New function call model in the .NET Semantic Kernel

Developed by Sergey Menshykh, a principal software engineer at Microsoft, the new model aims to simplify software developers’ use of AI models and increase interoperability between different AI connectors.

It is characterized by the fact that it has been designed to be connector and service agnostic. This enables developers to work with a uniform interface, regardless of which AI model they use. In practice, this means that the new function call classes can be used not only for Azure OpenAI and OpenAI AI connectors, but also for other future connectors. Microsoft plans to expand this support in future versions of the Semantic Kernel.

One of the key features is the ability to use multiple required functions at the same time. This was implemented from the beginning, while other features such as parallel function calls and concurrent execution of functions were postponed to future versions. The new approach ensures that the system remains adaptable to accommodate new requirements and developments in the AI ​​field.

In addition, developers can work not only through code, but also by configuring prompt templates. A presented configuration template allows to define the desired function in a semantic form, which increases the flexibility and usability of the model.

It is important to note that the current function call model represented by the class ToolCallBehaviorwill no longer be supported from mid-November 2024. Microsoft recommends that developers migrate their existing implementations to the new model as soon as possible to benefit from the new features and improved architecture.

For developers interested in using the new model, detailed information is available in a feature selection behavior article, as well as migration guidance to make the transition as smooth as possible. Microsoft actively encourages exchange with the developer community and offers support through the GitHub discussion channel.