpenRouter, or Cloudflare Workers AI.

Primitives:choicescorenoul

Technique worth stealing

Thresholds and --fail-on policy live in code, not prompts, so judgments become exit-code gates in CI and hooks.

Try it

npm install -g jevctl; jev auth login; jev verify "..." --evidence "..."

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.64
  • Shows a System One patternyes0.89
  • Handles uncertaintyunclear0.55
  • Measuredno0.10
  • Runnableyes0.98
  • Worth recommendingyes0.65
  • Model replicano0.03
  • Problem scopescore on a 0–2 scale1.27
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.