OpenJev (DiffusionGemma)
Jev- and OpenAI-Decisions-compatible System One server on DiffusionGemma 26B-A4B, also serving Laya, Clef and other open models.
Overview
OpenJev (razorback16) is an independent open-source System One decision server. You send a state and yes/no, choice or score questions and get a probability and confidence per answer in tens of milliseconds, read directly from model probabilities with no text parsing, so answers cannot go off-schema. Questions can also be about images. It speaks Jev's wire API, so TypeSafe SDKs work unchanged, and also OpenAI's Decisions API, so client.decisions.create works too.
The default model is DiffusionGemma 26B-A4B (Apache-2.0) read as a diffusion canvas, via vLLM on NVIDIA GPUs or MLX on Apple Silicon. It also serves other open models: Laya, Verdict, CLM, JevK5, Clef and Clef-flash, each with its own input limits and hardware. A free hosted instance is offered on Codiv. Per the list summary, the documented vLLM setup requires custom patches. It is for teams wanting a self-hosted endpoint for several open decision models.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Yes/no question about text or image state | yes · no |
| Choice | Pick one of up to 255 options | — |
| Score | Rate on ordered levels | — |
How it works
- Deploy the server with vLLM (NVIDIA) or MLX (Apple Silicon).
- Choose a model id such as openjev-latest or laya-1.0.
- Call it with the TypeSafe or OpenAI SDK.
- Use probabilities and confidence in your app.
Strengths
- Compatible with both Jev and OpenAI Decisions APIs.
- Multiple open models behind one endpoint.
- Image questions supported.
Caveats
- vLLM setup needs custom patches per the list summary.
- Large models need high-end GPUs.
Getting started
Use the free hosted endpoint https://api.codiv.ai/v1/systemone, or deploy the server locally with vLLM or MLX as described in the README.At a glance
- Type
- Library
- Status
- beta
- Language
- Python
- Runtime
- vLLM, MLX, HTTP server
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
An independent System One decision server compatible with Jev’s API, running an open DiffusionGemma model.
logicrw/awesome-jev-projectsOpen, Jev-compatible System One decision server on DiffusionGemma.
wh000wh000/awesome-jev-liveIndependent Jev-compatible server over DiffusionGemma/vLLM; the documented setup requires custom vLLM patches.
cobanov/awesome-jevOpen, Jev-compatible System One decision server on DiffusionGemma.
hellogumbo/awesome-jevJev-compatible decision server on DiffusionGemma 26B-A4B through vLLM, including questions about images. TypeSafe's SDKs work against it unchanged.
MrJev/awesome-jev
Frequently asked questions
Which SDKs work with it?
TypeSafe's SDKs and OpenAI's client.decisions.create.
Which models can it serve?
DiffusionGemma 26B-A4B by default, plus Laya, Verdict, CLM, JevK5, Clef and Clef-flash.
Is there a hosted version?
Yes, on Codiv with a free token allowance.
Related projects
Go harness for System One models: hosted Jev, open Kev in-process via llama.cpp, and Cloudflare Clef, with typed helpers and a CLI.
ServicebetaGoTypeScript library for one-line local decisions with open models (kev-0.6b, kev-4b, open-jev), no API keys or cloud.
ProjectbetaTypeScriptSwift 6 bridge that runs System One decision models behind Apple's Foundation Models API, on device with Laya or via hosted endpoints.
ProjectbetaSwiftZero-dependency npm package that runs Laya natively or as WebAssembly in Node, Bun and browsers, drop-in for Jev's /v1/systemone.
ProjectbetaJavaScriptLocal-first desktop AI workbench for llama.cpp, vLLM and SGLang that includes a JEV typed-decision page and /v1/systemone gateway.
ServicebetaRuns the open LAYA decision model under Bun with native ONNX inference behind a TypeSafe-compatible API, plus an English/Persian playground.
LibrarybetaTypeScript