Comment by ipotapov

Regarding the user instruction aggregation process in the agent loop, I'm curious how you manage context retention in multi-turn interactions. Have you explored any techniques for dynamically adjusting the context based on the evolving user requirements?