Microsoft's Chat Copilot now allows users to test plugins via Semantic Kernel, according to a new support page on the company's learning platform. This new feature is expected to enhance the functionality of the AI-based chat tool.
Chat Copilot on Semantic Kernel
Chat Copilot, not to be confused with Microsoft 365 Copilot, is a reference application for building chat experiences that uses the Semantic Kernel platform with an AI agent. The application was designed to demonstrate how various concepts of the platform, including plugins, planners, and AI memories, can be integrated to create a comprehensive conversational experience. Users can create new conversations with an agent and request it to perform tasks using ChatGPT plugins. Chat Copilot allows you to import your own OpenAI plugins and test them in a safe environment.
To use Chat Copilot, users need to run two components locally on their machine: a React web app that provides a user interface for interacting with the Semantic Kernel, and a .NET web service that provides an API for the React web app to interact with the Semantic Kernel. The necessary software for running the app includes Visual Studio Code, Git, .NET 6.0, Node.js, and Yarn.
Customizing Chat Copilot
Chat Copilot can be customized according to the user's needs. Most of the customization is done in the app settings file, located in the webapi folder and named appsettings.json. Users can define which models to use, choose a planner, and change the system prompts. The application has been designed and tested with OpenAI models from either OpenAI or Azure OpenAI. Users can also choose between two different planners: action and sequential. The action planner is the default planner, used for a plan with only a single step, while the sequential planner allows the agent to string together multiple functions.