Dev.to
6/16/2026

My AI agents spend 10 minutes every night rewriting their own memory.
Short summary
Lorekeeper is an open-source memory system for AI agents that consolidates and improves memories during idle time through automatic reflection cycles. Using local embedding models and cron-based scheduling, it maintains 84.6% Recall@1 on 300+ memories—substantially outperforming naive storage systems that degrade from 80% to 40% recall after just 200 items. The system prioritizes privacy by running entirely locally with optional cloud support, while enabling agents to autonomously discover new semantic connections between related memories.
- •Open-source memory layer automatically refines agent memories through idle-time reflection cycles; scores, merges duplicates, and archives stale facts
- •Achieves 84.6% Recall@1 and 96.6% Recall@5 on 300+ memories; naive storage degrades from 80% to 40% recall after just 200 items
- •Privacy-first design runs locally with optional cloud; enables agents to discover semantic connections between memories, analogous to human sleep consolidation
Generated with AI, which can make mistakes.
Is this a good recommendation for you?



