it a good practice to dynamically generate the Graph inside the Configuration?

[emanueleielo]

I’m designing a system using LangGraph where each assistant could have different configurations (e.g., model, temperature, available tools, system prompt, etc.).

Right now, the Configuration in configuration.py only sets the LLM and basic parameters.
However, I was thinking:
would it be a good practice — and aligned with LangGraph best practices — to make the Configuration class also responsible for dynamically generating the corresponding Graph, based on the provided configuration parameters?

This would allow each assistant to have a tailored graph automatically built according to its own needs, making the architecture more modular, scalable, and clean.

Is this recommended?
Would it fit well into LangGraph’s design philosophy, or is there a better separation of concerns that I should consider?

Thanks in advance for any advice!