This project Shows how to document a REST API using OpenAPI so it is friendly and ready to be understood and directly controlled by an AI model.
In order to:
- Use domain language to execute commands directly on the API
- To create Domain related protocols
- To translate those protocols into Compliant code or scripting language that works.
This project intent is to show best practices in documenting an API, its entities, the protocol usage (REST), using standards (OpenAPI). In order to enable an AI to understand and communicate most precisely, accurately, with a user. IMPORTANT: the AI will make decisions and will call directly functions on the service API.
The Solution is made of 4 C# .net 9.0 projects
- LightControllerAPI: The REST web service (Minimal API endpoints with OpenAPI generated definintion) Is the service we want the AI to control
- AIKernelClient: The controller application talking to the AI: a cross platform MAUI (Multi platform) application that controls the service through prompts Using Semantic Kernel with a connector to OpenAI (Chat GPT 4o Mini) it relays service calls from the AI.
- Plugin creation and propmting occurs in the class Services/SemanticKernelService.cs injected to the MainPageViewModel.cs
- SemanticKernelService.cs contains the hardcoded credentials to OpenAI (not an elegant solution but easy to replace with yours to test with)
- MainPageViewModel will initialize the SemanticKernelService (InitializeKernelAndPluginAsync) and will call it to prompt the AI (GetResponseAsync)
- Speech to Text is also implemented in the MainPageViewModel.cs class, by dependency injection.
- WPFPassiveClient: A visualization application that displays the changes done by the controler on the service by polling at regular intervals the changes made to the service resources (lights) (WPF application).
- LightsAPICommon: A shared Entities library (with seeded data) is only shared with AIKernelClient for it contains prompts. None of the entities are used in the client
