Statecomp cuts long agent memory by half while speeding up processing
StateComp: Learning When to Compress History in Long Horizon Agents
Artificial Intelligence
Summary
Many AI agents need to remember lots of past actions to do long tasks well, but remembering too much can slow them down. The authors created StateComp, a method that learns exactly when it's safe to compress old memories based on the agent’s current situation. This avoids getting rid of important information too soon or holding onto unnecessary details for too long. Their tests showed StateComp halves the memory needed and speeds up processing by over 12 times without hurting how well the agent performs.
What this means in practice
- •For software engineers building conversational agents: Reduce memory and processing load in chatbots by dynamically compressing conversation history based on current dialogue context.
- •For robotics developers: Improve efficiency in robots executing long tasks by selectively summarizing past actions when they are no longer critical.
Authors
Mingxuan Wang, Hongyue Chen, Yinglong Guo, Fei Luo, Chao Ning, Bo Wang, Guorun Yao, Yanbiao Ma, Jungong Han
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.