GLOVE: Global Verifier for LLM Memory-Environment Realignment

Most existing memory-enhanced Large Language Model (LLM) approaches implicitly assume that memory validity can be established either through external evaluators that provide task-specific success signals or through internal model cognition, such as reflection, for editing memory entries. However, these assumptions often break down in practical environments with dynamic drifts. We propose the Global Verifier (GLOVE), a plug-and-play module for LLM memory systems that establishes a relative notion of truth to achieve memory-environment realignment in the face of environmental drifts. Through active probing to detect inconsistencies between retrieved memories and fresh observations, GLOVE enables memory-environment realignment by verifying and updating memory without access to task-specific ground-truth supervision or strong reliance on model introspection. We evaluate GLOVE across a spectrum of tasks ranging from web navigation and discrete planning to continuous and embodied control, covering both simulated benchmarks and real-world robotic deployment. Across all domains, GLOVE improves adaptation across various LLM memory designs under both explicit and implicit drift, often recovering performance from near-failure to high success rates. These results suggest a practical pathway to memory-environment realignment for self-evolving cognitive agents.