browser-laya
Runs the open Laya decision model entirely in a browser tab with WebGPU and ONNX Runtime Web: no server, no API key.
Overview
browser-laya runs Laya, the open non-autoregressive System 1 decision model, entirely in a browser tab. You give it a state and typed questions and it scores every option in one forward pass, never generating text. After the roughly 900 MB weights are cached, no request leaves the page; WebGPU is recommended.
The repository contains @wexare/laya-web (tokenizer, sequence builder, ONNX Runtime Web session and decoding), a Vite playground for pasting state and questions, and tools that generate golden fixtures from the reference Python laya package to check correctness. It supports Laya's choice, score and noul questions, returning a calibrated probability of true for noul. The README's example routes a wrong-size shoe exchange to returns at 0.895.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which team should handle this? | returns · shipping · billing |
| Yes / no | Might this customer leave? | yes · no |
How it works
- Load the page and cache the Laya weights
- Enter state and typed questions
- Model runs in-browser via ONNX Runtime Web
- Read probabilities in the page or library
Strengths
- Fully client-side and private
- No API key or server
- Fixture tests against the Python reference
Caveats
- 900 MB first download
- Laya's quality, not Jev's
Getting started
Try https://wexare-ai.github.io/browser-laya/ or install @wexare/laya-web from npm.At a glance
- Type
- Service
- Status
- beta
- Language
- TypeScript
- Runtime
- browser (WebGPU / ONNX Runtime Web)
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Run Laya, the open-source System One decision model, entirely in your browser. Typed questions in, calibrated probabilities out. WebGPU, no server, no API key.
hellogumbo/awesome-jev
Frequently asked questions
Does any data leave the browser?
Not after the weights are cached.
Which model runs?
Laya, an open System 1 decision model.
How big is the download?
About 900 MB the first time.
Related projects
Mangaba AI's local, calibrated typed-decision model for Portuguese text, trained to optimize the Brier score; proprietary since v5.
ProjectproductionPython- blink★ 22
Open one-pass typed-decision model with a dependency-free C99 runtime that also builds to a 66 KB WebAssembly module.
ServicebetaC - Foq★ 11
Local open alternative to Jev using a 1.58-bit ternary 8B model to answer booleans, choices, scores and Pydantic extraction in ~25 ms.
ProjectbetaPython llama.cpp-based C++ server that turns any LLM into a System One decision model, reading answers from next-token logits.
ProjectbetaC++Three open 4B decision models for English, Turkish and German that serve the Jev /v1/systemone API on a CPU in about 3 GB of RAM.
ServicebetaPythonJapanese System One decision model on ModernBERT-ja (314.6M) returning typed choice, score and bool distributions with no generation.
ProjectbetaPython