COALA

Mariana Souza describes a three-index Elasticsearch architecture for persistent LLM agent memory that achieves 0.89 recall without tenant leaks. The system separates episodic, semantic, and procedural…