@desplega.ai/laya

Typed decisions from a laya checkpoint, in Node or Bun. You give it text or a JSON object and questions (choice, noul, score); it answers with probabilities, not generated text. One ONNX encoder pass plus a small head, no generation.

npm install @desplega.ai/laya

import { createAgent, defineQuestions } from "@desplega.ai/laya";

const agent = await createAgent({ checkpoint: "english", modelDir: "./bundles/english/fp32" });
const questions = defineQuestions({
  team: { type: "choice", instructions: "Which team handles this ticket?", criteria: { billing: "charges", technical: "bugs" } },
});
const [result] = await agent.predictBatch(["I was charged twice"], questions);

Node 22+ or Bun 1.4+. Pass a local fp32 ONNX bundle as modelDir (export one from upstream's public checkpoints with tools/export), or omit modelDir and createAgent downloads Desplega's prebuilt bundle from the public Hugging Face repo desplega/laya-onnx, verified against pinned SHA-256s. No token is needed; HF_TOKEN is optional (rate limits, or a private mirror).

The untyped vendored surface is available as @desplega.ai/laya/raw.

Full guide with real outputs: docs/usage.md. Apache-2.0, vendored from upstream laya-ts; see NOTICE.