@nodd/browser

Run a nodd model in the browser with transformers.js (onnxruntime-web: WASM by default, WebGPU opt-in). Inference runs in a Web Worker; model files are cached with the Cache API, so a model loads offline after the first visit.

npm install @nodd/browser

import { nodd } from "@nodd/browser";

const m = await nodd.load("/models/comment_moderation/v3"); // a `nodd export` folder
const d = await m.decide("Buy cheap followers at ...");
// { label: "spam", probabilities: {...}, confidence: 0.99, source: "micro", ... }
if (!m.isConfident(d)) { /* below the calibrated threshold: escalate */ }
m.dispose(); // stops the worker

Options: device: "auto" | "webgpu" | "wasm" (auto = wasm, measured faster than WebGPU at this size), dtype: "q8" | "fp32", onProgress({ loaded, total }), worker, cache, ortWasmPaths (default /ort/).

Serving:

• The export folder must be served from the page's own origin (never the Hub or a CDN).
• Serve onnxruntime-web/dist/ort-wasm* at ortWasmPaths.
• Optional: Cross-Origin-Opener-Policy: same-origin and Cross-Origin-Embedder-Policy: require-corp enable multi-threaded WASM.

The package ships as plain ESM that uses new Worker(new URL("./worker.js", import.meta.url)), so your bundler (Vite, webpack 5, …) compiles the worker and its dependencies.

On a server, use @nodd/node: same API, native CPU.