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

RoMem addresses a critical limitation in current LLM agents: the 'recency bias' of vector-based memory and the computational cost of managing Knowledge Graphs (KGs). By applying continuous phase rotation (reminiscent of Rotary Positional Embeddings but for temporal KGs), it allows for a more fluid representation of time. Despite the technical novelty, its defensibility is currently low (4) due to being a fresh research release (0 stars, 1 day old) without a surrounding ecosystem. The 5 forks indicate immediate interest from the research community, likely as a baseline for other papers. Frontier labs represent a medium risk; while they currently favor massive context windows or simple RAG, the shift toward 'agentic memory' makes TKG-like structures attractive. However, the high platform risk comes from vector database providers (Pinecone, Milvus, Weaviate) who are actively incorporating temporal decay and graph-native features into their managed services. If the 'continuous phase rotation' approach proves superior, it will likely be absorbed as a standard embedding transformation rather than a standalone product.