Why use this instead of reading the official docs?

The docs give the contract; this makes the docs-only details interactive. Parallel fan-out, input-only billing, Noul having no confidence, and code owning the workflow are all runnable in the lab instead of asserted.

When should I pick Jev over a generative model?

Use Jev for classify, route, score, and verify on hot paths with closed schemas and calibrated probabilities. Prefer a generative model for prose, code, open answer spaces, counting, or date math.

Where does my API key go and is it logged?

Set TYPESAFE_API_KEY or a key file under ~/.typesafe or ~/.config/typesafe with mode 600. The key is never printed, never returned by /api/health, and never committed.

What is the cheapest way to try it without a key?

Clone, install requirements, and open the operator lab or smart-home demo. Offline tabs, cost lab, patterns, jaggedness, and fixtures all run keyless; only the live judge needs a key.

What under-known facts does the repo highlight?

That Jev is a function call rather than a chatbot, that questions run in parallel with a fan-out about 12 times cheaper than serial, that output is free while Noul has no confidence field to copy onto a Choice, and that there is no fine-tune so answers are shaped by state, instructions and criteria.

How do I run the lab and demos?

Clone the repo, create a virtual environment, pip install -r requirements.txt and run python app.py. The smart-home demo is at /home and fans out 13 questions in one call while code flips the lights, and the operator lab, cost lab, patterns and jaggedness tabs run keyless. Docker works too via docker compose up --build.

What does the TypeScript harness CLI add?

The packages/js CLI packages the same System One questions for agent loops, with npx jev doctor, npx jev route and npx jev guard commands for model routing, semantic retrieval, LLM guardrails and reasoning-trace classification. It is wired into loop-engineering as loop-jev.

CommunityJev Practices & Patterns

Jev — cobusgreyling

Unofficial operator companion showing how Jev works as a function call, with an offline lab, smart-home demo, TypeScript harness CLI, and portable agent skills.

See highlights

Open github.com

Stars: 136
Language: Python
License: MIT
Published by: Community
Author: cobusgreyling
Last verified: Oct 2026

cobusgreyling/Jev

User repository on GitHub

View on GitHub

Unofficial TypeSafe Jev showcase — System One decisions, not chat.

Forks: 30
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

This is one of several Jev listings for the same project. The canonical entry is Jev — mayank953.

About this project

Unofficial operator-level companion for TypeSafe Jev Choice · Score · Noul · parallel fan-out · confidence as a second axis Jev 1.13 · released 15 September 2026. What people miss · Announcement · Docs · Models · Article · Showcase framework. Most posts stop at “fast structured output.” This repo surfaces the docs-only details that change how you build: Jev is a function call, questions run in parallel, output is free, confidence is not probability, and code owns the workflow.

Quoted from the project's README. Read the full README.

Tagscommunity awesomejev catalog-import

Works forJev for Classification

Used inJev for Software & DevTools

Languages: Python

Highlights

Smart-home demo fans out 13 questions in one call while code owns lights and workflow decisions.
Lab tabs cover cost, confidence routing, composite scoring, jaggedness, guardrails, and scenarios.
Harness CLI packages routing, guardrails, retrieval, and trace classification for agent loops with doctor checks.
Four installable skills cover primitives, speculative fan-out, guardrails, and confidence-lane routing.

Watch out

Explicitly unofficial and not a replacement for the TypeSafe SDK; live judging needs a key while offline tabs run without one. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev_local

Local unofficial /v1/systemone adapter that serves Choice, Score, and Noul probabilities via first-token LLM judgments or a fine-tuned ModernBERT scorer, including Japanese docs.

Jev Practices & Patterns#community#awesomejev#catalog-import

Community

Argos1111/jev_localPython

jev-on-a-laptop

Unofficial laptop study of parallel constrained decoding: prefill once, share the KV cache across fields, and decide all schema fields in one batched forward pass without writing JSON.