The Hypermedia Constraint (formerly known as HATEOAS) is a constraint that is used to provide direction to the user-agent.
By including links in returned representations, the server can remove the burden from the user-agent of determining what actions can be taken based on the current application state and knowing who to interact with in-order to achieve that goal.
As the server has no knowledge of the user-agent’s current state other than what it receives in a request, it is important that the user-agent tries to avoid using state other than the representations returned from the server. This ensures that the available actions provided by the server are based the most complete understanding of the user-agent state as possible.
A user-agent conforming to the Hypermedia constraint acts like a state machine, where state transitions are caused by following links available in the current representation. The returned representation becomes the new state.
The benefits of this approach can be a very lightweight user-agent. It requires very little code to manage state as its actions should be based purely on the received response and the link that retrieved that response. The user agent code becomes declarative and reactive, rather than imperative sequences of GET this then do this and then do that, you simply have the mechanics for following links and many instances of WHEN you receive this THEN do that.
For an example of how this works, you need look no further than your web browser and a web site that doesn’t use Javascript. The browser presents you with options based on links in the HTML. When you follow that link, the browser replaces its current state with the new state retrieved when you followed the link. The back button works (or at least should) because you are retrieving the state from a link in your history. The browser should not care how you got to the page, as the state should be based entirely on the retrieved representation.
This “state management” model can be very limiting, as your current application state is based on a single server response. However, complex applications can be built by using a set of user-agents working together. This is part of what AJAX achieves in that it allows a distinct user-agent to be used to make separate requests and therefore, in effect, manage another state machine. Unfortunately, most of the time people resort back to an RPC style when they start making javascript requests, which is unfortunate considering the natural asynchrony of Javascript.