GPT Actions: Connecting to External Services

Everything you’ve built so far — instructions, knowledge files, capabilities — works within ChatGPT’s world. Actions break that boundary. They let your GPT reach out to the internet, call APIs, and interact with external services in real time.

This is how you build GPTs that check live inventory, look up customer records, create calendar events, or send notifications.

🔄 Quick Recall: In the previous lesson, you configured capabilities and conversation flows. Actions go further — instead of using built-in tools, you connect your GPT to any external service with an API.

What Are GPT Actions?

Actions convert natural language into API calls. Here’s the flow:

1. User says: “What’s the current stock price of Apple?”
2. GPT translates: Calls a stock price API with the parameter “AAPL”
3. API responds: Returns {"price": 198.50, "change": "+1.2%"}
4. GPT presents: “Apple (AAPL) is currently trading at $198.50, up 1.2% today.”

The user never sees the API call. They just have a natural conversation, and the GPT handles the technical work behind the scenes.

The OpenAPI Specification

Actions are defined using the OpenAPI specification (formerly called Swagger). It’s a standard format that describes an API’s endpoints, parameters, and responses.

Here’s a simplified example for a weather API:

openapi: 3.1.0
info:
  title: Weather API
  version: 1.0.0
servers:
  - url: https://api.weather.example.com
paths:
  /current:
    get:
      summary: Get current weather for a city
      operationId: getCurrentWeather
      parameters:
        - name: city
          in: query
          required: true
          schema:
            type: string
          description: The city name
      responses:
        '200':
          description: Weather data returned

Don’t worry about writing this from scratch — you can ask ChatGPT itself to generate an OpenAPI schema for any API you describe.

✅ Quick Check: What is an OpenAPI specification? (Answer: A standard format in JSON or YAML that describes an API’s endpoints, parameters, request formats, and response structures — used by GPT Actions to know how to call external services.)

Setting Up an Action

In the GPT Builder, go to Configure → scroll to Actions → click Create new action.

Step 1: Define the schema

Paste your OpenAPI specification. If you don’t have one, describe the API to ChatGPT:

“Generate an OpenAPI 3.1 specification for a weather API. The endpoint is GET https://api.weather.example.com/current with a required query parameter ‘city’ that returns temperature, condition, and humidity.”

Step 2: Configure authentication

Choose from:

• None — For public APIs with no authentication
• API Key — You provide a key that’s sent with every request
• OAuth — For services where each user authenticates individually

Step 3: Add instructions

Tell your GPT when and how to use the action:

When the user asks about weather:
→ Use the getCurrentWeather action with the city name
→ Present the results as: "In [city], it's currently
   [temperature]°F and [condition]. Humidity is [humidity]%."

Authentication Deep Dive

API Key is the most common choice. You get a key from the API provider, paste it into the GPT builder, and every request includes it automatically.

OAuth is for advanced cases where each user needs their own credentials — like accessing their personal Google Calendar or Slack workspace.

✅ Quick Check: When should you use OAuth instead of an API Key for your GPT Action? (Answer: When each user needs to authenticate with their own account — like accessing their personal email, calendar, or workspace — rather than sharing a single API key.)

Practical Action Ideas

When Actions Aren’t the Right Choice

Actions have limitations:

• Latency: API calls take time — responses are slower than knowledge-file lookups
• Reliability: External APIs can go down, rate-limit, or change
• Complexity: Debugging API issues requires some technical knowledge
• Privacy: Data flows through OpenAI’s servers to the API

For static information that doesn’t change often, knowledge files are simpler and more reliable. Use Actions when you specifically need live, real-time data.

Practice Exercise

Even without a custom API, you can practice Actions:

1. Open your GPT in Configure → Actions → Create new action
2. Ask ChatGPT: “Generate an OpenAPI specification for a public joke API at https://official-joke-api.appspot.com/random_joke"
3. Paste the generated schema into the Action editor
4. Set authentication to None
5. Add instructions: “When the user asks for a joke, use the joke action”
6. Test in the Preview panel

This gives you hands-on experience with the Action setup flow using a simple, free API.

Key Takeaways

• Actions convert natural language into API calls, letting GPTs interact with external services
• They use the OpenAPI specification format — ask ChatGPT to generate schemas for you
• Three authentication options: None, API Key, and OAuth
• Use Actions for live data; use knowledge files for static reference material
• Always add instructions telling the GPT when and how to use each action

Up Next

In the next lesson, you’ll learn the systematic approach to testing and debugging your GPT — turning a rough prototype into a polished, reliable tool.