Meilisearch
Rust-based open-source search. Simple API, fast typo tolerance, growing fast.
VISIT MEILISEARCHKey takeaway: Reach for Meilisearch when you want a search server running in an afternoon and nobody on the team wants to become a relevance engineer. Its default ranking, typo tolerance and prefix search are the best out-of-the-box behaviour in the category. Against Typesense the gap is narrow: Meilisearch for the smoother start, Typesense when you are porting an Algolia front end.
Quick facts
- CategoryOpen-source server
- EngineRust
- PricingFreemium
- LicenseMIT
- Created2018
- GitHub stars50.2k
- Vector supportYes
- Edge-readyNo
What it is
Meilisearch is a Rust-based open-source search engine. Strong on typo tolerance, simple API, and the cleanest first-run experience in the category. Meilisearch Cloud for managed hosting. Native vector search and hybrid retrieval as of 2024.
Best for
- Greenfield search with the fastest time-to-first-query
- Apps that prioritise typo tolerance and instant-search UX
- Teams allergic to Java-stack operational complexity
- Hybrid retrieval (lexical + vector) for AI-aware search
When not to pick it
Skip Meilisearch for very large corpora (50M+ documents) where Elastic / OpenSearch is more proven. Skip if your team is already on Algolia and migration cost outweighs the savings.
My take
Meilisearch is the most polished open-source search engine in 2026. Rust performance is real; the DX is genuinely leading for self-hosted.
Defaults you rarely need to override
Meilisearch ships an opinionated ranking pipeline, words then typo then proximity then attribute then sort then exactness, applied in that order and adjustable per index. The effect is that a fresh index returns sensible results the first time you query it, which is not true of most engines. Prefix matching is on by default, so search-as-you-type works without a separate autocomplete index, and typo rules scale with word length rather than applying a flat edit distance. Configuration is a short list of index settings: searchable attributes, filterable attributes, sortable attributes, synonyms, stop words. There is no analyser chain to assemble and no mapping to get wrong. That narrowness is deliberate and it is why teams with no search background ship something decent quickly. It also means the ceiling on custom relevance sits lower than Vespa or Elasticsearch, where ranking is programmable rather than ordered.
Scaling limits and the operational shape
Meilisearch is a single binary with an LMDB-backed store on local disk, which makes small deployments trivial: one container, one volume, one API key scheme. That simplicity is also the constraint. Indexing runs as an asynchronous task queue, so bulk writes complete later and you poll for status rather than getting write-then-read consistency, which catches teams out on first integration. Horizontal scale is the weaker story: past tens of millions of documents, Elasticsearch or OpenSearch is the more proven answer, and heavy sustained write throughput on a large index makes the single-writer design felt. Hybrid retrieval combining lexical ranking with embeddings works well and Meilisearch can call an embedding provider for you. Meilisearch Cloud exists if you want the same engine without owning the disk, the backups and the upgrade path yourself.
Frequently asked questions
Is Meilisearch free?
Yes. The engine is open source under the MIT licence and free to self-host, which is more permissive than the GPL that Typesense and Manticore ship under, so it raises fewer questions in a legal review. You pay for the machine and the time to run it. Meilisearch Cloud is the paid managed option if you would rather not own backups, upgrades and disk.
Meilisearch or Typesense?
Meilisearch if you want the best defaults and the quickest path to a working search box, and if MIT sits better with you than GPL. Typesense if you are migrating from Algolia, need InstantSearch compatibility, or want a more conventional clustered deployment. Performance and feature coverage are close enough that the decision rarely turns on benchmarks.
How many documents can Meilisearch handle?
Millions comfortably on sensible hardware, with RAM and disk being the limits worth watching. It gets less comfortable in the tens of millions, where Elasticsearch or OpenSearch have the more proven track record for sharding and sustained write throughput. If your corpus is under a few million documents, Meilisearch will not be the thing that breaks first.
Links
Compare Meilisearch side-by-side
Similar tools you should also consider
If Meilisearch is your pick, the next conversation is short
The 30-min call is where your search choice becomes a real architecture, a relevance-tuning plan, and a price range you can take to your stakeholders.