ChatGPT Plugin Quickstart using C# and ASP.NET Core

This is a quickstart for sample for creating ChatGPT Plugin using GitHub Codespaces, Visual Studio or VS Code, and Azure. The sample includes templates to deploy the plugin to Azure Container Apps using the Azure Developer CLI. To gain access to ChatGPT plugins, join waitlist here!

Overview

The main project provides an API to query a products catalog. The products are loaded from the products.json file in the Data folder.

To test the API, navigate to the Swagger definition at http://localhost:7000/Swagger. There, you can review the available API endpoints and even run a test query. What you see for the spec (or more specifically, what you see when you open the definition at http://localhost:7000/swagger/v1/swagger.json) is what ChatGPT sees and uses to understand the API.

7000 is the port number for the local development server. If you have a different port configured on your machine, be sure to replace any mention of 7000 with the port you are using.

The API and API description are all standard. The ChatGPT plugin-specific part of the project is the ai-plugin.json file located in wwwwroot\.well-known. This is the manifest ChatGPT uses to find out information about your plugin. If you're having issues, you may need to update the port in this file with the port configured on your machine. You can also use the dev tunnels feature in Visual Studio to expose a public endpoint you can use for testing.

Getting started

1. 📤 One-click setup: Open a new Codespace, giving you a fully configured cloud developer environment.
2. 🪄 Make an API: Add routes in Program.cs, done in a few minutes even without ASP.NET Core minimal API experience, thanks to GitHub Copilot.
3. ▶️ Run, one-click again: Use VS or VS Code's built-in Run command and open the forwarded port 8000 in your browser.
4. 💬 Test in ChatGPT: Copy the URL (make sure its public) and paste it in ChatGPT's Develop your own plugin flow.
5. 🔄 Iterate quickly: Codespaces updates the server on each save, and VS Code's debugger lets you dig into the code execution.

Run

Run in Codespaces

1. Click here to open in GitHub Codespaces

   

2. Open Codespaces Ports tab, right click 8000, and make it public.

3. Copy the Codesapces address for port 8000

4. Open Chat GPT and add the plugin with the Codespaces address

5. Run a query for 'hiking boots'

Run in Dev Container

1. Click here to open in Dev Container

   

2. Hit F5 to start the API

3. Open Chat GPT and add the plugin with localhost:8000

4. Run a query for 'hiking boots'

Run Locally

1. Clone the repo to your local machine git clone https://github.com/timheuer/openai-plugin-aspnetcore
2. Open repo in Visual Studio or VS Code
3. Hit F5 to start the API
4. Open Chat GPT and add the plugin with localhost:8000
5. Run a query for 'hiking boots'

Deploy to Azure

NOTE: If you are running locally, then you first need to install the Azure Developer CLI

Deploy with Azure Developer CLI

1. Open a terminal
2. Run azd auth login
3. Run azd up
4. Copy the endpoint printed to the terminal
5. Open Chat GPT and add the plugin with that endpoint
6. Run a query for 'hiking boots'

Deploy with GitHub Actions

NOTE: Due to a security restriction Codespaces in the browser may not work for you. If you see an error when running the following commands, then please open the project with VS Code by choosing the Connect to Codespace option.

1. Fork this repo to your own account
2. Open your fork in Codespaces, Dev Container or Local
3. Open a terminal
4. Run azd auth login
5. Run azd pipeline config
6. Click on the printed actions link. Scroll to the bottom of the logs to find the endpoint.
7. Open Chat GPT and add the plugin with that endpoint
8. Run a query for 'hiking boots'

What's next?

Explore the API in ChatGPT. How well does it understand your intentions and call the API with appropriate parameters? How well does it summarize the response?

Try different phrases. "Where can I find a pair of hiking boots?" and "How expensive are hiking boots?" Learning how well ChatGPT takes cues from the conversation and translates them into API calls can be beneficial for improving the design of your API.

The entire interaction is based on the metadata of the schema that is sent to ChatGPT. One experiment to try is to enhance the descriptions of the API, even adding property-level descriptions to the schema, that explain more explicitly what the properties do. You can even add comments with examples to better train the model on how to call the API.