We built a persistent agent memory layer on Elasticsearch with 0.89 recall

Software Development, Developer Tools & Environments, Artificial Intelligence(elastic.co)view on HackerNews
Elasticsearchhybrid retrievalRRFrerankersupersessiondecayDLSagent memorypersistent memorymulti-tenantvector databaseNS

Author: showmypost

Date: 6/18/2026

Article Summary:
This article describes the implementation of a persistent agent memory layer on Elasticsearch, a hybrid retrieval and DLS system that uses three indices, hybrid retrieval with RRF and a reranker, supersession, decay, and per-user DLS isolation.