Meilisearch, a lightning-fast search engine designed to integrate seamlessly with applications and websites, has recently gained attention for its hybrid search capabilities. As the search engine market continues to evolve with AI-powered features, developers are sharing their experiences with Meilisearch in production environments and comparing it to alternatives like Typesense, Elasticsearch, and emerging solutions like Orama.
Production Readiness and Performance
Meilisearch has been production-ready since version 1.0, with several developers reporting successful implementations handling millions of documents. One user mentioned deploying it for a corpus of 7 million articles with good results, while another noted easy setup on an $8/month Hetzner instance handling a million records. However, questions remain about high-availability options, with some users noting that running multiple synchronized instances appears to be the only solution for redundancy.
The search engine's memory usage has sparked discussion, with some users observing high memory consumption (3GB+) even for small instances. A Meilisearch team member clarified this behavior:
Actually, Meilisearch uses the RAM available by design because it uses LMDB (which is memory-mapped) under the hood for key-value storage. It's a feature, not a bug. The cool thing about that is that it is the OS that will get to choose which process to allocate memory.
This design choice means Meilisearch leverages available system memory for performance but relies on the operating system to manage memory allocation between processes.
Hybrid Search Implementation
A significant topic of discussion centers around Meilisearch's approach to hybrid search, which combines traditional full-text search with semantic (vector-based) search. This implementation differs from competitors like Typesense, which uses Reciprocal Rank Fusion (RRF) to combine results.
The conversation revealed tensions between different approaches to hybrid search, with one commenter (who was later identified as a Meilisearch team member) criticizing Typesense's fusion search method as pretty much useless because you would always have one or the other search strategy that would give you awful results. This sparked a response from a Typesense representative defending their approach as well-researched and documented in academic papers.
For developers considering implementing hybrid search, one commenter advised: One thing to always dig into is how your hybrid search solution filters the vector search index. This is not at all standardized, often overlooked, but when you want 'top X most similar to query by embedding, but also in Y category/match Z search terms' it's the core thing your hybrid search is doing.
Indexing Speed and Document Updates
Several users highlighted challenges with Meilisearch when handling frequently changing documents. One noted that when documents change often and search results need to reflect those changes quickly, it ends up with pending tasks for hours. For static or rarely changing content, however, Meilisearch received praise for its performance and ease of setup.
The Meilisearch team pointed to improvements in their latest release (v1.12), which includes a new indexer version that is much faster, leverages high usage of parallel processing, and reduces disk writes. They also emphasized their disk-first approach, where content is written to disk rather than kept primarily in memory, allowing for instant reboots and easier upgrades without reindexing.
Alternative Solutions
The discussion revealed several alternatives to Meilisearch that developers are considering. Typesense was frequently mentioned, particularly for high-availability scenarios. Others pointed to Tantivy-based solutions like Quickwit and ParadeDB, though concerns were raised about Quickwit's future after being acquired by Datadog. A newer entrant, Orama, was highlighted for its compact size (under 2KB) and support for full-text, vector, and hybrid search that can run in browsers, servers, or edge networks.
For those seeking to combine full-text search with vector embeddings, suggestions included Elasticsearch, Vespa, and Typesense, though opinions varied on implementation quality and ease of use.
As search technology continues to evolve with AI capabilities, Meilisearch's approach of maintaining backward compatibility while improving performance positions it as a contender in this space, though developers should carefully consider their specific needs around document update frequency, memory usage, and high availability when choosing a search solution.
Reference: Meilisearch: AI-powered search in GA
 |
|---|
| A sleek movie search application interface illustrating how various search technologies, like Meilisearch, assist users in efficiently accessing relevant content |