Tag
The article critiques treating fact staleness as relevance decay in agent memory, advocating for splitting memory by type at write time and using write-time contradiction checks for facts instead of decay, citing Mem0's implementation. It warns that decay works for episodic data but not for facts, which can lead to agents confidently stating contradictory information.
agentmemory is an open-source library that provides natural temporal memory for AI agents like Claude Code, Hermes, and OpenClaw. It uses a three-tier architecture with hybrid retrieval (BM25, vector, knowledge graph) and Ebbinghaus decay, achieving ~92% fewer tokens and 200x more tool calls before context limits.
This article explains that AI agents don't actually get dumber over time; instead, their underlying memory accumulates corrupted context from stored assumptions, summaries, and contradictions, leading to performance degradation. Most systems lack the ability to revise or forget information, causing decay.