From legacy MLForm

Older examples used a class-shaped API:

const mlForm = new MLForm("/api/predict");
await mlForm.toHTMLElement(schema, container);

Use mountForm instead:

import { mountForm } from "mlform/kit";
import { transport } from "./transport";

const mounted = mountForm(container, {
  transport,
  schema,
});

Migration map:

Legacy	Current
new MLForm(url)	An application-owned Transport plus mountForm(...)
toHTMLElement(...)	mountForm(...)
inputs collection	fields collection
outputs collection	reports collection
field type	field kind
field title	field label
onSubmit callback	hooks.afterSubmit or await mounted.submit()
mlform/extensions strategies	Definitions and presenters packaged through defineMLFormPlugin(...)

Legacy schema:

const oldSchema = {
  inputs: [{ type: "text", title: "Prompt" }],
  outputs: [{ type: "classifier", title: "Prediction" }],
};

Current schema:

const schema = {
  fields: [{ id: "prompt", kind: "text", label: "Prompt", mappedTo: "prompt" }],
  reports: [{ id: "prediction", kind: "classifier", label: "Prediction", mappedTo: "prediction" }],
};

The legacy class posted to a URL itself. Now that request lives in an application-owned transport, and the response must contain report envelopes:

const transport = {
  async submit(request) {
    return callBackend(request.modelValues, { signal: request.signal });
  },
};

There are no built-in HTTP, streaming, retry, cache, tracing, or circuit-breaker layers to migrate. Keep those concerns in the application client that implements submit.