WebHarbor: Docking Real Websites for Evolving GUI Agent Environments
arXiv cs.AI · 3 Nov 2026, 04:57 AEST
WebHarbor is a suite of 153 executable and dockerized website environments designed for GUI web agents. It aims to provide realistic and interactive environments that evolve with new tasks while maintaining visual fidelity and functional correctness.
Key facts
- GUI web agents require realistic and interactive environments that remain stable during both evaluation and training.
- We propose WebHarbor, a suite of 153 executable and dockerized website environments that faithfully reconstruct the visual interfaces, content, and functional workflows of real-world web services under deterministic control.
- The resulting suite comprises 3,518 tasks spanning multi-step and post-authentication workflows, where existing works often fail to cover.
- Beyond reliable evaluation, WebHarbor provides high-fidelity synthetic web environments for cost-efficient reinforcement learning (RL).
Read aloud
WebHarbor is a suite of one hundred fifty three executable and dockerized website environments designed for GUI web agents. It aims to provide realistic and interactive environments that evolve with new tasks while maintaining visual fidelity and functional correctness. GUI web agents require realistic and interactive environments that remain stable during both evaluation and training. We propose WebHarbor, a suite of one hundred fifty three executable and dockerized website environments that faithfully reconstruct the visual interfaces, content, and functional workflows of real-world web services under deterministic control. The resulting suite comprises three thousand five hundred eighteen tasks spanning multi-step and post-authentication workflows, where existing works often fail to cover. Beyond reliable evaluation, WebHarbor provides high-fidelity synthetic web environments for cost-efficient reinforcement learning RL.
Sources in this cluster
- WebHarbor: Docking Real Websites for Evolving GUI Agent Environments
arXiv cs.AI · 3 Nov 2026, 04:57 AEST