6 Augmenting Prompts with Kernel Functions and Plugins
This chapter covers
- What is a simple agent?
- Building native and semantic plugins
- Kernel functions composition
- Out-of-the-Box plugins
At the beginning of this chapter, we explained how Robby, the robot car, can take advantage of senses like eyes and legs, by cleverly using native functions. If in the examples from the previous sections we started timidly with one function (semantic or native), now it's time to equip Robby with various tool sets and unleash him into the world. We'll do this via plugins, a construct that allows us to group native or semantic functions by specific domains, such as robot car movement, fire or rain detectors, sensors handling, and more. Semantic Kernel has a comprehensive library for creating and registering plugins, and we will explore the most important of them, building simple agents, but first, let’s see what do we understand by a simple agent.
6.1 What is a Simple Agent?
A Semantic Kernel agent is a software entity performing tasks autonomously or semi-autonomously. It combines AI reasoning (making decisions) with real-world actions executed through plugins (collections of native or semantic functions), and with chat history that acts like a short-term memory. Unlike simple chatbots, agents actively invoke plugins to interact with external systems, retrieve data, or run workflows.