Jev Opensources
Jev Opensources is where the open-source side of this ecosystem is documented: 3 repositories that let you run typed decisions on your own hardware or benchmark them head-to-head against Jev. Each page is written from the repository itself — the numbers, the commands and the licences, with the source linked at the bottom.
Why it belongs next to a Jev case library: the same workload (a bounded question, an answer with a probability, code acting on it) can be served three ways — Jev's hosted API, an open checkpoint on your laptop, or a local runtime that talks to both. The repositories below cover the open path end to end.
Laya-MLX
The MLX port that makes Laya a local runtime: typed decisions in milliseconds on a Mac, with no PyTorch, no Transformers runtime and no cloud API in the loop.
Apple Silicon runtimegithub.com/mizorewww/laya-mlx
Laya (official package)
The open-source decision engine Jev's ideas are compared against: one forward pass per typed question, three checkpoints, and a router that picks the right one per request.
Python · 100+ languagesgithub.com/NandhaKishorM/laya
Laya vs Jev (T-Rex arena)
A side-by-side arena rather than a spec sheet: local Laya and hosted Jev make the same choices in Chrome's T-Rex game, and the run is scored on agreement, latency and crashes.
Head-to-head benchmarkgithub.com/virajbhartiya/laya-vs-jev
What each repository covers
- Laya-MLX — the Apple Silicon runtime: 13.4 ms median for one short English decision, 0 output tokens, no PyTorch or cloud API at inference time.
- Laya (official package) — the decision engine itself: one forward pass per typed question, 33 ms on a T4, three checkpoints and a language router, Apache-2.0.
- Laya vs Jev — the head-to-head arena: local Laya against hosted Jev in Chrome's T-Rex game, with latency charts, crash replays and an explicit note about what the score does and does not measure.
How to read these pages
Each repository page states what it is, the measured numbers with their test conditions, how to run it, and what a Jev-shaped workload gains from it. Where a repository reports its own limits — one machine, assisted scores, untested macOS versions — that is kept in, because a benchmark number without its conditions is marketing.
Keep browsing: all 1173 Jev cases · cases by use case · the Laya guide · Laya vs Jev comparison · awesome Jev list
Last updated: 2026-09-23 · sources & corrections · every card links to its author's original post