appendix B OpenAI GPT Actions
GPT Actions (previously ChatGPT plugins) let AI models work with external APIs to access real-world data and services. Although Actions can make AI applications more capable, they’re currently limited to OpenAI’s ecosystem, which is why this content is in this appendix instead of chapter 13.
Actions connect GPTs to outside systems through APIs. You can link GPTs to databases and other services using OpenAPI specifications that define how to access external resources.
In chapter 13, we cover function calling, which differs significantly from Actions. Function calling formats AI outputs so your code can process them easily. GPT Actions do something else entirely: they give AI direct access to external systems and data. You’ll need OpenAPI specs for Actions, whereas function calling just needs structured output rules. Use Actions when you want AI to work with live data, and pick function calling when you need to process AI responses in your code.
Here’s how GPT Actions work:
- Set up an API.
- Make a Custom GPT.
- Include your OpenAPI spec with authentication.
- Let users interact with the API using natural language.
Our tests showed that Actions work well for specific tasks, especially exploring APIs. We built an Action to examine the OpenAI API, which showed available models through simple questions instead of code. It works like an interactive guide that helps developers understand an API without reading documentation or writing test code.