ReSum
A method that periodically summarises interaction histories to support longer agent tasks.
ReSum periodically condenses interaction history into summaries so an agent can continue longer tasks. The basic method works without training; ReSum-GRPO uses reinforcement learning to adapt agents to reasoning over the compressed context.
Source: ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization.
Referenced in An Empirical Study of Harness Design for Coding Agents.