Instead of Elasticsearch: Meilisearch

Why Meilisearch is better

• Meilisearch says search-as-you-type answers come back in under 50 milliseconds, with typo tolerance and ranking enabled by default.
• Elasticsearch means learning the JVM, Lucene, shards, replicas, mappings and analyzers. Meilisearch is one binary with no runtime dependencies.
• InstantSearch adapters and SDKs for most languages make a search bar a short job, where Elasticsearch needs query tuning for end-user search.
• Keyword, semantic and hybrid search with built-in embedding generation work without extra plugins.

When to keep Elasticsearch

Keep Elasticsearch for log and observability data, analytics aggregations, billions of documents, or complex query needs such as nested document queries. Since 2024 it is available under the AGPL again, so licensing is less of a reason to leave.

Switching

This is a rewrite of the search layer, not a drop-in. Export documents as JSON or NDJSON, define searchable, filterable and sortable attributes, and replace Elasticsearch queries with Meilisearch's simpler search API. Most site-search front ends move over by swapping the InstantSearch adapter.

Watch out

The open-source Community Edition runs on a single node. Replication and sharding for high availability need Meilisearch Cloud or the Enterprise Edition, which is under the BUSL licence. Typesense's comparison also notes that Meilisearch replication uses a fixed write leader with no automatic failover. If you need self-hosted failover, look at Typesense.

Also worth a look

• Typesense A GPL-licensed search engine with built-in clustering and automatic failover; pick it if self-hosted high availability matters.