provide more than buildtime chosen models?

[maxandersen]

Currently model name is build time only.

Makes perfect sense for enabling devmode fetching of the model and even possibly buildtime include the model (for jllama usecase) and definitely makes sense to keep.

But it excludes ability to i.e. use ollama or openai compatible endpoint and swap to a newly available model.

usecases where this is useful, is i.e. i'm writing a ai-agent using granite3-dense:8b but now a granite3-dense-much-better:8b becomes available; or just lets say I finetune a model for my usecase and make it available under a different name - today I'll have to rebuild my ai-agent based on quarkus-langchain4j to make use of that better model - even though nothing else changed.

I can use langchain4j directly and make this work but within quarkus langchain4j I didn't spot a way to do this without having to give up on all the other declarative goodness.

[geoand]

Definitely something to think about here...