Overview
Field: NLP Authors: Fei Tang, Huawen Shen, Zhiqiong Lu, Zhengxi Lu, Pengyuan Lyu, et al. Published: 2026-08-25 arXiv: 2608.24848
Abstract (translated from Chinese summary)
Web agents that act from rendered pixels avoid the fragility and heavy token cost of reading a page's HTML or accessibility tree, but training them depends on large amounts of high-quality interaction trajectories, and how to produce such data at scale remains an open problem. Public datasets typically contain only a few thousand trajectories drawn from a fixed and narrow set of websites, and even recent automated synthesis pipelines stay bound to predefined site lists or tutorial sources, so the number of distinct websites the agent ever sees barely grows.
The authors present BrowserForge, a framework that generates web interaction data at scale by driving many browser sandboxes in parallel over the open web. BrowserForge couples three components:
1. Open-web sourcing stage — exposes the agent to hundreds of thousands of real, publicly reachable websites. 2. Sandbox cluster manager — schedules hundreds of concurrent browsers and achieves high utilization. 3. Proposer-solver dual-agent loop — converts raw pages into executable tasks, then collects verified trajectories.
Original Abstract (excerpt)
> Web agents that act from rendered pixels avoid the fragility and heavy token cost of reading a page's HTML or accessibility tree, but training them depends on large amounts of high-quality interaction trajectories, and how to produce such data at scale remains an open problem. Public datasets typically contain only a few thousand trajectories drawn from a fixed and narrow set of websites... We present BrowserForge, a framework that generates web interaction data at scale by driving many browser sandboxes in parallel over the open web.
---
*Auto-collected on 2026-08-27*