Long-term memory significantly enhances language-model agents but introduces a critical failure mode: retrieval surfaces relevant yet stale, deleted, or mutually inconsistent facts. A query about a current address may retrieve the correct address alongside deleted ones. The answer is present, but the memory set is unsafe. We reframe agent memory as a governed object-selection problem and present MemGov, a substrate-agnostic middleware layer between existing retrieval pipelines and agent context. MemGov wraps retrieved evidence as explicit memory objects, detects must-link and cannot-link constraints, and applies deterministic constraint-aware repair, returning either a certified safe memory set or an auditable obstruction certificate rather than silently passing unsafe combinations. Evaluations on controlled governance scenarios and MemoryAgentBench-style end-to-end tasks, deliberately separating retrieval quality from governance decisions, demonstrate that while raw retrieval recovers answer-bearing evidence successfully, it simultaneously returns detected unsafe memory combinations in 47-51% of cases. The same retrieval substrates governed by MemGov reduce unsafe merges to 0% while preserving answer utility within about 2-4 percentage points, with fully auditable reasoning for all repair and refusal decisions. Additional diagnostics assess eligibility, temporal constraints, and paired answer retention.

Power distribution networks are evolving due to the integration of distributed energy resources (DERs) and increased customer participation. To maintain optimal operation, minimize losses, and meet varying load demands, frequent network reconfiguration is necessary. Traditionally, the reconfiguration task relies on optimization software and expert operators, but as systems grow more complex, faster and more adaptive solutions are required without expert intervention. Data-driven reconfiguration is gaining traction for its accuracy, speed, and robustness against incomplete network data. Large language models (LLMs), with their ability to capture complex patterns, offer a promising approach for efficient and responsive network reconfiguration in evolving complex power networks.In this work, we introduce LLM4DistReconfig, a deep learning-based approach utilizing a fine-tuned LLM to solve the distribution network reconfiguration problem. By carefully crafting prompts and designing a custom loss function, we train the LLM with inputs representing network parameters such as buses, available lines, open lines, node voltages, and system loss. The model then predicts optimal reconfigurations by outputting updated network configurations that minimize system loss while meeting operational constraints. Our approach significantly reduces inference time compared to classical algorithms, allowing for near real-time optimal reconfiguration after training. Experimental results show that our method generates optimal configurations minimizing system loss for five individual and a combined test dataset. It also produces minimal invalid edges, no cycles, or subgraphs across all datasets, fulfilling domain-specific needs. Additionally, the generated responses contain less than 5% improper outputs on seen networks and satisfactory results on unseen networks, demonstrating its effectiveness and reliability for the reconfiguration task.