aiqbench analysis
StateComp introduces a novel framework for managing context in long-horizon AI agents by dynamically determining when historical interactions can be safely compressed based on the agent's current state. This approach significantly reduces token usage and accelerates representation extraction without compromising task performance, addressing critical efficiency bottlenecks in extended agent workflows. The method leverages hidden representations from frozen language models to balance information retention with computational overhead.
StateComp: Learning When to Compress History in Long Horizon Agents
arXiv:2609.27298v1 Announce Type: new Abstract: Long-horizon agents continuously accumulate interaction history during task execution, yet the importance of past interactions changes as the agent state evolves. Existing context management methods largely compress history based on fixed windows, periodic schedules, or current relevance, overlooking a more fundamental question: when has a past interaction become safe to replace? Premature compression may remove information still needed for future actions, while overly conservative retention leads to substantial context overhead. To address this, we propose State Conditioned Compression (StateComp), a framework that determines when historical interactions can be safely compressed according to the current agent state. StateComp constructs KEEP and READY supervision through a two-stage annotation procedure and trains an imbalance-aware router on hidden representations from a frozen language model. A bounded state representation further reduces the cost of evaluating long histories, while adjacent READY interactions are grouped into continuous spans and replaced with compact summaries during execution. Experiments on WorkBuddyBench show that StateComp reduces total agent and summarization tokens by 52.27% while maintaining task performance, and achieves a 12.67-fold speedup in representation extraction.
Abstract:Long-horizon agents continuously accumulate interaction history during task execution, yet the importance of past interactions changes as the agent state evolves. Existing context management methods largely compress history based on fixed windows, periodic schedules, or current relevance, overlooking a more fundamental question: when has a past interaction become safe to replace? Premature compression may remove information still needed for future actions, while overly conservative retention leads to substantial context overhead. To address this, we propose State Conditioned Compression (StateComp), a framework that determines when historical interactions can be safely compressed according to the current agent state. StateComp constructs KEEP and READY supervision through a two-stage annotation procedure and trains an imbalance-aware router on hidden representations from a frozen language model. A bounded state representation further reduces the cost of evaluating long histories, while adjacent READY interactions are grouped into continuous spans and replaced with compact summaries during execution. Experiments on WorkBuddyBench show that StateComp reduces total agent and summarization tokens by 52.27% while maintaining task performance, and achieves a 12.67-fold speedup in representation extraction.
Artificial Intelligence (cs.AI)
arXiv:2609.27298 [cs.AI]
(or arXiv:2609.27298v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.27298 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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.
Full text under the publisher's open licence · 268 words
Related coverage
When History Is Multimodal: Rethinking Context Management for Long-Horizon Agents
arXiv · Sep 1, 2026
HiMPO: Hindsight-Informed Memory Policy Optimization for Less-Entangled Credit in Long-Horizon Agents
arXiv · Jun 17, 2026
Long-Horizon Embodied Decision-Making via Multimodal Memory Compression
arXiv · Aug 4, 2026
When Can Agents Forget Their Reasoning? ICLR for Long-Horizon Agent Context Compression
arXiv · Sep 25, 2026