What is OpenJevSwift?

A native Swift implementation of OpenJev, the open, Jev-compatible System One decision server. Ask a model typed questions about a piece of text — yes/no, which of several options, or how much on a scale — and get probabilities back instead of generated text.

Which models can it serve?

On a Mac, one openjev binary serves DiffusionGemma 26B-A4B or JevK5 on MLX, or Verdict or Laya on Core ML. The same libraries answer requests inside an iPhone or Mac app.

How do I start the server?

On an Apple silicon Mac with macOS 15 or later, download the Metal Toolchain with xcodebuild -downloadComponent MetalToolchain, clone the repo, run swift build -c release --product openjev, then .build/release/openjev serve --backend verdict.

Does it need an API key?

No. The model runs locally: the first start downloads Verdict's converted Core ML package, tokenizer and calibrator (about 310 MB) into Application Support, and each file's SHA-256 is checked. The server then listens on 127.0.0.1:8080.

Why read answers from probabilities instead of text?

Because an answer cannot go off-schema: a noul's answer is the probability of yes, and a choice's or a score's covers every option with a confidence. Nothing is generated, so there is no text to parse.

Can I run a single decision without a server?

Yes. openjev decide answers one request without starting a server and prints the bytes the server would send; you pipe the JSON request in on stdin. Ctrl-C stops the server gracefully.

Is it affiliated with TypeSafe AI?

No. The README states OpenJevSwift is an independent project, not affiliated with or endorsed by TypeSafe AI, Google DeepMind, NVIDIA, or the other models' authors. It is Apache-2.0 and keeps upstream's exact wire API.

Jev directory

CommunityJev Repos & SDKs

OpenJevSwift

A native Swift implementation of OpenJev, the Jev-compatible typed-decision server: send a state and typed questions over a Jev-compatible HTTP API and read every answer from model probabilities, on Apple silicon Macs or inside iPhone and Mac apps.

See highlights Open github.com

Stars: 10
Language: Swift
License: APACHE-2.0
Reading time: 2 min
Published by: Community
Author: algorythm-canada
Last verified: Oct 2026

algorythm-canada/openjevswift

Organization repository on GitHub

View on GitHub

Native Swift implementation of OpenJev, the Jev-compatible typed-decision server: yes/no, choice and score answers read from model probabilities, on Apple silicon Macs and in iPhone apps. Serves Verdict, Laya, DiffusionGemma and JevK5. Apache-2.0.

Forks: 2
Last pushed: Updated Oct 2026

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

Tagscommunity catalog-import cobanov yibie swift on-device coreml

Works forJev for Classification Jev for Scoring Jev for Decision gating

Used inJev for Software & DevTools

Languages: Swift
Platforms: macOS API CLI
Capabilities: Open sourceRuns locallySelf-hosted

Highlights

One openjev binary serves Verdict or Laya on Core ML, or DiffusionGemma 26B-A4B and JevK5 on MLX.
A noul answer is the probability of yes; choice and score answers carry a probability per option plus a confidence.
First start downloads Verdict's converted Core ML package, tokenizer and calibrator (about 310 MB), checking each SHA-256.
Upstream's exact wire API means TypeSafe's SDKs work against it unchanged; curl localhost:8080/v1/systemone answers requests.
Needs an Apple silicon Mac, macOS 15+, Xcode 27 and the Metal Toolchain; Apache-2.0, independent of TypeSafe AI.

Quickstart

bash

xcodebuild -downloadComponent MetalToolchain
git clone https://github.com/Algorythm-Canada/OpenJevSwift.git
cd OpenJevSwift
swift build -c release --product openjev
.build/release/openjev serve --backend verdict

Watch out

Grounded on a truncated README and GitHub metadata captured 2026-10-15; the digest cuts off in the app-embedding section, so the library path is described only in outline. The README itself states the project is independent and not affiliated with TypeSafe AI, Google DeepMind, NVIDIA, or the served models' authors.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

X: A computer assistant that listens

From the Jev guides

Original write-ups on what this entry is for.

The Open Reimplementations: Local Models That Speak System One

Sep 22, 2026

Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.

Jev Repos & SDKs#open-models#research#local

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai