No-LLM memory layer trades accuracy for cost and privacy

RE-call, a memory system that never calls an LLM, stores raw conversation turns instead of distilled summaries. On the BEAM benchmark, it scores 0.408 on temporal reasoning versus Mem0's 0.567, because it can't disambiguate multiple date instances. The trade-off avoids LLM API costs and keeps data local, appealing for privacy-sensitive or high-volume use cases.