Time is Not a Label: Continuous Phase Rotation for Temporal Knowledge Graphs and Agentic Memory

TL;DR

RoMem introduces a continuous phase rotation method for temporal knowledge graphs, enabling better distinction between persistent and evolving facts without deletion, improving temporal reasoning and memory accuracy.

Contribution

It presents RoMem, a novel module that uses geometric shadowing via continuous phase rotation to handle temporal facts in structured memory systems.

Findings

Achieves state-of-the-art 72.6 MRR on ICEWS05-15.

Delivers 2-3x MRR and accuracy improvements on temporal reasoning tasks.

Generalizes zero-shot to unseen financial domains.

Abstract

Structured memory representations such as knowledge graphs are central to autonomous agents and other long-lived systems. However, most existing approaches model time as discrete metadata, either sorting by recency (burying old-yet-permanent knowledge), simply overwriting outdated facts, or requiring an expensive LLM call at every ingestion step, leaving them unable to distinguish persistent facts from evolving ones. To address this, we introduce RoMem, a drop-in temporal knowledge graph module for structured memory systems, applicable to agentic memory and beyond. A pretrained Semantic Speed Gate maps each relation's text embedding to a volatility score, learning from data that evolving relations (e.g., "president of") should rotate fast while persistent ones (e.g., "born in") should remain stable. Combined with continuous phase rotation, this enables geometric shadowing: obsolete facts are rotated out of phase in complex vector space, so temporally correct facts naturally outrank contradictions without deletion. On temporal knowledge graph completion, RoMem achieves state-of-the-art results on ICEWS05-15 (72.6 MRR). Applied to agentic memory, it delivers 2-3x MRR and answer accuracy on temporal reasoning (MultiTQ), dominates hybrid benchmark (LoCoMo), preserves static memory with zero degradation (DMR-MSC), and generalises zero-shot to unseen financial domains (FinTMMBench).