Using AI Tools and Calling an AI API from Python Code

AI platforms and tools: what they are

An AI tool is an app that uses a trained AI model to do a job for you. An AI platform is a bigger service where you can use many such models, test them and, if you are a developer, connect them to your own programs.

• Chat and writing AI: answers questions, drafts and summarises text, explains code.
• Image AI: makes or edits pictures from a description.
• Voice AI: turns speech into text or text into speech.
• Translation AI: changes text between languages.
• Coding assistants: suggest and explain code.

Using AI tools well

The question you type is called a prompt. Good prompts say: who the AI should act as, what job to do, and in which format (for example, "Explain rain to a Class 6 child in 3 short lines").

1. Check the answer. AI can sound sure and still be wrong. Compare with a book or a teacher.
2. Protect private data. Do not paste passwords, ID numbers or other people's private details.
3. Be honest. If AI helped with homework, say so, and learn from the answer, do not just copy it.
4. Pick the right tool. A text AI is not a calculator and an image AI is not a fact-checker.

What is an API? The waiter picture

An API (Application Programming Interface) is a set of rules that lets one program ask another program for something. In the restaurant picture: your program = customer, API = waiter, AI service = kitchen. You do not go into the kitchen; you give the waiter an order in the agreed form, and a dish comes back.

Most AI services offer a web API. You send a request to an address (a URL, also called an endpoint) over the internet and get a response. Data usually travels as JSON, a simple text form like {"prompt": "hello"}.

Calling an API in Python

The common Python library is requests. A call has four parts: Key, URL, Data, Send.

import requests
key = "YOUR_SECRET_KEY"
url = "https://api.example.com/v1/chat"   # example address, not real
data = {"prompt": "Explain rain in one line"}
reply = requests.post(url, json=data, headers={"Authorization": "Bearer " + key})
print(reply.json()["answer"])

• requests.post sends data (POST). requests.get only asks for data (GET).
• headers carry extra information such as the key.
• reply.status_code tells if it worked: 200 OK, 401 wrong or missing key, 429 too many requests, 500 problem at the service.
• reply.json() turns the JSON text into a Python dictionary, so ["answer"] picks one value.

Real services use their own address and field names; read their documentation and copy the pattern, not the example above.

Keys, errors, cost and limits

Keep the key safe. Never write it in code you share or upload. Store it in an environment variable and read it with os.environ["AI_KEY"].

Handle errors. Networks fail. Use a timeout and check the status:

try:
    reply = requests.post(url, json=data, headers=h, timeout=10)
    reply.raise_for_status()      # stops on 401, 429, 500...
    print(reply.json()["answer"])
except requests.RequestException as e:
    print("Could not reach the AI service:", e)

Cost and limits. Many services charge per request or per amount of text (tokens) and limit how many requests you can send per minute. Test with short prompts and do not call in an endless loop.

Try it

In the 3D free-play step send each of the three requests, then press "Try a wrong key". Predict the status code before you read it. Then write on paper the six lines of code from memory and mark K, U, D, S, R next to the right lines.

At home with a grown-up: try one real AI tool. Write a vague prompt first, then a clear one with who, what and format, and compare the answers.

Key formulas and definitions

• API call = Key + URL + Data + Send, then Read the reply (K-U-D-S-R)
• requests.post(url, json=data, headers=...) sends; reply.json() reads
• Status codes: 200 OK, 401 wrong key, 429 too many requests, 500 service problem
• Good prompt = who + what + format

Worked examples

Common mistakes

• Pasting the secret key into code that is shared or uploaded online. Use an environment variable.
• Trusting every AI answer without checking it.
• Using requests.get when the service needs requests.post with JSON data.
• Not handling errors: no timeout, no status check, and endless loops that hit the limit.