All these samples rely on FE to orchestrate everything. We should have a mix on samples where the FE does what a production grade agent will do i.e pass audio back and forth while the major work is handled by Backend including initialisation params, system prompt etc.