Add an action assistant
to your web app.

What is AI Action SDK?

An SDK is a software development kit: code you add to an application you are building. This SDK provides the assistant interface, an action registry, argument checks, and confirmation cards. You connect those actions to the features and APIs your application already has.

What you can try today: this page uses fixed commands and fictional records. The ZChat and MyLiveChat integrations support real read actions or dashboard navigation. Model inference and production write actions require additional application integration.

See what the assistant is allowed to do.

The application defines this action catalog. Requests to run arbitrary code or unregistered actions are rejected by the SDK.

Choose your starting point.

Try a working local command first, then connect the assistant to your application’s existing features.

<div id="assistant"></div>
<p id="result" role="status">Ready.</p>
<script src="richscripts-actions.js"></script>
<script>
RichScriptsActions.mount({
  target: '#assistant',
  welcome: 'Type "Show status" to try a read action.',
  resolve: async ({ message }) =>
    message.trim().toLowerCase() === 'show status'
      ? { action: 'show_status', arguments: {} }
      : { message: 'Try "Show status".' },
  actions: [{
    name: 'show_status',
    readOnly: true,
    parameters: { type: 'object', properties: {}, required: [] },
    execute: () => {
      // Replace this sample read with an authorized host API call.
      document.querySelector('#result').textContent = 'Demo status: ready.';
      return 'The status is displayed in your application.';
    }
  }]
});
</script>

Connect the complete sample workflow.

The starter ZIP now includes an authenticated ASP.NET page, model resolver, proposal executor and session data store. Configure your own model key to try real language-to-action requests on fictional records.

1. Install server-sample/ inside your existing signed-in ASP.NET application.
2. Merge the example model settings into your protected configuration.
3. Open ConnectedDemo.aspx after signing in.
4. Ask for a change, review it, and confirm. The server consumes the stored proposal and updates only the sample session.

These server examples are downloadable source, not active RichScripts customer endpoints. Model calls incur your provider costs. The public demo above remains local and free of API calls.

Build an assistant that can do useful work.

RichScripts AI Action SDK is a developer component for adding a conversational interface to a web application. Instead of treating every request as an open-ended chat, the application defines a small catalog of actions with named parameters. These actions can search records, change the visible view, prepare a note or request an update to an existing item. The assistant helps users reach the features that already exist in your product.

The preview on this page runs entirely in the browser using fictional support requests and a scripted resolver. It demonstrates the interaction without sending messages to a language model. Reloading the page restores the sample inbox. An integration can keep explicit local commands or supply a server-side resolver backed by the model or local inference service chosen by the developer. The browser communicates with your application endpoint; model credentials remain on the server.

Action schemas describe allowed argument types and values. The prototype validates flat string, integer and boolean fields, rejects unknown fields, and checks required arguments. It does not implement the complete JSON Schema standard. Your backend must repeat validation and use the authenticated user's identity to select records and check permissions. Browser-side checks improve interaction but cannot authorize a database change.

Read-only actions can execute immediately. Actions that modify state show the proposed arguments and ask the user to confirm or cancel. A production executor should bind this approval to the proposed action, enforce it on the server and use idempotency keys for retries. The connected sample binds proposals to the signed-in user, keeps separate pending operations across tabs, supports server cancellation, checks their five-minute expiry and returns cached results for duplicate execution requests. Its session data is fictional; real applications still need database transactions, tenant authorization and durable idempotency. The SDK reports action activity through callbacks so developers can connect their own audit trail.

The initial release uses vanilla JavaScript and an isolated component style, with no framework dependency. It includes a model resolver, a sample executor and a connected ASP.NET page, with fictional session data for installation inside an authenticated application. This is a developer preview intended for evaluation and integration experiments. Pricing and a production support policy will be published after the component has been tested against real application workflows.