How does this compare to calling the hosted Jev API?

Both score every option in one forward pass with temperature-scaled softmax per question and no text generation, so answers stay within supplied options. The encoding differs: DeBERTa packs pairs with slot tensors while Qwen variants use delimiter blocks with a derived causal mask.

Which of the three models should I load?

The small default model suits constrained devices with a large context window, the larger Qwen-class model is most accurate but needs a capable GPU, and the DeBERTa-class model has a short context with multiple weight precisions. Pass an alias or any repo whose config carries the expected section.

What does wiring it into my app look like?

Install the library with its Transformers.js peer dependency, load with model, dtype, device, and progress options, then call decide with state plus array or keyed questions. Concurrent calls queue serially and oversized questions throw.

Will a long document silently corrupt my questions?

State text truncation defaults to cutting trailing tokens with an error-throwing opt-in, alongside token budgets, temperature defaults per family, and caller-text escaping so inputs cannot forge delimiters. Check token counts before sending long states.

Which model is the open-jev default and how big is it?

kev-0.6b, a Qwen3-0.6B-Base model with an 8192-token context, about 0.34 GB at q4f16. The two alternatives are kev-4b and the DeBERTa-based open-jev.

How do context and option limits differ by model?

open-jev has a 512-token context and allows 2 to 255 choice options and 2 to 10 score levels. The kev models allow an 8192-token context, 1 to 255 choice options, and 2 to 255 score levels.

How do I check a state against the token budget up front?

Call jev.countTokens(text) to get the token count without markers, then compare it to maxStateTokens. The default is 256 for open-jev and 8192 for kev.

CommunityJev Demos & Experiments

Open Jev — nico-martin

Browser-focused TypeScript library for typed decisions that runs open model reproductions on-device via Transformers.js, returning one calibrated distribution per question.

See highlights

Open github.com

Stars: 41
Language: TypeScript
License: MIT
Published by: Community
Author: nico-martin
Last verified: Oct 2026

nico-martin/open-jev

User repository on GitHub

View on GitHub

open-jev is a browser-focused TypeScript library for typed decisions: one piece of text (the state) plus any number of typed questions go in, and one forward pass returns a calibrated probability distribution per question. Nothing is generated, so an answer is always one of the options you provided.

Forks: 9
Last pushed: Updated Sep 2026

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

About this project

open-jev is a browser-focused TypeScript library for typed decisions: one piece of text (the state) plus any number of typed questions go in, and one forward pass returns a calibrated probability distribution per question. Nothing is generated, so an answer is always one of the options you provided. It runs open reproductions of the shape of TypeSafe AI's Jev "System One" model via @huggingface/transformers (Transformers.js). Everything happens on-device: WebGPU when available, WebAssembly otherwise.

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

Tagscommunity jevbest catalog-import

Works forJev for Classification

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

State plus any number of typed questions resolves in one forward pass entirely on-device over WebGPU or WASM.
Choice, score, and yes/no builders with fully typed answers flowing option literals into results.
Model aliases span small, large, and DeBERTa-class weights with dtype, device, and cache-aware loading.
Keyed or array questions, token budgets, truncation modes, and serial queuing for app integration.

Watch out

On-device reproductions of the Jev shape, not the hosted model; accuracy and fit depend on the chosen open weights and quantization. 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.

Other listings of this project

Forks and copies of this project in the directory. This entry is the canonical one.

Open JEV

An independent local alternative to Jev: it takes English or Chinese context, questions, and candidate answers, and returns probabilities over the supplied candidates from a frozen Qwen3-4B backend.

Jev Repos & SDKs#community#python#open-models

Community

zhihz/openjevPython

open-jev

Open harness that renders allowed JSON answers on a diffusion canvas, denoises for a fixed step budget, then picks the top allowed tokens from final logits with no training.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

JoshuaSP/open-jevPython

open-jev

Unofficial PyTorch reconstruction of the ideas behind Jev, encoding state once and answering each question through small typed readout heads in a single parallel pass.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

kyegomez/open-jevPython

More Jev resources like this

laya-server

Self-hosted Docker service and web UI that packages Laya structured-decision inference behind a compatible System One endpoint.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

1Panel-dev/laya-serverPython

jev-browser-skill

Standalone CLI skill that drives isolated Playwright Chromium with Jev choosing scoped in-page actions, defaulting to Vercel AI Gateway with TypeSafe as an option.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

hqman/jev-browser-skillTypeScript

jev-explained

Interactive playground that teaches Jev by running real typed noul, choice, and score questions and showing the resulting probabilities.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

davila7/jev-explainedTypeScript

Jevbridge

Zero-dependency Node 22 adapter that puts typed Jev decisions beside any LLM, with swappable backends, confidence gates, MCP tools, an ACP sidecar, and offline-ready recipes.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

gamesonrblx/JevbridgeTypeScript

ulka

Local browser-agent extension where FX plans tasks and Jev picks constrained actions from observed elements, with model calls routed through Vercel AI Gateway.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

razaanstha/ulkaTypeScript

jev-design-test

Design sampler that turns a one-line prompt into a canvas of full-page screens by asking Jev to pick among fixed shadcn layouts instead of writing code.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

bhaiG-de/jev-design-testTypeScript

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

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai