GitHub · Serving and API

zhangcy122/OpenJev

Reading a claim about order invariance, with the caveat that 33 stars is not yet a community.

The repository's GitHub card, read on 2026-09-27

What it does

Answers the same three question types as Jev — choice, noul, score — from open models, and adds a distillation loop that keeps reasoning traces that worked as permanent fast paths, so a later call of the same shape skips the exploration. It reports sub-35 ms decisions and perfect option-order invariance on its own benchmark.

The repository's own description: “Self-evolving cognitive decision engine & TypeSafe Jev alternative. Deliberative decision flywheel ('explore first, crystallize later' System 2→1) with 100% option-order invariance. Typed probabilistic API (Choice, Noul, Score) for Open LLMs, Laya (ModernBERT), & commercial Jev with calibrated logprobs and adaptive safety guards.”

How it works

The engine reads calibrated logprobs instead of generated text, normalises them so shuffling the options does not change the answer, and can run fully local or sit in front of a commercial Jev deployment as a guard for borderline classifications.

The repository, by the numbers

Stars

Forks

Language

Python

Last push

2026-09-26

Read from the GitHub API on 2026-09-27. Stars and the last push move daily — quote them with the date, the way we do.

What we checked

the README's benchmark table and the queries it is built on
the install line and the licence field GitHub reports
the star count and the last-push date, read from the API
the advertised modes: pure-local, and a guard in front of a hosted model

What we did not check. We did not install it or run the benchmark, and its table — including the row for TypeSafe's live model — is the author's measurement on a 30-query set rather than ours.

Open the repository ↗All builds

The rest of the topic

Laya, the reference →

Start here

Laya, the reference What it is, what we measured running it, and what it is not.
Every Laya guide, by question The whole topic as a contents page: five explainers on the model, two on running it, and the columns.
Laya and Jev, compared The two models on the same questions, including the ones Laya loses.

Columns

Built with Laya 22 you are here Ports, servers, agents and games, with what we checked.
What X is saying 28
What people filmed 16
What Hacker News made of it 11
Docs, weights and reading 15

The topic

Every record, by kind of work The same rows cut a second way, across every column.
What self-hosting costs Hardware, memory, and the failure modes of running it yourself.
What the licence allows Apache-2.0 weights, three checkpoints, and the limits that come with them.
Laya with MLX The Apple-silicon port: what changes, and how MLX, Core ML and MPS differ.

By kind of work

Runtime ports 26 Getting the weights onto someone else's hardware: MLX on a Mac, Core ML on the Neural Engine, ggml in C++, candle in Rust, ONNX in Node.
Serving and API 11 Standing the model up as something else can call, usually behind Jev's own wire format.
Agents and tools 8 Typed decisions inside an agent loop: routing, evals, guardrails, memory, and the plumbing around them.
Browser and computer use 2 Deciding the next click, key or field on a real screen.
Games and demos 10 The model playing something you can watch. Most of this model's traffic arrived through one of these.
Benchmarks and evaluation 30 Numbers somebody actually ran — including the ones that say Laya lost.

What it does

How it works

The repository, by the numbers

What we checked

More like this

receptron/laya

codejunkie99/keel

lkarlslund/laya.cpp