When should Jev replace LLM reasoning inside an agent loop?

Only the selection bottleneck with distinct observable branches and the right answer included. Keep LLM planning, recovery, and completion outside, and hand back control on novelty, stale candidates, or delayed feedback.

Do the benchmark wins hold across every task?

By task. The paired runs show large call and token drops on WebShop and FrozenLake, a mixed result on Terminal-Bench, and almost no gain on WebArena. Coverage alone is not success.

What hardware do I need to train versus run a released model?

Train a small adapter on the bundled tickets for CPU, or load a released 4B or 27B checkpoint with enough GPU memory. Released weights download on first use, then run behind a local demo server.

How far should I push autonomy once the demo works?

Start D0 LLM-only, then shadow, one-step, routine default, and bounded subgoal only with paired reward and cost evidence. Without that evidence the extra model is overhead.

How do I install and run it locally?

Use Python 3.12 or newer, clone the repository, make a venv and pip install -e .; start a model server, then jevany demo --base-url http://127.0.0.1:8008 --text-only and open http://127.0.0.1:8090 to edit a state or options and see the decision change.

What are the measured demo savings?

On WebShop, actions fall from 9 to 5, LLM calls from 9 to 4, tokens from 38,852 to 14,256 and time from 18.54 to 7.83 seconds. On FrozenLake, LLM decision calls fall from 4 to 1 and tokens from 2,338 to 663. On Terminal-Bench sqlite-db-truncate, tool commands fall from 13 to 7 and time from 187.9 to 144.7 seconds.

What license is it under, and what is the default model?

Apache-2.0. The current default release is JevAny-Qwen3.8-27B, while the 30 archived examples come from an earlier compatible checkpoint, so the demo replays are not all from the default release.

CommunityJev Demos & Experiments

JevAny

Open training and deployment stack for System 1 decision models: one state-plus-question-plus-options API that returns a choice with probabilities for tickets, tools, or robot actions.

See highlights

Open github.com

Stars: 72
Language: Python
License: APACHE-2.0
Published by: Community
Author: weitianxin
Last verified: Oct 2026

weitianxin/JevAny

Organization repository on GitHub

View on GitHub

Open infrastructure for training, evaluating, and deploying System 1 decision models across language and multimodal backbones.

Forks: 10
Last pushed: Updated Oct 2026

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

About this project

🇺🇸 English | 🇨🇳 简体中文 🎮 Results & Demos | ⚡ Quickstart | 💻 Run locally | 🤗 Models | 📊 Benchmarks | 📚 Docs. JevAny is open infra for System 1 decision model training and deployment, covering data preparation, model adaptation and evaluation. Use a released model or train on your own data to route support tickets, select tools, or choose a robot's next action. One API takes the state, question and candidate options, then directly returns a choice and its probabilities.

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

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification

Used inJev for Software & DevTools

Languages: Python

Highlights

Single decision API takes state, question, and candidate options, then returns a choice with probabilities.
Golden rules restrict use to two to four valid branches with distinct outcomes and measured ranking errors.
Paired agent harnesses compare LLM-only against LLM plus Jev at equal reward with traces and timing ratios.
Python 3.12 package with local playground, train-on-own-data path, and released Qwen-based checkpoints.

Watch out

Demos are archived replays from an earlier checkpoint, and paired results vary by task, so read the linked protocol and appendix before generalizing. 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

jevify

Toolkit that makes a model behind an OpenAI-compatible endpoint or running in-process answer typed choice, score, and yes/no questions from next-token probabilities in Jev API shape.

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

Community

fidecastro/jevifyPython

jev-forge

164

End-to-end toolkit that synthesizes decision data, trains and calibrates candidate scorers, evaluates fixed splits, and serves Jev-compatible inference for interactive web decisions.

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

Community

zwliJay/jev-forgePython

stuntd

Self-hosted proxy that records typed decisions, distils them into small heads on a frozen encoder, and answers locally when confident.

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

Community

bladedevoff/stuntdPython

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

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