Open multimodal web agent built on GPT-4V that autonomously completes real-world web tasks end-to-end with vision + DOM grounding.