OpenSearch¶

Summary

OpenSearch is a first-class, shipped engine (MVP1), served by the same ElasticAdapter as Elasticsearch. The study workflow is identical — pick the cluster, run the loop.

Version support¶

Versions	Status
2.x	supported
3.x	supported

OpenSearch and Elasticsearch share one adapter implementation. Where the DSLs diverge, the adapter probes capabilities and adapts; you don't choose a code path.

What RelyLoop tunes¶

The same query-time surface as Elasticsearch — field boosts, function scores, fuzziness, minimum_should_match, tie-breakers, and hybrid weights — varied together by Optuna's TPE sampler and scored with ir_measures.

How RelyLoop relates to OpenSearch's own tooling¶

OpenSearch ships the most overlapping native tooling of any engine, so it's worth being precise about the boundary:

Search Relevance Workbench (SRW) has GA query sets, judgment lists, A/B comparison, LLM-as-judge, scheduled experiments, and UBI judgments — but its optimizer is a 66-cell grid restricted to hybrid weights, and it has no apply path by explicit RFC decision.
The OpenSearch Relevance Agent does conversational DSL recommendations for OpenSearch-only shops, but does not run multi-thousand-trial sweeps.

RelyLoop's slice is the autonomous, full-search-space Bayesian loop plus the Git-PR apply path — and the fact that the same loop runs on ES and Solr too. The citation-backed matrix is in the repo's comparison doc.

Gotchas¶

Local dev runs without security

The bundled Compose OpenSearch has the security plugin disabled for local development. Operator clusters with auth are configured per cluster with credentials from mounted secret files.

UBI judgments shipped (MVP2). OpenSearch's COEC click model is GA upstream; RelyLoop's own UBI-derived judgments (read from ubi_queries + ubi_events) landed in MVP2, with a hybrid UBI + LLM mode.