# The OpenAI API Platform

**Learning Objectives:**
- Navigate the OpenAI API platform and create API keys
- Set spend limits to control costs
- Track usage across models and projects

---

## Platform Overview

The OpenAI API platform at [platform.openai.com](https://platform.openai.com) is your control center for building with OpenAI models. It's separate from ChatGPT — the platform is where you manage API keys, monitor usage, set budgets, and test models programmatically.

![OpenAI API Platform Dashboard](assets/openai_api_platform.png)

*The OpenAI API platform home page. The dashboard shows token usage, request counts, your monthly budget, and credit balance. The left sidebar organizes tools into **Create** (Chat, Agent Builder, Images) and **Manage** (Usage, API keys, Batches).*

Key areas in the sidebar:

| Section | What It Does |
|---------|-------------|
| **Chat** | Test prompts interactively (like ChatGPT but with API parameters) |
| **API keys** | Create and manage your secret keys |
| **Usage** | Monitor spending, tokens, and requests |
| **Batches** | Track asynchronous batch processing jobs |

---

## Getting Started: API Keys

To make API calls, you need an API key. Go to **API keys** in the left sidebar and click **Create API key**.

**Important security rules:**
- Never commit API keys to git. Add `.env` to your `.gitignore`.
- Store keys in environment variables or a `.env` file.
- If a key is leaked, revoke it immediately from the platform.

The code examples in this course use `python-decouple` to load keys from a `.env` file at the repository root:

**Full code**: [02_openai_hello on GitHub](https://github.com/finm-33200/ai_inclass_examples/tree/main/basic_llm_api/02_openai_hello)

```python
from decouple import Config, RepositoryEnv
from openai import OpenAI

# Load API key from .env file
config = Config(RepositoryEnv(".env"))
client = OpenAI(api_key=config("OPENAI_API_KEY"))
```

Your `.env` file should look like:
```
OPENAI_API_KEY=sk-proj-...
```

---

## Setting a Spend Limit

Before writing any code, set a monthly budget. In the dashboard screenshot above, notice the **April budget: $13.77 / $450**. This is a hard cap — once you hit it, API calls will fail rather than silently running up a bill.

To set your budget:
1. Go to **Settings** (gear icon, top right)
2. Navigate to **Billing > Budgets**
3. Set a monthly limit

**Why this matters:** A single misplaced `while True` loop calling GPT-4o can burn through credits fast. A $10-50 monthly budget is plenty for coursework. You can always increase it later.

---

## Tracking Usage

The **Usage** page gives you a detailed breakdown of your API spending.

![OpenAI API Platform Usage Page](assets/openai_api_platform_usage.png)

*The Usage page shows total spend, token counts, and request volume over time. You can filter by date range, project, and API capability. The breakdown by **API capabilities** (Responses and Chat Completions, Images, etc.) helps identify where your budget is going.*

Key metrics to monitor:
- **Total Spend**: Your dollar cost for the selected period
- **Total Tokens**: Combined input + output tokens across all requests
- **Total Requests**: Number of API calls made
- **Spend by capability**: See if costs come from text generation, images, or other services

### Model Pricing

To estimate cost, multiply tokens by the model's pricing:

| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|-------|----------------------|----------------------|
| GPT-4o-mini | $0.15 | $0.60 |
| GPT-4o | $2.50 | $10.00 |
| GPT-4.1 | $2.00 | $8.00 |
| GPT-4.1-mini | $0.40 | $1.60 |

```{tip}
Start with `gpt-4o-mini` for development and testing. It's fast and cheap. Only move to more expensive models once you've validated your approach works.
```

---

## Key Takeaways

1. **Set a budget first.** Before writing any API code, configure a monthly spend limit on the platform. This protects you from runaway costs.

2. **Track usage.** Use the platform's Usage page to monitor spending by model, project, and date range.

3. **Start cheap, scale up.** Use `gpt-4o-mini` during development. Move to larger models only after validating your approach.