@mailwoman/neural-web

Deprecated. The browser runtime moved into @mailwoman/neural. This package is a re-export shim so existing imports keep working; it will be removed a major later.

Migrating

Install @mailwoman/neural and onnxruntime-web, then change the specifier:

-import { loadNeuralClassifierFromURLs } from "@mailwoman/neural-web"
+import { loadNeuralClassifierFromURLs } from "@mailwoman/neural/web-loader"

WebONNXRunner and its options live at @mailwoman/neural/web-onnx-runner; the tokenizer, classifier and soft-feature channels at @mailwoman/neural/browser. Every export keeps its identity, so a mixed-specifier tree behaves the same during the move.

Nothing else changes:

import { loadNeuralClassifierFromURLs } from "@mailwoman/neural/web-loader"

const { classifier } = await loadNeuralClassifierFromURLs({
	modelURL: "/static/mailwoman/model.onnx",
	tokenizerURL: "/static/mailwoman/tokenizer.model",
	runner: {
		// Optional. Point this at wherever you serve onnxruntime-web's .wasm assets if your
		// bundler doesn't put them in the default location.
		wasmPathsRoot: "/static/ort/",
	},
})

const tree = await classifier.parse("123 Main St, Springfield, IL 62704")

Install the runtime yourself

onnxruntime-web is an optional peer dependency of @mailwoman/neural, not a transitive one. It no longer arrives on its own — install it alongside:

npm install @mailwoman/neural onnxruntime-web

That is the point of the move rather than an inconvenience of it. A Node service installs onnxruntime-node, a browser app installs onnxruntime-web, and neither pays for the other. Whichever you bundle, you chose it.

Why the packages merged

@mailwoman/neural used to depend on onnxruntime-node, which ships native binaries a browser bundler cannot parse — so the browser runtime lived over here, and the boundary between the two was held by hand: a re-export list, a lint rule, a webpackIgnore comment. Each held until someone added an import.

The boundary is a resolution contract now. @mailwoman/neural/onnx-runner carries a browser export condition, so a bundler resolves the Node runner to a counterpart instead of following it, and both runtimes can live in one package.

WebONNXRunner still implements the same structural NeuralRunner interface as the Node runner — that part was never the problem.

License

AGPL-3.0-only.