So far, you've likely interacted with AI through a chat window, typing a message and getting one back. The Claude API lets you build applications that do the same thing, but programmatically. Instead of you typing into a website, your code sends the prompts and receives the responses.

This opens up a world of possibilities. You can build tools that summarize articles, write code, analyze customer feedback, or power a chatbot on your own website. The core idea is simple: your application makes a request to a specific web address (an endpoint) provided by Anthropic, and Anthropic's servers send back Claude's response.

This type of architecture is known as a RESTful API. Think of it like ordering food at a restaurant. You (the client application) give your order (the API request) to the waiter, who takes it to the kitchen (the server). The kitchen prepares the food (processes the request) and the waiter brings it back to your table (the API response). You don't need to know how the kitchen works, just how to place your order.

Anthropic offers a family of models through the API, each with different strengths. You might choose a model based on its reasoning ability, speed, or cost. The most powerful models are great for complex tasks, while smaller, faster models are perfect for things like simple chat or content moderation.

The main models you'll encounter are from the Claude 3 and 3.5 families, such as Haiku, Sonnet, and Opus. Haiku is the fastest and most compact, Sonnet offers a balance of intelligence and speed, and Opus is the most powerful model for highly complex tasks. For most use cases, Sonnet is an excellent starting point.

To use these models, you need a way to prove to the API that you're an authorized user. This is done with an API key.

An API key is a unique string of characters that you include with every request you send. Think of it like a secret password for your application. When Anthropic's servers receive a request, they check for a valid API key. If it's there, they process the request. If not, or if it's invalid, they deny it.

Let's look at the structure of a basic API call. You'll send your request as a block of JSON data. This JSON object needs to contain a few key pieces of information: the model you want to use and the messages you want to send.

Let's break that down:

• "model": This tells the API which Claude model to use. Here, we've specified a version of Sonnet 3.5.
• "max_tokens": This sets a limit on the length of the response. A token is roughly a word or a piece of a word.
• "messages": This is an array containing the conversation history. For a simple query, it just has one object.
• "role": This indicates who is speaking. In this case, it's the "user".
• "content": This is the actual prompt you're sending to Claude.

You also need to include your API key in the request headers, which are like a set of instructions that travel with your request. Specifically, you'll add a header called x-api-key.

The response from the API will also be in JSON format, containing Claude's reply.

Notice the role is now "assistant", indicating the response is from Claude. The answer to your question is inside the content array. This request-response cycle is the fundamental building block for any application you create with the Claude API.

Ready to check your understanding?

Now you have a solid grasp of what the Claude API is and how it works at a high level. You understand the client-server relationship, the role of different models, and the basic structure of an API call.