Open-Source Jev Alternatives and Reproductions

NandhaKishorM/laya

Non-autoregressive decision engine over 100+ languages: three checkpoints and a router that detects the script and dispatches per request. Its benchmarks end with a limits section naming the datasets it does not generalise to and the headline figure that came from a training split.

jaredpalmer/kev

Jev-style decision models from 0.5B to 8B, built as LoRA adapters on Qwen and served behind a Jev-compatible /v1/systemone API.

TheoLeeCJ/SemIf-OpenJev

Jev-style decisions from a frozen 4B model on a single RTX 3090, with a browser demo. Formerly OpenJev.

Contrastive-LM/CLM

CLM-8B, a contrastively trained System One model served behind a TypeSafe-compatible API, with data, weights and a fine-tuning tutorial.

deepopen-com/deepopen

A router and presets over Convai's Laya checkpoints, packaged as its own engine.

TianyuCodings/NanoJev

0.6B parallel decision model with an end-to-end training pipeline.

bespokelabsai/nimble

Open data, training recipe, and a 9B model for Jev-style choice and true/false decisions on Apple Silicon or NVIDIA GPUs.

StartLuxLabs/StartLux-Decision

Typed decision models from 0.8B to 27B, each answer carrying a probability per option, with a chess duel demo.

firelex/jeff

Qwen3.5 and Gemma 4 fine-tunes for zero-shot classification that take Jev's request shape. The author reports about 22 ms per decision on an RTX PRO 6000 and 28 ms on an M4 Max under MLX.

nokia-applied-research/AnyJev

Turns any open-weights LLM into a typed decider by averaging the option logits over permutations and subtracting a label-free prior, so the answer barely moves when you reorder the options.

ollaya-dev/ollaya

Pulls and serves open decision models locally the way Ollama serves language models.

vinnylarouge/jevlike

Train a small model that scores a changing list of text options in one pass.

Mapika/decider

A family of System One-style models that never generate text: one forward pass over a state and typed questions returns a probability distribution per question. Ships ten text games and a Super Mario Bros agent where each move is one typed decision over the legal actions.

receptron/laya

Runs Laya, an open Jev-compatible System One model, from Node.js and TypeScript through ONNX Runtime.

Rizzo-AI-Academy/rizzo-flow

Local typed decisions from Spark-X2.5-4B over llama.cpp, serving both its own schema and Jev's /v1/systemone. Its /v1/models alias says in its description that it is not answered by Jev, and every published result names its dataset by hash.

feder-cr/jev

Serves /v1/systemone and /v1/models for yes/no questions only — other types are refused — from a GGUF model through llama.cpp on CPU, offline once the file is downloaded.

wfzyx/von

Non-autoregressive System One model with Python and TypeScript clients, published on Hugging Face under Apache 2.0. The author reports sub-25ms inference.

mode-io/vllm-jev

Serves decision models natively on vLLM, answering candidate probabilities over the System One endpoint.

Valen-Team/Valen

Multimodal System One model, text and images and video in and decision probabilities out, with training and evaluation code and weights on Hugging Face.

githubnext/localjev

Local Jev-compatible /v1/systemone server for Bun that asks DiffusionGemma for probabilities, from GitHub Next.

razorback16/openjev

Jev-compatible decision server on DiffusionGemma 26B-A4B through vLLM, including questions about images. TypeSafe's SDKs work against it unchanged.

featherless-ai/simple-jev

Turns open Hugging Face models into a Jev-style classifier endpoint by reading next-token logits, with a public demo API.

taeold/djev-run

Serves DiffusionGemma-Jev behind a compatible API on Cloud Run, with a small game demo on top.

Yinsongxu/LLM2Jev

Adapts a local language model into a Jev-style decision engine, answering runtime-defined Choice, Score and Noul questions through SGLang.

Zefan-Cai/Open-Jev

Open typed-decision models that return probabilities over candidates directly, without generating or parsing an answer, with a project site and reports.

PostHog/jeeves

A Jev-style classifier that reasons before it decides: Qwen3.5-9B with a LoRA and a pointer head, trained with SFT and CISPO, plus a diffusion drafter. From PostHog.

mohit67890/imajev

Small open models at 2B, 4B and 9B that read the photos, records and text a business already has and answer in the options you set, with a probability on each and an explicit "can't tell" that sends the rest to a person.

malevrigns/agent-jev

A 0.6B decision model on a Qwen3 backbone with weights on Hugging Face: state in, a distribution over your options out, nothing decoded.

ekzhang/openjev-sglang

Jev-compatible API server running an open model on SGLang.

hr98w/jev-visual

Multiple typed questions about one image in a single pass, on Qwen3.5-0.8B with MLX on Apple Silicon. States plainly that it explores the pattern and does not claim to reproduce Jev's architecture or training. Chinese and English.

logan-markewich/jeff

Self-hosted implementation of Jev's System One API on the 400M-parameter GLiFormer model. The official SDK works after changing the base URL.

Heman10x-NGU/openJev-verdict-2.0

A 151M non-autoregressive decision model on ModernBERT with calibrated uncertainty and an in-browser WebGPU playground. Its LICENSE is not recognised as the Apache 2.0 its badge claims.

jiwidi/jiwo

Small open decision models fine-tuned from Qwen3.5 (0.8B and 4B, Apache-2.0) with a server that speaks Jev's POST /v1/systemone format: one forward pass, a probability per option.

MCG-NJU/VideoJev

Open 4B and 35B-A3B System One models for video, which answer temporal-grounding and multiple-choice questions by scoring candidates instead of generating text. With training code and a paper; needs an NVIDIA GPU.

LiQiiiii/Jeva

Open 4B System One model with Jev's Choice, Noul and Score question types, plus evaluation and fine-tuning code and a paper. Needs an NVIDIA GPU.

OmniJev/OneJev

Multimodal decision model answering typed questions about screenshots, photos, video and text in one forward pass. English, Chinese and Japanese.

SiliconLabAI/OpenJev

Approximates the System One contract on top of any logprob-capable model: a fixed answer space, each option scored independently, all questions in parallel.

kshetrajna12/reflex

A small open decision model for your own GPU: fixed answer options in, per-option percentages out, with no free text so it cannot answer off the list.

tinnel123666888/OmniJev

Omni-modal decision model from Beijing Zhongguancun Academy and CASIA: typed questions about images, video, screens and robot scenes, with weights on Hugging Face.

zwliJay/jev-forge

End-to-end toolkit for the other direction: synthesise decision data, train a calibrated candidate scorer on it, evaluate it, and serve it behind a Jev-compatible endpoint.

allebee/jevk5

Open-weight decision model that serves the /v1/systemone request shape from your own GPU, with weights on Hugging Face.

SimpleJev/JevAny

Infrastructure for training, evaluating and deploying decision models across language and multimodal backbones, with calibrated LoRA checkpoints on a 27B backbone. English and Chinese.

Shanghua-Gao/RSI-Jev

Typed-decision checkpoints produced by a self-improving loop of agents that register predictions before spending GPU time, published with the code that produced them.

deepanwadhwa/OpenDecision

Open-source semantic decision engine: state, a question in natural language, and answer criteria in; a structured decision out.

1Panel-dev/laya-server

Self-hosted API and web interface for Laya's checkpoints, speaking /v1/systemone behind its own API keys.

BudEcosystem/Bud-Decision-Studio

Desktop app and local server for open Jev-style decision models on macOS, Windows and Linux: pick a model, ask typed questions, read every answer as a chart, or call it over a /v1/systemone-compatible API. No licence file yet.

lyuyiqi/open-jev-fast

Faster inference backend for Open-Jev-27B: fused CUDA kernels, a prefix tree and CUDA graphs. Needs an Open-Jev install and its weights.

akivet/Mica

HTTP server for Mica, a 4B open decision model (Apache-2.0, now v0.5 Preview) run as GGUF through llama.cpp: yes/no, a choice among 2 to 255 options, or a score, in TypeSafe's /v1/systemone shape. The Docker image needs a CUDA GPU.

kouhxp/gutsy

Local decision model behind the Jev request shape: no generation, no network calls, no per-call cost.

slavabars/decision-cascade

Local decision engine with trainable Noul, Choice and Score heads over a text encoder, ECE and Brier calibration, a cheap-to-strong cascade and a POST /v1/systemone server. Documentation in Russian.

bnsd55/jevmlx

Jev-style typed decisions from local MLX models on Apple Silicon.

ikermoel/open-alternative-jev

Typed, calibrated decisions from any open-weights model in one forward pass, as the Python package open-alternative-jev.

HITsz-TMG/JevEmbed

Choice, score and noul decisions read out of an embedding model of your choosing, with a Python API, a CLI and an optional HTTP server.

alanhuangyoo/OpenJev

Open decision models (1.7B to 8B) trained by distillation for general decisions and browser-agent steps, served behind the POST /v1/systemone shape, with training code and results. Apache-2.0.

iapp-technology/openthai-systemone

Thai and English decision model with a slot-softmax head whose request and response shape mirrors the official API, so existing SDK code can point at it.

nace-ai/drex-decision-models

Nace.AI's open decision models, Drex v1.5 on a 9B Qwen-based backbone and Drex DLM on NVIDIA's Efficient-DLM-8B diffusion backbone, each answering POST /v1/systemone from a Python server or forks of llama.cpp and Ollama. Formerly drex-dlm; Drex DLM weights under CC-BY-NC-4.0.

lateos-ai/reflex

GGUF-native Rust and CUDA engine built for cold-start latency, with a system1 command that scores your candidates from a local model.

OmniJev/PlayJev

Multimodal decision model that plays browser games from raw pixels, with weights and a hosted demo.

jwuthri-cloud/SelfJev

Self-hosted 4B decision model on Qwen3.5 with weights on Hugging Face, a server that answers Jev's request format, a Python SDK and fine-tuning tools; its tests run TypeSafe's own SDK against the server.

Mushroom-Systems/lichen

Drop-in replacement serving /v1/systemone from open weights on your own hardware, returning the same choice, noul and score shapes.

alvarobartt/sys1

System One compatible API in Rust for open decision models such as Laya.

Abhinavexists/lev

A LoRA adapter on Qwen3.5-4B that answers typed questions from logits it has already computed, served over /v1/systemone so existing clients work against it, with the harness that measures it.

fidecastro/jevify

Makes a model you already serve answer typed questions in one pass the way Jev does, and measures how well it manages it.

adventists-ai/duplexjev

Open 4B and 32B speech decision models that answer typed questions about raw audio (end of turn, intent, emotion) in one pass, served with vLLM, with Pipecat and LiveKit plugins.

intikhab49/open-jev-typed-decision-engine

A 150M encoder trained to answer typed questions in one pass, with the training notebook written to run on a free GPU.

sabeel111/OpenSourceJev

Research experiment in local System One decisions through llama.cpp logits projection on consumer hardware.

hunkim/solar-mini4-jev

Puts Upstage's Solar Mini4 behind Jev's /v1/systemone shape, bring your own Upstage key. Its benchmark grades against a third-party judge rather than a peer model's answers, and every published figure recomputes from the artifacts committed with it.

mpnikhil/dev-0.4b

A 399M bidirectional encoder with one universal choice head, answering Choice, Noul and Score in a single forward pass. Every README figure reconciles to an evaluation JSON shipped in the repository.

zhangcy122/OpenJev

Asks an Ollama or OpenAI-compatible model to write a likelihood score per candidate, then softmaxes them with a fixed temperature. PolyForm Noncommercial, not an open-source licence.

IamBusy/OpenJev-Vision

Encodes an image once and answers several typed questions from the shared distribution. Ships synthetic scenes, trained readouts, a dataset and reproducible evaluations.

FLock-io/this-that-model

A 1.9B typed-decision model with an arXiv paper behind it: one forward pass, no decoding loop, and a /v1/systemone endpoint.

hotchpotch/bekko-system-one

Small decision models from 17M to 400M for yes/no, choice, score and reranking, with a training toolkit, Python inference and a browser app.

Argos1111/jev_local

A local /v1/systemone server with two backends: an LFM model zero-shot, and a fine-tuned ModernBERT-Ja cross-encoder. Japanese documentation.

PsiACE/dohnuts

Small multimodal models for direct decisions on text, documents and images. Its model card publishes the benchmark it loses and states that its confidence field is not a measured probability of correctness. Weights are CC BY-NC-SA.

manjunathshiva/opendecider

Calibrated decision models from 400M to 80B on CPU or NVIDIA, with weights on Hugging Face and a Colab notebook.

ncchinh/quyet

A family of open decision models that pick one option per question with calibrated probabilities and answer in TypeSafe's /v1/systemone shape. Apache-2.0.

I-You-Dream/jev-from-scratch

Fine-tunes Qwen3.5-4B with QLoRA to answer Choice, Noul and Score questions from candidate likelihoods, with scripts for data, training, calibration and comparison. Chinese; no licence.

hawkymisc/typed-decision-bert

A local server that speaks Jev's /v1/systemone shape from a BERT encoder, with a numbered account of every request it refuses that Jev might accept.

0xBakeer/arbiter

Serves typed-decision models, Laya or your own, behind a Jev-compatible API on an NVIDIA GPU or a Mac.

blockbrain-ai/cygnet-recipe

Typed decisions from a frozen Gemma-4-12B-it by reading one option-letter token over stock vLLM, packaged for JevBench.

yzfly/edgejev

Converts open Jev reproductions such as Laya, Kev and PlayJev to ONNX with INT8 quantisation and serves them offline on a CPU behind a local /v1/systemone endpoint, with no torch at runtime. Chinese.

yoheinakajima/glance

Asks a frozen open vision-language model typed questions about an image and reads the answer out of one forward pass, with a calibration harness around the readout.

jlt-commons/lev

System One decision engine in Clojure on Jolt, answering typed questions over a state with calibrated probabilities.

ankit-aglawe/tinyjev

A small decision model for a laptop that answers the three primitives and is built around knowing when to ask a person.

chengyongru/fastjev

An independently maintained SemIf fork packaged SDK-first, for deploying an open decision model on your own infrastructure.

Xiaooolong/vev

Jev-style decision models that also take images, with open weights on Qwen3.5 for your own GPU.

afshinm/laya-mps

Runs Laya's typed-decision checkpoints on Apple Silicon through Metal Performance Shaders, with a lower-memory mode.

edgelesssys/privatemode-decisions

Gets a choice and a probability per option out of any OpenAI-compatible model in one forward pass, for Edgeless Systems' confidential-computing API or any endpoint you set.

QiqianFu/Qev

Qwen fine-tuned into decision models at 2B, 4B and 9B, with the training data and evaluation code. English and Chinese.

9pings/notjev

Serves the Jev request shape from any OpenAI-compatible endpoint by presenting options as single letters and reading the letter mass out of logprobs.

abhishek085/open-spark-jev

A local decision model for NVIDIA DGX Spark, labelled from policy engines and solvers rather than an LLM judge. Its evaluation protocol records the time its own corpus leaked most of the test set into training.

NotXf1le/choosekit

Scores a finite set of choices against a model you already run and returns a typed decision with a probability distribution, from text or images. Backends for llama.cpp, Ollama and OpenRouter, and a choosekit-mcp package exposing the same choice as one read-only MCP tool.

lawrence3699/jev-style

Small Qwen3.5 decision models at 0.5 and 1.3 GB in 4-bit, with a systemone-compatible local server, agent skills and a Claude Code guard.

gulucaptain/Chinese-Jev

Tools to train, calibrate and evaluate Chinese decision models on a Laya base, with released datasets, a held-out benchmark and open medical-domain weights.

arjun988/Kev

Open System One engine for typed choice, score and noul answers with calibration.

yijunyu/jev-rs

Rust engine that answers the three primitives from any language model in one prefill, reading probabilities rather than generating.

zhulinchng/jevper

Independent implementation of the documented System One wire format over an OpenAI-compatible chat model.

shibing624/train-jev

Training and serving code for Jev-style models on Qwen: LoRA and a decision head distilled from Jev 1.13's output distributions, served on a /v1/systemone-compatible endpoint. Needs a GPU.

Singularity-Compute/super-decider

Open-weights decision model fine-tuned from DiffusionGemma 26B-A4B that writes a short thought only when unsure, with a Docker runtime serving POST /v1/systemone. Self-hosting needs a 48 GB GPU; the vendor also sells a hosted API.

leesk212/JEV-CPU

A CPU port of SemIf that swaps only the model loader and reuses the scoring code unchanged, so you can read a decision out of a small model's option logits on a laptop with no GPU.

nvkudva/ollajev

Pulls System One decision models from Hugging Face and serves them, Ollama style, behind TypeSafe's routes and shapes, so the stock SDK works by changing TYPESAFE_BASE_URL.

aovestdipaperino/laya-rust

Pure-Rust inference for Laya on candle: ModernBERT-large with its RL decision head, no Python at serving time.

pst2154/Nemotron_Jev

Browser lab for asking typed questions of a dense diffusion model and inspecting the distributions, with a disclaimer that it is neither Jev nor a TypeSafe service.

SihanLv/AudioJev-Inference

Python library and HTTP service for AudioJev: give it an audio clip and several typed questions in one request, and get option probabilities, yes/no judgments or ordinal scores back. Needs an NVIDIA GPU with BF16.

SamratDuttaOfficial/WaterSheep

An open-weight model that answers Choice, Score and Noul questions, plus multi-label ones, with a probability for every option. Serves a local POST /v1/systemone endpoint; its README says TypeSafe's Python SDK works against it by changing the base URL. Also runs in the browser through ONNX.

leofmarciano/groot

Open typed-decision model on mmBERT encoders with a long-state memory, served behind POST /v1/systemone, plus a pure Bend implementation of the forward pass.

ali-master/usejev

Runs Laya under Bun with native ONNX inference behind a TypeSafe-compatible API, so the official SDK reaches it with a baseURL. Ships an English and Persian playground.

bradAGI/ruling

Turns any open chat model into a local decision engine: a state and typed questions in, a probability for every option out, no text generated. Not affiliated with TypeSafe.

redwolf2019/laya-rs

Pure-Rust runtime and HTTP server for Laya on ONNX Runtime, CPU-first, defaulting to the multilingual model, with portable builds for Linux, Windows and macOS. Chinese and English docs.

sgyaqing/verdict

Rust serving engine for open System One models (decider-2b, Von, Laya) behind a Jev-compatible /v1/systemone API, on GPU or CPU. The author reports large throughput gains over the models' own servers.

yohanargentina-oss/Foq

Local ternary 8B decision model that answers Choice, Noul and Score questions through its own API and marks low-confidence answers needs_review. Loading the weights needs a llama.cpp fork.

apertixsk/cerno-inference

Inference code for 4B, 9B and 27B Jev-style decision models with FP8 weights on Hugging Face, taking the /v1/systemone request shape through vLLM or llama.cpp. Needs a 24 GB GPU.

DreamBlooms/dohnuts.cpp

Native C++ inference for Dohnuts on the released Qwen3.5-0.8B base, so the same decisions run on a CPU.

FuturixAI-and-Quantum-Works/carve-jev

Open Jev-like model family that reads one-token decisions from Qwen3.8 under vLLM and serves a TypeSafe-shaped API, with one-command setup.

chaitin/Decis

Self-hosted server that answers Jev's /v1/systemone route from open weights (Laya, kev, jeff), one Docker image per engine, behind its own API key.

wonseokjung/connect-tycoon

Gymnasium company simulator for training a decision model with imitation and then PPO, asking a person when unsure; every decision uses Jev's Choice request and answer shape, so Jev itself can be tested in it too. Korean.

Algorythm-Canada/OpenJevSwift

Native Swift implementation of the OpenJev typed-decision server: yes/no, choice and score answers on Apple platforms, with a triage demo app.

getainode/jebadiah

Open System One decision model on Qwen backbones, publishing the trainer, data builders, evaluations and the record of every run, with weights on Hugging Face and a server that answers Jev's /v1/systemone shape.

itsoltech/basal-rs

Rust inference engine and server, on CUDA, Metal and Intel Arc, for Remek Kinas's Bielik-based Basal decision models, serving the System One contract at /v1/systemone alongside Basal's own endpoint. Polish.

aminry/decisio

Self-hosted server that answers typed questions with a probability for every option from one forward pass of a frozen Qwen or Gemma base, and can fit a correction to labelled examples of your own question.

SergeyBazhmin/krasivo-jev

Training data and models for a one-pass decision model: 100 Hugging Face datasets converted to one state, question, options and label format, Qwen3.5-2B LoRA models with a pointer head, held-out benchmarks, and a vLLM plugin with a client that returns SDK-shaped answers. No licence yet.

Sivasuthan9/fovea

Open 4B decision model for images, with an SDK and a local server that answers TypeSafe's System One API, so typed questions can be asked about a picture.

emnlmn/snap

Local, deterministic typed decisions from one forward pass of an open model, serving the Jev request shape.

FeiLiuEM/open-medical-jev

Two frozen off-the-shelf models answer a yes/no judgment per option, and their agreement becomes a confidence, an auto-release gate and a guaranteed candidate set. Nothing is trained. Its figures are measured against Jev 1.13.0 on national medical exams and are author-reported. Chinese and English.