AMBER: Training Long-Horizon Web Agents through Append-Only Memory

Chinmay Savadikar , Zhaoyu Zhang , Mingyu Zhao , Shuang Xie , Han Li , Tianfu Wu , Lingyun Wang

🏛 Institutions: North Carolina State University , Shopify
📅 Date: October 5, 2026
📑 Publisher: arXiv
💻 Env: Web
🔑 Keywords: AMBER append-only memory long-horizon web agents WebArena WorkArena

TLDR

Proposes AMBER (Append-only Memory Bank for Evidence Retention), an architecture that replaces overwrite memory mechanisms with structured append-only evidence recording for long-horizon web navigation. AMBER prevents critical historical evidence and error feedback from being discarded during multi-step execution, boosting task completion across WebArena and WorkArena.

Open paper Report issue