RelyLoop is public

RelyLoop is now open source under Apache 2.0. It's the only tool that runs autonomous, full-search-space Bayesian optimization across all three major open-source search engines — Elasticsearch, OpenSearch, and Apache Solr — and ships the winning configuration as a Pull Request to your config repo, where your existing approvers and CI decide what reaches production.

The shape is a Karpathy-style loop — propose, evaluate, select, repeat — with Optuna's TPE sampler tuning field boosts, function scores, fuzziness, mm, tie-breakers, and hybrid weights together, scored against your judgments with ir_measures. The loop ends at the PR: RelyLoop never sits on the live search-serving path.

RelyLoop is alpha. MVP1 — the full loop — shipped, and MVP2 is already underway: the Apache Solr adapter and UBI-derived judgments (including the hybrid UBI + LLM mode) have landed. The Roadmap tracks live status. We're single-vendor-stewarded for now and actively looking to broaden the maintainer team.

Try the Quickstart, and tell us what breaks on GitHub.