Repository navigation
Provide more standardized configuration of the kernel #3459
Dr. Damir Dobric (ddobric)
started this conversation in
General
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Currently, the configuration of the SK is done by using the following approach:
This is all fine for learning and testing purposes. I'm looking for the following configuration:
Explenation
The currently provided approach is very important. The developer must change the code when changing the service and model. Once the application goes through the DEVOPS lifecycle, it should be possible to configure it by configuration only, without changing the code. It might be that SK already contains this, but I simply did;t see it.
If you provide the proposed constructor, the developers can decide how to materialize the kernel's configuration. It can use DI, can load settings from DB, environment variables, etc. This has already been an approach in place fro decades in. NET. Following your code samples shipped through the VS extension, it is can be noticed that 80% of the code of every sample is plumbing the configuration (files, secrets, environment variables, etc.) in a very unusual way.
Bootstrapping the NET application is an established process in enterprises, and the SK should also align with it.
It is not the task of the SK team to provide a new type of application. I see in the documentation very often "AI Apps". This is the wrong direction. The correct thinking would be, "We build .NET application, and the SK is a part of it leveraged through SK-SDK".
We are happy with possible new AI application types; however, that one should not slow us down today.
All reactions