Design Multimodal Agent Architecture with BeeAgent

Outline an agent architecture using the BeeAgent framework that leverages Gemini 2.5 Pro for visual and language understanding. The architecture should include modules for: `Perception` (interpreting visual scenes and language instructions), `Planning` (generating high-level goals and action sequences), `Action Execution` (interacting with the simulated environment via tools), and a `Memory Management` module using Postgres and pgvector. Describe how Gemini 2.5 Pro will bridge the visual and language modalities within BeeAgent's workflow.