[dave_meshimize]

Would be interested to know if this architecture facilitates dynamic context injection from external knowledge sources without inflating the payload again.