The OpenAI API: Your First Call

The OpenAI API: Your First Call

The chat completions API is the workhorse of the OpenAI platform. You send an array of messages and receive a generated reply. It is stateless: the model does not remember earlier calls, so your application must feed all the context it needs.

Authentication and Setup

Create an API key in the platform dashboard and pass it through the OpenAI client library. Environment variables keep the key out of your code and out of version control. Never commit a key to a repository.

from openai import OpenAI

client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4o",
    temperature=0.2,
    max_tokens=200,
    messages=[
        {"role": "system", "content": "You translate Java errors to plain English."},
        {"role": "user", "content": "NullPointerException on line 14"}
    ]
)
print(response.choices[0].message.content)

Messages and Roles

The three standard roles are system, user, and assistant. System sets the behavior, user carries the request, and assistant supplies previous turns so the conversation has memory. A typical loop appends the latest user message, sends the whole history, and appends the reply.

Costs and Limits

You pay per token for both input and output. Long histories and large outputs are the main cost drivers. Watch the usage dashboard and set budget alerts before building anything serious.

Error Handling

Rate limits, invalid keys, and network failures happen. Catch the exceptions, retry with backoff for transient failures, and never crash just because one call failed. A small wrapper around the client keeps this logic in one place.

Key Points

• Sending messages returns model output; the API is stateless.
• Keep API keys out of code and out of the repo.
• Tokens drive cost; watch usage and set alerts.
• Handle rate limits and retries gracefully.