Organization repository on GitHub
Run Laya, the open-source Jev-compatible System-1 decision model, from Node.js / TypeScript via ONNX Runtime
- Forks
- 79
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
Jev directory
CommunityJev Repos & SDKs
@receptron/laya
A Node and TypeScript package that runs Convai's Laya checkpoint through ONNX Runtime, with the same system_one shape as the Python reference.
- Stars
- 956
- Language
- TypeScript
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- receptron
- Last verified
- Sep 2026
About this project
Run Laya — the open-source, Jev-compatible System 1 decision model by Convai Innovations — from Node.js / TypeScript. Laya does not generate text. You hand it a state (a ticket, an email, a JSON object) and typed questions, and it returns every answer with calibrated probabilities in one forward pass: choice — pick one option, with a probability per option. score — an expected level on an ordered rubric, with the distribution. noul — a calibrated P(true) for a yes/no statement.
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- TypeScript
- Platforms
- API
- Capabilities
- Open sourceRuns locally
Highlights
- Requires Node.js 20 or newer. PyTorch is not needed at runtime.
- ONNX weights are about 1.7 GB and cache under ~/.cache/receptron-laya.
- The README says answers match the Python implementation to four decimal places.
- Three questions in one systemOne call take about 140 ms on a warm Apple-silicon CPU.
- State truncates at 512 tokens on the English checkpoint, and about 20 choice options is the model's own recommendation.
Quickstart
ts
import { Laya } from "@receptron/laya";
const laya = await Laya.load();
const result = await laya.systemOne(
{ subject: "Refund not received" },
{ churn_risk: { type: "noul", instructions: "Is the customer likely to cancel?" } },
);Watch out
MIT for the package. Laya weights are Apache-2.0 from Convai Innovations. This is the Node runtime, separate from the Python Laya training repo.
Questions about this entry
Short answers drawn from the linked project or page.
No. It runs the open Laya checkpoint locally through ONNX Runtime, so there is no API key and no network traffic once the weights are cached. The request and response shape matches the TypeSafe system_one API and the Python reference implementation; running it yourself is the point.
About 1.7 GB of fp32 ONNX weights downloaded from Hugging Face on first use and cached under ~/.cache/receptron-laya, plus roughly 2 GB of RAM for the loaded model and a few hundred MB per batch of questions. The cache location can be overridden with LAYA_CACHE.
No, and it fails loudly rather than trimming. State is truncated to 512 tokens on the English checkpoint, and a choice question's options must fit head_max_len of 192 tokens or systemOne throws. The model's own recommendation is fewer than about 20 options per choice.
Yes. export/export_onnx.py builds the bundle from the Hugging Face checkpoint and prints the max logit difference versus PyTorch, and Laya.load({ modelDir }) takes a local directory. You can also pin a revision for reproducibility or select a multilingual subfolder.
npm install @receptron/laya on Node.js 20 or newer, then await Laya.load(), which downloads about 1.7 GB of fp32 ONNX weights from Hugging Face on first use and caches them under ~/.cache/receptron-laya, overridable with LAYA_CACHE.
modelDir for a local export, repo and subfolder to pick a bundle or the multilingual variant, revision to pin a commit, cacheDir, an HF token, an onProgress callback, executionProviders and sessionOptions.
Every question of one systemOne call is batched into a single run; a call with three questions takes about 140 ms on an Apple-silicon CPU once the model is warm.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
A local Jev-compatible POST /v1/systemone server in TypeScript that maps typed questions onto oMLX classification prompts and normalizes logprob-style answers.
Jev Models & Reimplementations#community#typescript#local
Community
githubnext/localjevTypeScript
githubnext/localjevTypeScriptA TypeScript package for one-line local AI decisions: calibrated choice, noul and score answers from open models such as kev-0.6b and kev-4b, with no API keys and no cloud, and models auto-downloaded on first use.
Jev Repos & SDKs#community#github#typescript
Community
amatelic/jevTypeScript
amatelic/jevTypeScriptHTTP server for the Mica v0.5 4B Preview decision model: GGUF weights run through llama.cpp and answer noul, choice and score questions with calibrated probabilities over the TypeSafe /v1/systemone format.
Jev Repos & SDKs#community#github#python
Community
akivet/MicaPython
akivet/MicaPythonServing code for the Jeff 1.0 family of Jev-style decision models: a state, a typed question and its options go in, and one forward pass returns a probability for each option. Four open models from 4B to 31B parameters are served the same way.
Jev Repos & SDKs#community#github#python
Community
j-geuter/jeff-servePython
j-geuter/jeff-servePythonRun Laya locally on Bun with native ONNX inference and Hugging Face tokenizers: an HTTP server implementing TypeSafe's decision API plus an English/Persian playground with 16 demos, usable through the official @typesafe-ai/sdk with a custom baseURL.
Jev Repos & SDKs#community#catalog-import#logicrw
Community
ali-master/usejevTypeScript
ali-master/usejevTypeScriptA frozen decision server that serves Gemma 4 31B — quantized to FP8 at startup — in TypeSafe's /v1/systemone format, reading the next-token distribution instead of generating tokens. It needs one FP8 NVIDIA GPU with 80 GB and has no authentication.
Jev Repos & SDKs#community#github#python
Community
krishna-gogineni-765/deck31bPython
krishna-gogineni-765/deck31bPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.