Skip to content

jev-cookbook

Fifteen runnable Node recipes using Jev via OpenRouter for triage, dedupe, PII scanning, reranking and a browser agent.

Articlebeta

Overview

Jev Cookbook by Jeroen Erne is a set of fifteen practical recipes that call Jev through OpenRouter's decisions endpoint. Each recipe ships a runnable script, a small labelled dataset and measured results, covering support triage, database indexing, a file organiser, tagging, category trees, duplicate detection, PII column scanning, bank transactions, invoice extraction, search, log triage, moderation, lead scoring, a browser agent and a Gmail labeller for your own inbox.

It is aimed at developers learning to use typed decisions in real jobs. Every recipe routes on confidence and measures its errors, sending low-confidence answers to human review while deterministic code owns thresholds and actions. The README quotes Jev's price as $0.042 per million input tokens with free output, and estimates 2 to 8 cents per 1,000 items across the recipes. A sibling project, Jev Browser Control, turns the browser recipe into an MCP server.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich team should handle this message?billing · technical · sales
Yes / noDoes the customer ask for a refund?—
Yes / noAre these two records duplicates?(inferred)—

How it works

  1. Pick a recipe and set OPENROUTER_API_KEY
  2. Send items as state with Choice or Noul questions
  3. Accept high-confidence answers automatically
  4. Route low-confidence items to a human review band and compare against labels

Strengths

  • Each recipe includes measured results on labelled data
  • Low cost estimate: 2 to 8 cents per 1,000 items
  • Confidence routing pattern reused across domains

Caveats

  • Runs through OpenRouter's alpha decisions endpoint
  • Small datasets per recipe

Getting started

Set OPENROUTER_API_KEY and run any recipe script with Node; requests go to https://openrouter.ai/api/alpha/decisions.

At a glance

Type
Article
Status
beta
Language
JavaScript
Runtime
Node scripts
Has code
yes
Has tests
yes

Use cases and integrations

How curated lists describe it

  • Developer education: 15 runnable Node recipes that route support tickets, file documents, categorize bank transactions and label Gmail with Jev `Choice` and `Noul` questions, sending low-confidence answers to human review.

    yibie/awesome-jev
  • Community cookbook of 15 runnable recipes where Jev picks categories, tags, dates, duplicates and next browser actions while deterministic code owns thresholds, review bands and every action.

    cobanov/awesome-jev
  • Practical, tested recipes for TypeSafe's Jev decision model on OpenRouter: support triage, database indexing, file organizing, tagging, taxonomies, dedupe, PII detection, extraction, search re-ranking and a browser agent.

    hellogumbo/awesome-jev
  • Fifteen runnable recipes calling Jev through OpenRouter - support triage, dedupe, PII column scanning, reranking, moderation and more - each shipping the output it produced beside it.

    MrJev/awesome-jev

Frequently asked questions

What kinds of decisions are covered?

Ticket routing, tagging, duplicates, PII, transactions, extraction, reranking, moderation, lead scoring and browser actions.

How does it handle mistakes?

Each recipe routes on confidence and sends uncertain answers to human review.

Which key do I need?

An OpenRouter API key, since recipes call Jev through OpenRouter.

Related projects

  • Two beginner Python apps using Jev: a message sorter and a support ticket router that send uncertain items to human review.

    ProjectprototypePython
  • Python sandbox turning Jev issue-category and reproduction judgments into testable routing proposals, with offline replay.

    ProjectprototypePython
  • .NET 10 and React 19 ticket triage where Jev answers five typed questions in one call and .NET business rules make the final routing.

    ProjectprototypeC#
  • Data-incident router and use-case guide: registry facts first, then Jev Choice, Score and Noul with confidence-gated human review.

    ProjectprototypePython
  • Small Python demos of Jev: parallel Noul ticket checks with measured speedup, and a three-way Choice citation and claim checker.

    AppprototypePython
  • Six side-by-side demos where Jev makes the decision, Claude or Kimi writes the words, and code owns control flow.

    AppprototypeTypeScript