One of the primary purposes of an API is to describe how computer applications and software developers may access a set of (usually third party) functions (for example, within a library) without requiring access to the source code of the functions or library, or requiring a detailed understanding of the functions' internal workings. The software which provides the functionality described by the API is said to be an implementation of the API. The API itself is abstract, as it is an interface. A reference implementation of an API is the implementation created by the designer of the API, or one which other implementations of the API are expected to be compared against.
For example, an API might describe how an application may call an icon-drawing function within a graphics library, for displaying icons in the screen. A programmer can write a program which calls the icon-drawing function described in the API. When compiled the compiler will link against the API. When executed, the program will use the implementation of the API (a library) to draw the icon.
Computer programs often use the operating system's API to allocate memory and access files. Many types of systems and applications provide and implement APIs, such as graphics systems, databases, networks, web services, and even some computer games.
In many instances, an API is often a part of a Software development kit (SDK). An SDK may include an API as well as other tools and perhaps even some hardware, so the two terms are not strictly interchangeable.
There are various design models for APIs. Interfaces intended for the fastest execution often consist of sets of functions, procedures, variables and data structures. However, other models exist as well, such as the interpreter used to evaluate expressions in ECMAScript/JavaScript or in the abstraction layer, which relieves the programmer from needing to know how the functions of the API relate to the lower levels of abstraction. This makes it possible to redesign or improve the functions within the API without breaking code that relies on it.
Two general lines of policies exist regarding publishing APIs:
Some companies guard their APIs zealously. For example, Sony used to make its official PlayStation 2 API available only to licensed PlayStation developers. This is because Sony wanted to restrict how many people could write a PlayStation 2 game, and wanted to profit from them as much as possible. This is typical of companies who do not profit from the sale of API implementations (in this case, Sony broke-even on the sale of PlayStation 2 consoles and even took a loss on marketing, instead making it up through game royalties created by API licensing). However, PlayStation 3 is based entirely on open and publicly available APIs.
Other companies propagate their APIs freely. For example, Microsoft deliberately makes most of its API information public, so that software will be written for the Windows platform. The sale of the third-party software sells copies of Microsoft Windows. This is typical of companies who profit from the sale of API implementations (in this case, Microsoft Windows, which is sold at a gain for Microsoft).
Some APIs, such as the ones standard to an operating system, are implemented as separate code libraries that are distributed with the operating system. Others require software publishers to integrate the API functionality directly into the application. This forms another distinction in the examples above. Microsoft Windows APIs come with the operating system for anyone to use. Software for embedded systems such as video game consoles generally falls into the application-integrated category. While an official PlayStation API document may be interesting to read, it is of little use without its corresponding implementation, in the form of a separate library or software development kit.
An API that does not require royalties for access and usage is called "open." The APIs provided by Free software (such as all software distributed under the GNU General Public License, for example glibc), are open by definition, since anyone can look into the source of the software and figure out the API. Although usually authoritative "reference implementations" exist for an API (such as Microsoft Windows for the Win32 API), there's nothing that prevents the creation of additional implementations. For example, most of the Win32 API can be provided under a UNIX system using software called Wine.
It is generally lawful to analyze API implementations in order to produce a compatible one. This technique is called reverse engineering for the purposes of interoperability. However, the legal situation is often ambiguous, so that care and legal counsel should be taken before the reverse engineering is carried out. For example, while APIs usually do not have an obvious legal status, they might include patents that may not be used until the patent holder gives permission