Your First API Call 🔑

Which model provider?

Every major provider works the same way: you send text over HTTPS, you get text back, you pay per token. This track uses Anthropic's Claude, because its Python library is clean, its documentation is excellent, and (relevant to this campus) its fastest tools are written in Rust. The shape of everything here transfers directly to OpenAI, Google, Mistral or a local model; only the import line and the model names change.

Step 1: get a key, and understand what it is

Sign up at console.anthropic.com, add a little credit (a few dollars lasts a long time at these token prices), and create an API key. It looks like sk-ant-... and it is, in effect, a password that can spend your money. Treat it exactly like one.

Step 2: keep the key out of your code

The key never, ever goes in a .py file. It goes in an environment variable, loaded from a .env file that git is told to ignore. This is exactly the pattern from Lessons 42 and 52, and here is why it exists.

# .env  (this file is SECRET and gitignored, never committed)
ANTHROPIC_API_KEY=sk-ant-your-real-key-here

# .gitignore  (add this on your very first commit, before you forget)
.env
.venv/
__pycache__/

Then load it. Two ways, both fine:

import os

# The bare way: the variable must already be in your shell's environment
api_key = os.environ.get("ANTHROPIC_API_KEY")
if not api_key:
    raise SystemExit("Set ANTHROPIC_API_KEY. See the README. Never hard-code it.")

print("Key loaded:", api_key[:7] + "..." if api_key else "MISSING")

# The convenient way: python-dotenv reads the .env file for you
from dotenv import load_dotenv      # pip install python-dotenv
import os

load_dotenv()                        # reads .env into the environment
api_key = os.environ["ANTHROPIC_API_KEY"]

Step 3: install the library

python3 -m venv .venv
source .venv/bin/activate        # .venv\Scripts\activate on Windows
python -m pip install anthropic python-dotenv

A virtual environment (Lesson 26), then the official SDK plus the dotenv helper. On your own machine this is two minutes. In the browser it cannot run at all, which is why this whole level has no ▶ buttons: real assistants need the network and a key, and the school keeps both out of your browser on purpose.

Step 4: the first call

import anthropic

# The client reads ANTHROPIC_API_KEY from the environment automatically.
client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-5",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "In one sentence, what is Python good for?"},
    ],
)

# The reply is a list of content blocks. Pull the text out.
for block in response.content:
    if block.type == "text":
        print(block.text)

Take that apart, because every field matters:

Choosing a model

For most of your Jarvis, claude-sonnet-5 is the right default: nearly the capability of Opus at a fraction of the price and latency. Reach for Opus on genuinely hard problems, Haiku for cheap bulk classification. A million tokens is a lot of text, so a few dollars of credit goes further than you would think.

A system prompt: telling it who to be

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=1024,
    system="You are Jarvis, a concise and slightly witty assistant. "
           "Answer in at most two sentences. Never invent facts.",
    messages=[{"role": "user", "content": "Should I bring an umbrella?"}],
)

print(response.content[0].text)

The system parameter sets the assistant's standing instructions: its persona, its rules, its constraints. It is separate from the conversation and applies to every turn. This is where your Jarvis gets its personality and its guardrails, and you will spend real time tuning it.

When it goes wrong

import anthropic

client = anthropic.Anthropic()

try:
    response = client.messages.create(
        model="claude-sonnet-5",
        max_tokens=1024,
        messages=[{"role": "user", "content": "Hello"}],
    )
    print(response.content[0].text)
except anthropic.AuthenticationError:
    print("Your API key is missing or wrong. Check ANTHROPIC_API_KEY.")
except anthropic.RateLimitError:
    print("Too many requests. Wait a moment and retry.")
except anthropic.APIStatusError as err:
    print(f"The API returned an error: {err.status_code}")
except anthropic.APIConnectionError:
    print("Could not reach the API. Check your internet connection.")

The library raises typed exceptions, exactly the sort you handled in Lesson 22. Catch the specific ones you can do something about: a bad key, a rate limit, a network drop. The SDK already retries transient failures a couple of times on its own.

from dotenv import load_dotenv
import anthropic

load_dotenv()
client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=500,
    system="You are a patient teacher. Explain clearly, use one analogy, "
           "and check for the most common misconception at the end.",
    messages=[{"role": "user", "content": "Why is the sky blue?"}],
)

print(response.content[0].text)

import anthropic

client = anthropic.Anthropic(api_key="sk-ant-api03-RealKey123")
response = client.messages.create(
    model="claude-sonnet-5", max_tokens=1024,
    messages=[{"role": "user", "content": input("Ask: ")}])
print(response.content[0].text)