**Note:** _Any_ web server technology can serve a Universal app as long as it can call Universal's `renderModule()` function.
The principles and decision points discussed here apply to any web server technology.
Universal applications use the Angular `platform-server` package (as opposed to `platform-browser`), which provides
server implementations of the DOM, `XMLHttpRequest`, and other low-level features that don't rely on a browser.
The server ([Node.js Express](https://expressjs.com/) in this guide's example)
passes client requests for application pages to the NgUniversal `ngExpressEngine`. Under the hood, this
calls Universal's `renderModule()` function, while providing caching and other helpful utilities.
The `renderModule()` function takes as inputs a *template* HTML page (usually `index.html`),
an Angular *module* containing components, and a *route* that determines which components to display.
The route comes from the client's request to the server.
Each request results in the appropriate view for the requested route.
The `renderModule()` function renders the view within the `` tag of the template,
creating a finished HTML page for the client.
Finally, the server returns the rendered page to the client.
### Working around the browser APIs
Because a Universal app doesn't execute in the browser, some of the browser APIs and capabilities may be missing on the server.
For example, server-side applications can't reference browser-only global objects such as `window`, `document`, `navigator`, or `location`.
Angular provides some injectable abstractions over these objects, such as [`Location`](api/common/Location)
or [`DOCUMENT`](api/common/DOCUMENT); it may substitute adequately for these APIs.
If Angular doesn't provide it, it's possible to write new abstractions that delegate to the browser APIs while in the browser
and to an alternative implementation while on the server (aka shimming).
Similarly, without mouse or keyboard events, a server-side app can't rely on a user clicking a button to show a component.
The app must determine what to render based solely on the incoming client request.
This is a good argument for making the app [routable](guide/router).
{@a universal-engine}
### Universal template engine
The important bit in the `server.ts` file is the `ngExpressEngine()` function.
The `ngExpressEngine()` function is a wrapper around Universal's `renderModule()` function which turns a client's
requests into server-rendered HTML pages. It accepts an object with the following properties:
* `bootstrap`: The root `NgModule` or `NgModule` factory to use for bootstraping the app when rendering on the server. For the example app, it is `AppServerModule`. It's the bridge between the Universal server-side renderer and the Angular application.
* `extraProviders`: This is optional and lets you specify dependency providers that apply only when rendering the app on the server. You can do this when your app needs information that can only be determined by the currently running server instance.
The `ngExpressEngine()` function returns a `Promise` callback that resolves to the rendered page.
It's up to the engine to decide what to do with that page.
This engine's `Promise` callback returns the rendered page to the web server,
which then forwards it to the client in the HTTP response.
**Note:** These wrappers help hide the complexity of the `renderModule()` function. There are more wrappers
for different backend technologies at the [Universal repository](https://github.com/angular/universal).
### Filtering request URLs
NOTE: The basic behavior described below is handled automatically when using the NgUniversal Express schematic. This
is helpful when trying to understand the underlying behavior or replicate it without using the schematic.
The web server must distinguish _app page requests_ from other kinds of requests.
It's not as simple as intercepting a request to the root address `/`.
The browser could ask for one of the application routes such as `/dashboard`, `/heroes`, or `/detail:12`.
In fact, if the app were only rendered by the server, _every_ app link clicked would arrive at the server
as a navigation URL intended for the router.
Fortunately, application routes have something in common: their URLs lack file extensions.
(Data requests also lack extensions but they're easy to recognize because they always begin with `/api`.)
All static asset requests have a file extension (such as `main.js` or `/node_modules/zone.js/dist/zone.js`).
Because we use routing, we can easily recognize the three types of requests and handle them differently.
1. **Data request**: request URL that begins `/api`.
1. **App navigation**: request URL with no file extension.
1. **Static asset**: all other requests.
A Node.js Express server is a pipeline of middleware that filters and processes requests one after the other.
You configure the Node.js Express server pipeline with calls to `server.get()` like this one for data requests.
**Note:** This sample server doesn't handle data requests.
The tutorial's "in-memory web API" module, a demo and development tool, intercepts all HTTP calls and
simulates the behavior of a remote data server.
In practice, you would remove that module and register your web API middleware on the server here.
The following code filters for request URLs with no extensions and treats them as navigation requests.
### Serving static files safely
A single `server.use()` treats all other URLs as requests for static assets
such as JavaScript, image, and style files.
To ensure that clients can only download the files that they are permitted to see, put all client-facing asset files in
the `/dist` folder and only honor requests for files from the `/dist` folder.
The following Node.js Express code routes all remaining requests to `/dist`, and returns a `404 - NOT FOUND` error if the
file isn't found.
### Using absolute URLs for HTTP (data) requests on the server
The tutorial's `HeroService` and `HeroSearchService` delegate to the Angular `HttpClient` module to fetch application data.
These services send requests to _relative_ URLs such as `api/heroes`.
In a server-side rendered app, HTTP URLs must be _absolute_ (for example, `https://my-server.com/api/heroes`).
This means that the URLs must be somehow converted to absolute when running on the server and be left relative when running in the browser.
If you are using one of the `@nguniversal/*-engine` packages (such as `@nguniversal/express-engine`), this is taken care for you automatically.
You don't need to do anything to make relative URLs work on the server.
If, for some reason, you are not using an `@nguniversal/*-engine` package, you may need to handle it yourself.
The recommended solution is to pass the full request URL to the `options` argument of [renderModule()](api/platform-server/renderModule) or [renderModuleFactory()](api/platform-server/renderModuleFactory) (depending on what you use to render `AppServerModule` on the server).
This option is the least intrusive as it does not require any changes to the app.
Here, "request URL" refers to the URL of the request as a response to which the app is being rendered on the server.
For example, if the client requested `https://my-server.com/dashboard` and you are rendering the app on the server to respond to that request, `options.url` should be set to `https://my-server.com/dashboard`.
Now, on every HTTP request made as part of rendering the app on the server, Angular can correctly resolve the request URL to an absolute URL, using the provided `options.url`.