API Endpoint

Programs need to be able to send each other data. A weather app on a phone doesn’t know anything about the weather by itself. It asks a computer sitting somewhere out on the internet and gets an answer back. To do that, it needs a fixed address to direct its question to. That’s exactly what an API endpoint is: a clearly defined address behind which a particular service waits for requests. It looks like an internet address, but it isn’t called up by humans in a browser — it’s called up by other programs.

Why entire companies consist of nothing but endpoints

Endpoints are the sales counter of the software industry. A provider like OpenAI or Anthropic doesn’t sell software you download. It provides an address you send text to, and it returns the response from its AI model. Billing happens per request. The model itself never leaves the provider’s data centers.

For developers, this is enormously convenient. A small company can use a language model whose training cost hundreds of millions of dollars. It doesn’t need graphics cards or machine learning specialists for that. All it needs is the address and an access key — essentially a kind of password for programs.

This convenience comes at a price, and that price is dependency. Anyone who builds their app on top of someone else’s endpoint is at that provider’s mercy. The provider can raise prices, swap out the model, or discontinue the service. This is exactly what fuels a lot of debate in the tech industry. Some companies therefore also run their own models on their own servers as a backup.

What actually happens during a request

A request consists of three parts. First, the address, for example something in the form api.provider.com/v1/messages. Second, the method, which states what should happen: retrieving data or sending data. Third, the actual data, usually in a format called JSON. JSON is a simple text notation that humans can still read and programs can easily process.

The server receives the request, checks the access key, and processes it. For an AI endpoint, that means: the model computes and generates a response. This response travels the same route back, likewise as JSON. Along with it comes a status code, a three-digit number. 200 means everything worked, 401 means the key was wrong, 429 means too many requests came in at once.

A common misconception: API and endpoint are not the same thing. The API is the entire interface with all its rules and capabilities. An endpoint is a single address within it. So an API usually has many endpoints — one for logging in, one for querying, one for deleting, for example. You can think of the API as a menu and the endpoint as a dish on it.

Endpoints in apps, price lists, and headlines

You use endpoints constantly without noticing it. When a website shows a Google map, it’s calling an endpoint. When a banking app loads your account balance, likewise. Chatbots on company websites also send every message to an AI provider’s endpoint.

In the news, the term usually comes up around pricing and outages. Headlines like “Provider cuts API prices by 60 percent” refer to the cost per request to such endpoints. If a central endpoint goes down, thousands of apps that rely on it stop working at the same time. Such outages are a regular topic in the tech press.

Anyone learning to program themselves runs into this quickly. Many public endpoints are free, for example for weather data, transit schedules, or Wikipedia content. You can try them out with just a few lines of code. It’s often the first moment your own code talks to the outside world.