AriadneMem: Threading the Maze of Lifelong Memory for LLM Agents

Long-horizon LLM agents require memory systems that remain accurate under fixed context budgets. However, existing systems struggle with two persistent challenges in long-term dialogue: (i) $\mathbf{\text{disconnected}}\mathbf{\text{evidence}}$, where multi-hop answers require linking facts distributed across time, and (ii) $\mathbf{\text{state}}\mathbf{\text{updates}}$, where evolving information (e.g., schedule changes) creates conflicts with older static logs. We propose AriadneMem, a structured memory system that addresses these failure modes via a decoupled two-phase pipeline. In the $\mathbf{\text{offline}}\mathbf{\text{construction}}\mathbf{\text{phase}}$, AriadneMem employs \emph{entropy-aware gating} to filter noise and low-information message before LLM extraction and applies \emph{conflict-aware coarsening} to merge static duplicates while preserving state transitions as temporal edges. In the $\mathbf{\text{online}}\mathbf{\text{reasoning}}\mathbf{\text{phase}}$, rather than relying on expensive iterative planning, AriadneMem executes \emph{algorithmic bridge discovery} to reconstruct missing logical paths between retrieved facts, followed by \emph{single-call topology-aware synthesis}. On LoCoMo experiments with GPT-4o, AriadneMem improves $\mathbf{\text{Multi-Hop}}\mathbf{\text{F1}}\mathbf{\text{by}}\mathbf{\text{15.2\%}}$ and $\mathbf{\text{Average}}\mathbf{\text{F1}}\mathbf{\text{by}}\mathbf{\text{9.0\%}}$ over strong baselines. Crucially, by offloading reasoning to the graph layer, AriadneMem reduces $\mathbf{\text{total}}\mathbf{\text{runtime}}\mathbf{\text{by}}\mathbf{\text{77.8\%}}$ using only $\mathbf{\text{497}}$ context tokens. The code is available at https://github.com/LLM-VLM-GSL/AriadneMem.