WebAnything creates scalable faithful web environments for agent training
WebAnything: From the Wild Web to Scalable Agent Environments
Artificial Intelligence
Summary
Training web agents to interact with websites needs real and repeatable environments, but building these manually for each site is hard and slow. The authors present WebAnything, a fully automatic system that recreates working versions of real websites without manual fixes after setup. WebAnything captures site structures and interactions, then verifies and repairs them to make reusable environments. Their tests show it works better than previous human-assisted methods and improves AI agent performance when trained with its data.
What this means in practice
- •For web developers: Build accurate web replicas automatically for testing web agents across diverse sites without manual coding or repairs.
- •For ai product teams: Improve AI assistant training by using scalable, realistic web environments from real URLs to increase task success rates.$Commercial implications: Enables selling AI assistants trained on faithful web environments that better handle real-world tasks, improving product reliability.
Authors
Tianci Xie, Zikun Cai, Chaoyun Zhang, Pinxue Guo, Bingxuan Li, Rui Zheng, Hang Yan
Abstract
Web agents require diverse, executable, and resettable website environments for reproducible training and evaluation. Human-assisted mirrors demand per-site engineering, while generative construction can diverge from specified real websites. Capture-and-replay preserves observations, with interaction coverage constrained by captured content and replay capabilities. We present WebAnything, a fully automated framework that reconstructs source-grounded, reusable Web environments from real URLs without per-site manual implementation or repair after initial setup and authorization. An Explorer-Builder collects browser-observable structures, resources, and interaction outcomes; an independent verifier checks frozen candidates and guides local repair or targeted source recapture. We construct 244 website artifacts across 14 application domains and 53 categories, supporting 12,984 executed and accepted tasks. Compared with human-assisted WebHarbor, WebAnything achieves higher aggregate scores on three fidelity metrics, with 12 wins, two ties, and no losses in a blind audit. Fine-tuning Qwen3.5-9B on 5,856 interaction trajectories improves REAL success by 9.9 percentage points, Odysseys-Easy Rubric Avg and Task Avg by 6.25 and 6.32 percentage points, and WebArena-Verified Overall success by 13.33 percentage points under the fixed harness. Code: https://github.com/china-qijizhifeng/Web-Anything