Memory as Action: Autonomous Context Curation for Long-Horizon Agentic Tasks

Researchers have developed Memory-as-Action (MemAct), a framework that treats working memory management as learnable policy actions for large language models, enabling joint optimization of information retention and task performance through end-to-end reinforcement learning. The MemAct-RL-14B model matched the accuracy of models 16 times larger while reducing average context length by 51%, addressing attention dilution in long-horizon tasks. The framework introduces Dynamic Context Policy Optimization to restore training efficiency without compromising reasoning integrity, with learned strategies that adapt to model capabilities and generalize across task complexities.

Computer Science Artificial Intelligence Submitted on 14 Oct 2025 v1 https://arxiv.org/abs/2510.12635v1 , last revised 7 May 2026 this version, v3 Title:Memory as Action: Autonomous Context Curation for Long-Horizon Agentic Tasks View PDF /pdf/2510.12635 HTML experimental https://arxiv.org/html/2510.12635v3 Abstract:Long-context Large Language Models, despite their expanded capacity, require careful working memory management to mitigate attention dilution during long-horizon tasks. Yet existing approaches rely on external mechanisms that lack awareness of the agent's reasoning state, leading to suboptimal decisions. We propose Memory-as-Action MemAct , a framework that treats working memory management as learnable policy actions. By formulating context management as in-place editing operations deletion, insertion , MemAct enables joint optimization of information retention and task performance through end-to-end reinforcement learning. To address the computational challenges of dynamic context updates, we introduce Dynamic Context Policy Optimization, which restores training efficiency without compromising reasoning integrity. Experiments show that MemAct-RL-14B matches the accuracy of models $16\times$ larger while reducing average context length by 51\%, with learned strategies that adapt to model capabilities and generalize across task complexities. Submission history From: Yuxiang Zhang view email /show-email/46b9f528/2510.12635 Tue, 14 Oct 2025 15:29:57 UTC 327 KB v1 /abs/2510.12635v1 Sat, 10 Jan 2026 01:44:56 UTC 374 KB v2 /abs/2510.12635v2 v3 Thu, 7 May 2026 13:18:53 UTC 371 KB References & Citations Loading... Bibliographic and Citation Tools Bibliographic Explorer What is the Explorer? https://info.arxiv.org/labs/showcase.html arxiv-bibliographic-explorer Connected Papers What is Connected Papers? https://www.connectedpapers.com/about Litmaps What is Litmaps? https://www.litmaps.co/ scite Smart Citations What are Smart Citations? https://www.scite.ai/ Code, Data and Media Associated with this Article alphaXiv What is alphaXiv? https://alphaxiv.org/ CatalyzeX Code Finder for Papers What is CatalyzeX? https://www.catalyzex.com DagsHub What is DagsHub? https://dagshub.com/ Gotit.pub What is GotitPub? http://gotit.pub/faq Hugging Face What is Huggingface? https://huggingface.co/huggingface ScienceCast What is ScienceCast? https://sciencecast.org/welcome Demos Recommenders and Search Tools Influence Flower What are Influence Flowers? https://influencemap.cmlab.dev/ CORE Recommender What is CORE? https://core.ac.uk/services/recommender arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs https://info.arxiv.org/labs/index.html .