Crafting Conversational Experiences with ChatGPT GPTs - A Guide to Using API Actions

What is a GPT?

Consider a GPT as a specialized chatbot, developed utilizing the comprehensive capabilities of ChatGPT, tailored for a particular function or target group.

You have the ability to integrate your own documents into GPTs, enabling users to engage with your data through conversation. By incorporating API actions into GPTs, their interactivity and usefulness are further enhanced, as this allows them to interact with your specific data conversationally through APIs.

This article demonstrates this process building a GPT capable of fetching images of dogs via an API, as depicted in the provided screenshots however a similar process can be used to interact conversationally with any API.

This article assumes some familiarity with building basic GPTs in ChatGPT.

How a GPT Chatbot Utilizes APIs to Enhance Conversations

When interacting with a GPT-based chatbot, users may not realize the sophisticated processes happening behind the scenes, especially when the chatbot accesses external information using an API (Application Programming Interface). Here's a breakdown of this process:

1. User Interaction: It all starts with the user sending a message to the chatbot.
2. Message Processing: This message is then internally reformatted by the chatbot's AI system to match the specific format required by the API it intends to use. This step is crucial as different APIs may have varied requirements for how requests should be structured.
3. API Communication: Once the message is appropriately formatted, it's sent as a request to the chosen API's endpoint. An endpoint here refers to a specific URL where the API can receive and process requests.
4. Receiving API Responses: The API processes this request and sends back data. This data is often in a coded form, such as JSON (JavaScript Object Notation), which might not be immediately understandable to a typical user.
5. Translation to Natural Language: The GPT chatbot then takes this coded response and translates it into a natural language answer. This step involves interpreting the data and converting it into a user-friendly response that appears seamless and coherent.
6. User Receives the Answer: Finally, this translated answer is presented to the user. The user receives a clear, understandable response, usually with no indication of the complex API interactions that occurred behind the scenes.

This entire process occurs rapidly, making the user experience smooth and interactive, while the chatbot efficiently bridges the gap between complex data sources and conversational language.

What is truly remarkable in this context is that the AI can execute these steps without needing specific programming for each individual task. Unlike traditional software that requires detailed coding for each function, GPT-based AI systems are designed to inherently understand and process a wide range of data types and sources. This adaptability and capability to handle complex tasks through conversational interaction are hallmarks of advanced AI, reflecting significant strides in the field of artificial intelligence.

Step-by-Step Integration of "Dog Image API" with a GPT

Preliminary Setup

Start with the API: Utilize the Dog Image API that provides various endpoints to retrieve dog images. The documentation at [Dog CEO's Dog API](https://dog.ceo/dog-api/documentation/) is your starting point.

Create a basic GPT for analyzing user uploaded images of dogs.

Enabling GPT Actions: After creating and testing a basic GPT, navigate to the GPT Actions admin area and select "Create a new Action".

Generating the Schema

Schema Creation: Select the "blank template" option in the actions admin area. Delete the placeholder Schema code. When the placeholder code has been deleted you will see a new button appear "Get help from ActionsGPT".

This will open the ActionsGPT GPT which you can use to generate the necessary OpenAPI Schema.

Paste the documentation URL into ActionsGPT and the Schema code will be generated. There may be errors in the code that need to be fixed.

Insert the schema code generated by ChatGPT into this template, as shown in the below screenshot, which details the schema for image retrieval actions.

Troubleshooting Errors: If there are errors, take note of the feedback, and refine the schema with the help of ChatGPT until the code is functional. Paste the error code into ActionsGPT and updated code will be generated. Repeat this process as many times as necessary until there are no more errors.

When the schema has been correctly implemented you will see a list of available actions below the Schema window and there will be no further error messages displayed.

Testing and Validation

Running Tests: With the schema in place, use the 'Test' buttons next to the actions, such as 'getImagesByBreed' or 'getImagesByBreed', to verify functionality.

Schema Corrections: Ensure that the paths match the API's specifications. In the case of the Dog API, ensure that the path uses "breed" rather than "breeds" to align with the API's endpoint requirements.

Action Review and Finalization

Privacy Policy: Before publishing the action, link the API privacy policy URL. This step is not necessary if you save the GPT only for your own personal use.

Completion and Deployment: With all corrections made and the schema validated, finalize the setup and prepare to publish the action, making it available for user interaction.

Example of a Successful Action

As evident in the below screenshot, a successful API interaction has been established. The "GPT Actions API Demo" GPT shows an output where the model has communicated with the Dog Image API and presented an image of Golden Retrievers, accompanied by a charming description. This result is the culmination of the steps outlined, showcasing how the GPT uses the schema to provide contextually relevant data to the user.

Significance of API Actions in GPT

By integrating API actions, we can extend the capabilities of GPTs beyond text generation. Through this step-by-step integration of the Dog Image API into a GPT, we have illustrated how ChatGPT can be transformed into an interactive agent that fetches and displays images in response to user inputs.

Connecting a GPT to Corporate Data

This approach highlights the significant capabilities of API integrations in enhancing AI interactions. It opens up new avenues for applying GPT models, such as linking with your organization's specific data or accessing information within your Microsoft Graph.

https://learn.microsoft.com/en-us/graph/use-the-api

https://developer.microsoft.com/en-us/graph/graph-explorer

Paul Hankin is the author of:

AI Adoption: A Practical Guide for Business

and

AI and Law: Navigating the Future