31 self-hosted Elasticsearch alternatives

Last reviewed Aug 26, 2026 · 456 words

The escape hatch from Elastic's licensing already exists: OpenSearch, the Apache-2.0 fork AWS started in 2021, keeps the query DSL, the dashboards, and most client compatibility. But before you migrate to it, ask the quieter question - whether you ever needed a Lucene cluster at all. In my experience most self-hosters running Elasticsearch are using about 5% of it, and that 5% fits in a 256 MB single binary.

What you gain and what you give up

You gain a genuinely open license (Elastic moved off Apache-2.0 in 2021; it has since added an AGPL option, but the ecosystem trust never fully recovered) and, if you switch engines rather than forks, a staggering drop in operational weight. Meilisearch runs in 256 MB and Typesense in 512 MB, both single binaries with typo tolerance and faceting built in. An OpenSearch node wants 2 GB minimum and cluster tuning before it wants anything else.

You give up the ecosystem: Kibana-grade dashboards, the aggregation depth, ingestion pipelines, and a decade of plugins. Meilisearch simply has fewer features than Elasticsearch, and Typesense's plugin ecosystem is smaller. If your workload is analytics-heavy rather than search-heavy, the lightweight engines are not a substitute.

Which replacement fits which user

Running log search or observability dashboards: OpenSearch (13,577 stars, Java) is the only option here that keeps your Kibana muscle memory, at the cost of being resource-heavy. Powering search in an application you build: Meilisearch (59,093 stars, MIT, Rust) if you want the easiest setup, Typesense (26,478 stars, GPL-3.0, C++) if you want open-source high-availability clustering without a paid tier. Logs at serious scale on a budget: Quickwit (Apache-2.0, Rust) searches directly on object storage and exposes an Elasticsearch-compatible API, though it is built for append-only data, not general search.

Migration means reindexing, with one exception

OpenSearch forked from Elasticsearch 7.10, so 7.x-era clients, index templates, and dashboards mostly carry over; the further your cluster drifted past 7.x, the more you patch. Everything else is a rebuild: neither Meilisearch nor Typesense imports an Elasticsearch index, and the query DSL does not translate. You re-push documents from your source-of-truth database and rewrite queries against a much smaller API - which typically takes days, not weeks, precisely because the APIs are smaller.

The default

If Elasticsearch is your log store, take OpenSearch and keep your dashboards. For everyone else - and I think that is most people reading this - Meilisearch is the default: easiest to run, lightest on RAM, and the features you will actually miss are ones you were not using.