Can I use the standard TypeSafe client against a local MoJev server?

Yes. It is wire-compatible with the TypeSafe Python SDK system_one API, so you run its server locally and point the official client at the local endpoint.

Do I have to retrain or register the answer options ahead of time?

Candidates are supplied with the request and read directly, with typed Choice, Noul, and Score questions packed into one model invocation over shared state.

How does the large context window actually work in practice?

It supports a large native context with YaRN scaling higher, while training truncates state inputs and inference defaults to a smaller window selectable with a context flag.

Are the accuracy and calibration numbers reproducible?

The repo reports accuracy with expected calibration error on a fixed evaluation set, plus consistency suites for order invariance, padding, and distractor swaps.

How do I put an image into the state?

Reference it by absolute path inside the state string, wrapped in the vision start, image pad, and vision end markers. The processor loads the image, expands it into visual patch tokens, and places them in the shared state branch of the tree mask, and multiple image markers can appear in one state.

What are the architecture and training objective?

State, each question, and each candidate all attend through TreePacked Attention, with a rank-512 readout converting the encoded context and candidate pair into one logit. Training combines a Plackett-Luce ranking objective with a Brier calibration loss, and each candidate attends to the state, its own question, and its own tokens.

How large a state can it handle?

A 262,144-token native context, extendable to 1,010,000 tokens with YaRN scaling. Training truncates state inputs at 16,384 tokens, so inference defaults to that window unless you pass a larger one with --context-tokens, and serving at a million tokens uses a YaRN configuration.

CommunityJev Demos & Experiments

Mojev

Local typed decision model that takes text or image state plus a runtime schema and returns probability distributions in one forward pass.

See highlights

Open github.com

Stars: 29
Language: Python
License: MIT
Published by: Community
Author: MoLeMo-Lab
Last verified: Oct 2026

MoLeMo-Lab/mojev

Organization repository on GitHub

View on GitHub

MoJev: typed, calibrated decisions in one forward pass.

Forks: 3
Last pushed: Updated Sep 2026

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

About this project

Typed, calibrated decisions in one forward pass. MoJev takes text or image state and a runtime-defined schema, then returns a probability distribution over the permitted values. It runs locally and is wire-compatible with the TypeSafe Python SDK's systemone API. Try MoJev on Hugging Face Spaces — text, images and custom candidates, scored on server-side ZeroGPU. No model download is needed in the browser. Upload one or multiple images alongside your context. Space source · Optional local browser runtime.

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

Tagscommunity jevbest catalog-import

Works forJev for Classification

Used inJev for Software & DevTools

Languages: Python

Highlights

Runtime-defined schemas with typed Choice, Noul, and Score questions answered together in one pass.
Text and image state support, with images referenced by path and expanded into visual tokens in shared state.
SDK-compatible local HTTP server plus command-line serving, metrics, consistency, and API verification commands.
Published checkpoint, training mixture, preprint, project page, and documented repository layout.

Watch out

Local checkpoint download and Python serving setup required; large-context serving needs explicit window flags. 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

open-jev-typed-decision-engine

Open 150M encoder that reproduces the typed noul, choice, and score interface in one forward pass, trainable on a free GPU and runnable locally for nothing.

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

Community

intikhab49/open-jev-typed-decision-enginePython

TypeLLM

941

Library that adds schema-guaranteed typed outputs to existing autoregressive LLMs without changing weights, served through SGLang or a hosted API.

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

Community

TypeLLM/TypeLLMPython

solar-mini4-jev

Drop-in service that exposes Upstage Solar models through the System One shape, accepting state plus noul, choice, or score questions.

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

Community

hunkim/solar-mini4-jevPython

open-medical-jev

Research recipe that fuses two frozen open readers into Jev-style yes/no judgments with a fit-free router, confidence gate, and candidate sets.

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

Community

FeiLiuEM/open-medical-jevPython

jev-robot-control

Simulation study that drives one xArm7 apple-to-plate task with Jev, a large reasoning model, and a small model through identical incremental controls.

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

Community

openroboto-ai/jev-robot-controlPython

OpenJev-Vision

Open research toolkit that encodes an image once and answers multiple typed choice, noul, and score questions from a shared distribution, plus a local 0.6B text-decision model.

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

Community

IamBusy/OpenJev-VisionPython

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