hybrid-recall

Elasticsearch blog post describes building a persistent agent memory layer with three memory types (episodic, semantic, procedural), achieving 0.89 recall on a QA eval with zero tenant leaks using hybrid recall and DLS isolation.