hcompai/surfer-h-cli

Leverages Holo Action Vision-Language Models (VLMs) with three-component architecture—policy for decision-making, localizer for UI element detection, and validator for success verification—enabling autonomous web task completion. Supports multiple deployment options including remote H Company API inference, local vLLM serving, Docker containers, and AWS SageMaker, with model variants ranging from 3B to 7B parameters optimized for accuracy or efficiency tradeoffs.

151 stars.