Feature Request: Load Agent Configuration (Kernel + Plugins) from YAML/JSON

[FabianSchurig]

Hi @moonbox3 and @eavanvalkenburg, Hi Semantic Kernel Team,

First, thank you for the amazing work you're doing with Semantic Kernel! The framework is incredibly powerful and has immense potential for streamlining the development of intelligent systems.

I wanted to propose a feature that would enhance usability and flexibility: the ability to load a complete Agent (Kernel + Plugins) with all its configurations from a YAML or JSON file. This would enable seamless configuration management and simplify setup for different environments or projects. The scope of this feature would include:

1. PromptTemplate: Already supported in Semantic Kernel, this ensures the conversational context is readily available.
2. Kernel Configuration: Including the Kernel's Service (such as ChatCompletion).
3. Plugins: Automatic integration of plugins as specified in the configuration.
4. Execution Settings: Any additional runtime or execution parameters.

Why is this useful?

A standardized configuration approach (e.g., YAML/JSON) would:

• Simplify deployment and scaling, especially in environments with multiple developers or runtime instances.
• Make it easier to reuse and share configurations between projects.
• Improve the integration of Semantic Kernel into larger pipelines or frameworks, such as Promptflow or other orchestration tools.

References and Related Work

I've already explored this idea in the context of Promptflow integration and drafted two tools that could serve as a starting point:

• KernelFactory: Handles Kernel creation with specific configurations.
• PluginManager: Manages plugin integration.

Additionally, the following resources from the Semantic Kernel repository and documentation provide insight into the current implementation:

• Agent Templates Documentation
• KernelPromptTemplate
• ChatCompletionAgent

Contribution

If this idea aligns with your vision for the Semantic Kernel, I'd love to contribute to its development. Please let me know if there is a similar feature planned or if we could at this feature together.

Looking forward to your thoughts and feedback!

Thanks,
Fabian

[eavanvalkenburg]

@markwallace-microsoft is working on this already, he can comment and maybe get some new ideas!

[FabianSchurig]

@eavanvalkenburg Thank you so much for the fast reply.

Hi @markwallace-microsoft, do you have a timeline for this feature and do you plan implementing all three .Net, Python and Java?
I could assist with Python.