Agentic tasks are inherently long-horizon and multi-turn, requiring constant context accumulation through interactions with the environment. Existing context compression methods lead to information loss and are constrained by rigid heuristic rules, misaligning with the agent's evolving reasoning focus. We propose Agentic Context Management (ACM), a framework that equips agents with purpose-built context editing tools for lossless context management.
Inspired by the interaction between short-term and long-term human memory, the agent autonomously decides when to compress its context, offloads discarded content to an external memory system, and queries it on demand for later retrieval.
Building on this framework, we further develop a post-training pipeline that constructs high-quality demonstrations of context management and enhances model performance on both agentic search and coding tasks.
Further analysis reveals that effective context management reduces peak token pressure, enables extended explorations, and yields more consistent solutions across independent trials. Code, data, and model checkpoints are available at GitHub.
Blogger's Review: The introduction of the Agentic Context Management (ACM) framework represents a significant advancement in context management for long-horizon tasks. By combining autonomous decision-making with external storage, ACM not only minimizes information loss but also enhances task flexibility and efficiency, which will have a profound impact on future intelligent agent research.