Elasticsearch powers a tenant‑isolated agent memory with 0.89 recall
The blog details a three‑index architecture that gives agents a persistent, long‑term memory store, supports content, temporal, and user‑specific queries, and isolates data per tenant. Using hybrid recall with a RRF reranker, it reaches 0.89 recall@10 on a 168‑question benchmark with zero cross‑tenant leaks. The complete code is on GitHub, ready for integration into any Elastic‑based AI stack.