Browser integration and privacy

The supplied browser client runs model compilation and simulation in a Web Worker backed by WebAssembly. Build the assets as described in installation, then bundle or serve the bindings/wasm/ JavaScript module and generated pkg/engine/ files together. Your build tool must support module workers and resolving new URL(..., import.meta.url).

import { createEngineClient } from './bindings/wasm/js/index.mjs';

const engine = createEngineClient();
try {
  const { result } = await engine.run({
    kind: 'run',
    source: modelText,
    request: runRequest, // A pharmflux.run/v0.1 request.
  });
  console.log(result.names, result.times, result.columns);
} finally {
  engine.dispose();
}

modelText may contain a .pfx model or a JSON model document. runRequest uses the same versioned format as the command-line program; the synthetic request is a complete example. The returned columns are typed arrays aligned to times; sides marks each sample's pre- or post-event side. run() returns a request ID and result, and accepts timeoutMs and onStarted options. Call dispose() when the client is no longer needed. The worker accepts model and request payloads of at most one million bytes each.

The public worker handles simulation requests only. The underlying WASM binding also has sensitivity and fit entry points, but this worker does not expose them. An embedding application would need to integrate those bindings explicitly before offering browser-based fitting.

Privacy boundary

For the supplied worker, model source and simulation requests are sent from the page to its local Worker and processed by WASM on the user's device. The worker does not upload them or the resulting trajectories to a PharmFlux service. This allows a browser application to keep a user's model and data local during simulation.

The browser still fetches the JavaScript and WASM assets. The embedding application controls where those assets come from and whether it uploads inputs or results, logs telemetry, or stores anything locally. Review those application behaviors before promising users that their data never leaves their device. Browser execution also does not replace application-level access control or clinical validation.