Visual Agent

An AI agent implementation paradigm that relies on multimodal large models to interpret screenshots and produce operation instructions. It requires no pre-authored interface rules and generalizes broadly, but each decision step requires a complete screenshot–upload–inference–return cycle.