[HUGGINGFACE]score: 0.63
SAM Framework Adds State-Adaptive Memory to Long-Horizon LLM Agents
May 22, 2026
SAM (State-Adaptive Memory) is a standalone framework that dynamically adapts which parts of an agent's interaction history are accessed based on the agent's evolving reasoning state, rather than truncating, compressing, or retrieving uniformly.
paper
HOW THIS AFFECTS YOU
●
builderYou can integrate SAM as a drop-in memory layer for long-running agents without redesigning the underlying LLM or tool-use pipeline.
●
researcherThis reframes long-horizon agent memory as a state-dependent access problem rather than a compression problem, offering a new modeling angle for agentic reasoning.