What is the TypeSafe Jev Decision Studio?

An interactive workbench for Jev's three decision primitives, built as a Next.js app and powered by typesafe/jev-1.13 via OpenRouter. It is meant for exploring, benchmarking and deploying Choice, Noul and Score decisions rather than generating text.

Which Jev primitives does the studio cover?

All three: Choice multi-class categorization (demoed over 95 self-hosted project categories in one forward pass), Noul binary evaluation with transfer curves and configurable thresholds, and Score ordinal rubric rating with probability breakdowns across tiers.

What do I need to run the studio?

Node 18.18+ or 20+, an OpenRouter account with credit and access to typesafe/jev-1.13, and a .env with OPENROUTER_API_KEY and JEV_MODEL. Then npm install and npm run dev, and open http://localhost:3000.

How fast are evaluations in the studio?

The README reports 400-600 ms end-to-end via OpenRouter inference, contrasting it with 1-4 second latency for generative LLMs asked to classify. That figure is the author's own measurement, not an independent benchmark.

Is this an official TypeSafe project?

Not as documented; it is a community workbench by Romain-Jochum that accesses Jev through OpenRouter. The README badges MIT, but the GitHub metadata lists no licence, which is worth checking before you depend on it.

Why use this instead of prompting a chat model to classify?

The README's pitch: Jev returns mathematically calibrated probability distributions directly, so there is no regex scraping or JSON repairing, and results are natively typed. Generative models add non-deterministic output and slower inference for the same task.

Can I point the studio at a native TypeSafe key?

The documented setup goes through OpenRouter: you set OPENROUTER_API_KEY and JEV_MODEL=typesafe/jev-1.13. The README excerpt in the digest does not describe a direct TypeSafe endpoint option or pricing beyond OpenRouter credit.

Jev directory

CommunityJev Demos & Experiments

TypeSafe Jev Decision Studio

An interactive Next.js workbench for Jev's three decision primitives through OpenRouter: rank 95+ categories with one Choice pass, test binary Noul decisions with transfer curves, and grade inputs on ordinal Score rubrics.

See highlights Open github.com

Stars: 1
Language: TypeScript
Reading time: 2 min
Published by: Community
Author: romain-jochum
Last verified: Oct 2026

romain-jochum/typesafe-jev-decision-studio

User repository on GitHub

View on GitHub

Fast, calibrated System One decision platform powered by TypeSafe Jev via OpenRouter. Sub-second logprob scoring, transfer curves, zero hallucinations.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity catalog-import logicrw calibration openrouter workbench

Works forJev for Classification Jev for Scoring Jev for Verification

Used inJev for Software & DevTools

Languages: TypeScript
Platforms: Next.js OpenRouter
Capabilities: Open sourceRequires API key

Highlights

Covers all three primitives: Choice over 95 categories in one pass, Noul binary verification with transfer curves, and Score ordinal rubric rating.
Talks to typesafe/jev-1.13 through OpenRouter; the README reports 400-600 ms end-to-end evaluation.
Pitch against generative LLMs: logprob-level scoring returns calibrated distributions directly, so no regex scraping or JSON repairing is needed.
Auto-detects server keys and shows a Key status badge; keys can also be configured in-browser via Settings.
Built with Next.js 15.5, React 19 and TypeScript 5.8; needs Node 18.18+ and an OpenRouter account with credit.

Quickstart

bash

git clone https://github.com/Romain-Jochum/typesafe-jev-decision-studio.git
cd typesafe-jev-decision-studio
npm install
npm run dev

Watch out

Grounded in the truncated README and GitHub metadata reviewed on 2026-10-15; the README is cut off inside the Choice example, its latency and 95-category figures are the author's own, the GitHub metadata lists no licence although the README badges MIT, and a funded OpenRouter account is required.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

Per-Tick Decisions in Games

Sep 28, 2026

A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.

Jev Demos & Experiments#gaming#agents#control

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai