p(the jev | what)
GitHub ↗

Keyboard shortcuts

⌘ / Ctrl K
Search documents
D
Toggle theme
L
Switch language
H
Go home
M
Copy article Markdown
Esc
Close dialog

WHAT-THE-JEV / RESOURCES

Resources

Decision models and the ecosystem taking shape around them.

Closed-Source Models

01

Solar Decide (Upstage)

Upstage's System One decision model on Solar Mini 4, with a 512K context, Korean support, and free output tokens.

↗
02

Hanzo Kai

Hanzo's closed decision model returning four typed answer kinds, supporting multi-decision runs and replay, with free output.

↗
03

Span-01 (Respan)

Respan's behavior classifier that returns present, absent, and not_observable probabilities in a single forward pass.

↗
04

Jev (TypeSafe AI)

TypeSafe AI's closed-source System One decision model returning typed answers with calibrated probabilities.

↗
05

OpenAI Decisions API (GPT-6 Luna)

OpenAI's public-beta typed decisions API with a single gpt-6-luna model and text-plus-image states.

↗
06

Liquid d1 (Liquid AI)

Liquid AI's TypeSafe SDK-compatible decision model family with three hosted variants covering text, image, and audio.

↗

Open-Source Models

07

chinese-laya (yanqiangmiffy)

Chinese fine-tune of Laya's multilingual checkpoint, with machine-translated data splits and author-reported test results.

↗
08

StartLux-Decision (StartLux Labs)

Typed decision family from 0.8B to 35B-A3B reading text, JSON, and images, with non-commercial weights.

↗
09

Nimble (Bespoke Labs)

Bespoke Labs' open Jev alternative: a Qwen3.5-9B one-step typed text decision model with public data and recipe.

↗
10

OpenJev (razorback16)

razorback16's open System One decision server; several unrelated projects share the OpenJev name.

↗
11

Lev (Interfaze AI)

Interfaze AI's System One decision model plus the levbench evaluation harness; adoption is early.

↗
12

Julia-1 (Supersonic Labs)

Supersonic Labs' 144.3M-parameter compact decision model that runs on CPU and in the browser.

↗
13

Jebadiah (Frontier Infra)

Frontier Infra's System One-style decision model with a full training, data, and evaluation pipeline.

↗
14

Rune 26B-A4B (Invergent)

Invergent's multimodal decision model reading text and images and returning probabilities in one pass.

↗
15

NeoHorse-Jev-4B (TokenRhythm)

TokenRhythm's prefill-only decision model for agent routing, tool selection, and scoring.

↗
16

Laya (ConvAI Innovations)

Multilingual non-autoregressive decision engine returning typed decisions with calibrated confidence in one forward pass across 100+ languages.

↗
17

Kev (Jared Palmer)

Small Qwen-based decision model family for self-training and local deployment, with three isolated question types and calibrated probabilities.

↗
18

decider (Mapika)

Mapika's Apache-2.0 decision model family spanning 0.8B to 35B variants, with the training recipe published in full.

↗
19

JevK5 (allebee)

Open decision model pairing distilled Qwen3.5 LoRA with option-letter logit readout; ranked first among open-source models on JevBench v1.4.

↗
20

Hopper (hopit-ai)

Single-forward-pass decision server with five models; its 4B adapters are restricted to research and demonstration use.

↗
21

SemIf (TheoLeeCJ, formerly OpenJev)

Zero-training approach reading option probabilities from frozen open models in one forward pass, with MIT-licensed code.

↗
22

Perplexity Decisions API (pplx-decider-v1-27b)

Multimodal decision model fine-tuned from Qwen3.8-27B, with Apache-2.0 weights and a commercial hosted API.

↗
23

Clef / Clef-flash (Cloudflare)

Cloudflare decision models at 27B and 9B with image input, 64K context, and Jev API compatibility.

↗
24

Tev1 (Together AI)

Open Qwen3.5 4B/0.8B LoRA decision weights supporting choice-type only, with a $17 training walkthrough.

↗
25

Prometheus / Prometheus-Eval

KAIST-led open-weight judge model family outputting scores and written judgments, not typed calibrated probabilities.

↗

Use Cases

26

Jev Trader (jarrodwatts)

An experimental Monad trading bot that asks Jev for one buy-or-sell decision per block.

↗
27

QuantDinger (OpenByteInc)

An open-source AI trading OS with Jev as one optional pre-trade decision gate.

↗
28

Hermes Jev Skills (kerpopule)

Eleven Jev-powered agent skills for Hermes, Claude Code, and Codex.

↗
29

jev-ultrafast (browser-use)

Official Browser Use agent making one Jev decision per cycle, with an LLM only for text entry.

↗
30

toolgate (RiskAverseTech)

Calibrated firewall for agent tool calls: seven Jev questions map through thresholds to allow/ask/deny verdicts.

↗
31

Inbox Zero (elie222)

Open-source email assistant using Jev as an optional classification backend for sorting and filtering.

↗
32

JEV Paper Radar (Eliot5566)

Daily radar screening the full arXiv feed with one Noul question per declared interest; author-reported figures.

↗
33

HA-Jev (AboveColin)

Home Assistant integration exposing Jev questions as sensors and actions, with 25 blueprints and a spend dashboard.

↗
34

JevTown (NevaMind-AI)

Pixel-art town simulation engine where Jev picks actions from engine-defined options; LLM handles dialogue. MVP stage.

↗
35

jev-tetris (thelau)

Plays Tetris with Jev under controlled measurement; a 23-line regex baseline beat the model in its own tests.

↗
36

Milvus bootcamp: Search with Jev

Nine official Milvus notebooks using Jev for judgment calls across a RAG pipeline, with pymilvus reranking merged.

↗
37

DeepEval (confident-ai)

Pytest-style LLM evaluation framework whose JevEval metric delegates scoring to Jev for calibrated probabilities.

↗
38

Semantic Router (aurelio-labs) (same-paradigm reference)

Pre-Jev decision layer routing via encoder similarity in tens of milliseconds; currently being rewritten for 1.x.

↗
39

RouteLLM (lm-sys) (same-paradigm reference)

Framework routing traffic between strong and weak models by difficulty; 85% cost savings officially self-reported.

↗

Integrations

40

Microsoft Agent Framework: agent-framework-typesafe

Official alpha package adapting TypeSafe System One models to Microsoft Agent Framework Python.

↗
41

DSPy TypeSafe Integration

Experimental Jev/TypeSafe integration in DSPy 3.4.0 with decision types and the ReAnchor calibrator.

↗
42

Mastra Classifier

Mastra's Classifier primitive for typed classification with decision models.

↗
43

No-Code Integrations (Zapier / Make / n8n)

Official TypeSafe/Jev integrations on the no-code platforms Zapier, Make, and n8n.

↗
44

OpenRouter Jev Hub

OpenRouter's Jev hub: concept guide, tutorial, cookbooks, and an interactive demo on one platform.

↗
45

Pydantic AI TypeSafeModel

Pydantic AI's TypeSafeModel derives Jev questions from output types, with LLM fallback.

↗
46

LiteLLM Jev Support

LiteLLM gateway support: a Jev pass-through endpoint and a relevance-compression guardrail.

↗
47

Vercel AI Gateway & AI SDK

Vercel serves Jev via AI Gateway, with AI SDK 7 evaluate API and eve framework integration.

↗
48

Cloudflare Workers AI: typesafe/jev

Jev served inside Cloudflare Workers AI, callable with a single env.AI.run() call.

↗
49

LangChain langchain-typesafe

LangChain's Jev-as-judge benchmark: five trace classes each judged 100 times on accuracy, variance, cost, latency.

↗
50

Spring AI TypeSafe (Spring AI Community)

Spring AI community integration: TypeSafeClient with Jev judge, guardrail, and evaluator adapters.

↗
51

LEAPERone Decisions API

OpenRouter-compatible gateway passing Jev decision requests through; Chinese docs only.

↗

Tools

52

Ollaya (ollaya-dev)

Local runtime for decision models, Ollama-style; serves TypeSafe's /v1/systemone wire format, written in Rust.

↗
53

Swama (Trans-N-ai)

Local AI runtime for Apple Silicon Macs, serving OpenAI-compatible and SystemOne decision endpoints.

↗
54

llama.cpp

llama-server exposes /v1/systemone since 2026-10-02, with five pre-converted decision-model GGUFs under ggml-org.

↗
55

SGLang

SGLang turns any chat model into a decision model via /v1/decisions and /v1/systemone; probabilities are uncalibrated.

↗
56

Ollama

Ollama 0.35+ serves Jev-style decision models locally on /v1/systemone, starting with nimble and tev1 models.

↗
57

TypeSafe AI Official GitHub (SDKs, adapter, agent skills)

Official TypeSafe AI GitHub organization hosting the Jev SDKs, LLM-provider adapter, agent skills, and evaluation source.

↗
58

jev-mcp (jkudish)

MCP server by jkudish exposing 12 Jev judgment tools to coding agents over four carriers or compatible endpoints.

↗
59

duckdb-jev (colliber)

DuckDB extension by colliber that poses a typed Jev question per row via four scalar SQL functions.

↗
60

jev-reranker (hotchpotch)

Python library by hotchpotch using Jev for RAG candidate reranking and relevance filtering with threshold-friendly scores.

↗
61

building-with-typesafe-jev (aaddrick)

Unofficial agent skill by aaddrick teaching Jev question design; self-reported evaluation gain of 0.65 to 0.96.

↗
62

openrouter-decisions Skill (OpenRouterTeam)

Official OpenRouter skill teaching agents to delegate judgment and computation to decision models via a seven-step method.

↗

Benchmarks

63

classifier-benchmark (jabr)

A head-to-head benchmark for choice, noul, and score classification models, with two hash-locked suites.

↗
64

System One Mosaic Benchmark (S1MB)

A leaderboard comparing Jev and open decision models across more than 100 benchmarks.

↗
65

jev-rerank-bench (anessbelbati)

A 14-dataset study of Jev as a reranker against Cohere, ZeroEntropy, and open models.

↗
66

jev-sec-bench (Gaurav-Gosain)

A blind security benchmark testing Jev on prompt injection and vulnerable-code detection.

↗
67

Jevals.com

Independent leaderboard scoring Jev and six LLMs on identical typed decision questions, with public per-decision probability data.

↗
68

jev-bench (PavelRavich)

Solo independent benchmark comparing Jev with two GPT-6 models on two 500-sample datasets.

↗
69

JevBench (Benchmark Heaven)

Benchmark Heaven's leaderboard for Jev-class decision models; four-axis geometric-mean score, cite with version.

↗
70

Decision Index

Benchmark for typed decision engines with a full reproduction kit and a public-private mixed suite.

↗
71

evals.typesafe.ai

TypeSafe's official workflow-evals site; figures are vendor-reported, workflow code is open-sourced.

↗