The introduction of edge intelligence technology to offload partial power services to the distribution station side serves as an effective technical pathway for distributed smart grid construction at the medium- and low-voltage distribution network level. This paper proposes a power service-oriented distribution and consumption edge intelligence agent, along with a three-phase autonomous optimization method for photovoltaic-storage integrated distribution transformer areas. Firstly, the definition of the distribution and consumption edge intelligence agent is established, with its architecture designed through domain-driven business modeling and event-driven state service mechanisms. Secondly, to address prominent three-phase asymmetry in distribution transformer areas, an improved three-phase distribution network branch power flow model is derived via convex relaxation. Finally, a three-phase coordinated multi-objective optimization model for photovoltaic-storage distribution transformer areas is developed based on the improved power flow model, integrated with the edge agent's state service mechanism for dynamic service processing. This achieves autonomous edge-side management encompassing localized state perception, event identification, optimized control, and closed-loop feedback. Case studies on a 21-node distribution transformer area and engineering validations demonstrate that the proposed edge intelligence agent effectively handles service requirements, while the proposed model and optimization strategy exhibit engineering-applicable accuracy and control effectiveness.