Algolia vs Meilisearch

Short version: Algolia and Meilisearch do the same core job (instant, typo-tolerant, faceted search over JSON documents) and they feel similar to integrate, so the choice is decided by three things that are not the search box. Algolia wins on merchandising, search analytics, A/B testing, global edge latency and enterprise contracts. Meilisearch wins on cost predictability and on being a binary you can run yourself, because it is priced per instance rather than per query. If you are a product team tuning conversion rates on a large catalog, Algolia is worth its bill; if search is a feature of your app and the bill is what brought you here, Meilisearch gets you most of the way for a flat number.

I run Layerbase, which hosts Meilisearch, so I have a stake in this one. The comparison below stands without it, and I have tried to be specific about the places Algolia is simply better rather than hand-waving them. Every price and limit here was checked against the vendors' own pages and is marked as of September 2026.

If you already know you want to move, the mechanics are in Migrating from Algolia to Meilisearch. If you are shopping the whole category rather than these two, Algolia alternatives covers Typesense and Postgres full-text search too.

Contents

• Feature comparison
• Relevance and the ranking model
• Typo tolerance
• Faceting and filtering
• Hybrid and semantic search
• Analytics, A/B testing and merchandising
• Indexing model and hard limits
• SDKs and the InstantSearch question
• Hosting and licensing
• What it costs at three sizes
• Where Layerbase fits
• Picking one
• FAQ

Feature comparison

The table is not balanced on purpose. Algolia has a larger product, and most of what it has that Meilisearch does not is the merchandising and experimentation layer that sits above search rather than search itself.

Relevance and the ranking model

Both engines rank with a cascading tie-break rather than a single blended score, which is why results from them feel more alike than results from, say, Elasticsearch.

Algolia's default order is typo, geo, words, filters, proximity, attribute, exact, custom. A criterion only comes into play when everything before it ties. The custom criterion at the end is where your business signal goes: popularity, sales rank, recency.

Meilisearch's defaults are words, typo, proximity, attribute, sort, exactness, and you can reorder them or append your own attribute rules per index. Same bucket-sort idea, different first move: Algolia leads with typo count, Meilisearch leads with how many query terms matched at all.

In practice the difference shows up on long queries where a term is dropped, and it is the reason a migration never reproduces the old ordering exactly. Plan to re-tune rather than to port. The thing to know is that Meilisearch exposes the rule list as an ordered array you can rewrite, while Algolia's first seven criteria are fixed and you tune by feeding customRanking and Query Rules. Algolia's approach is more opinionated and gives better defaults; Meilisearch's is blunter and gives you the knob.

Typo tolerance

This is the feature people actually mean when they say "we need real search," and both do it well out of the box. Both enable typo tolerance by default, both scale the allowance to word length so short words are matched strictly, and both let you disable it per attribute (product SKUs and part numbers are the usual case) and per index.

I have not found a query class where one is clearly better. Algolia's typo criterion runs first in the ranking, which means a typo-free match outranks everything else before any other signal is consulted; Meilisearch runs words first, so a result matching all your terms with one typo can beat a result matching fewer terms cleanly. Neither behaviour is wrong, and both are configurable. Do not pick on this axis.

Faceting and filtering

Algolia has one setting, attributesForFaceting, that controls both what you can filter on and what returns facet counts, with modifiers (filterOnly, searchable, afterDistinct) layered on top. Meilisearch splits the same job into filterableAttributes for filtering and faceting and sortableAttributes for sorting.

The bigger structural difference is sort orders. Algolia models each sort order as a replica index, so "price ascending" and "price descending" are separate indexes kept in sync with the primary, which counts toward your index budget and your record meter. Meilisearch sorts at query time against one index. For a catalog with four sort orders, that is four Algolia indexes versus one Meilisearch index, and it is one of the places the Algolia record meter surprises people.

Filter expression syntax differs and is mechanical to convert:

// Algolia
const { hits } = await index.search('wool jacket', {
  filters: 'category:outerwear AND price < 300',
  hitsPerPage: 20,
})

// Meilisearch
const { hits } = await client.index('products').search('wool jacket', {
  filter: 'category = outerwear AND price < 300',
  limit: 20,
})

Hybrid and semantic search

Both have an answer here now, and they are shaped differently.

Meilisearch has hybrid search built into the engine: you configure an embedder on the index, Meilisearch generates and caches embeddings for documents and queries, and a semanticRatio on the search call dials between keyword and vector results. The embedder can point at OpenAI, Cohere, Mistral, Gemini, Cloudflare, Voyage, Jina, AWS Bedrock, Hugging Face, or any REST API that returns vectors, per the AI-powered search docs. You bring the model and the key; the engine handles the plumbing. Meilisearch Cloud additionally offers managed embedders so you do not bring a key at all.

Algolia's semantic offering is NeuralSearch, and on the pricing page as of September 2026 it sits on the Elevate plan, which is an annual contract with custom pricing. If you want vector-flavoured relevance without a sales call, that is a real difference in favour of Meilisearch.

If you are still working out whether you need vectors at all, Full-Text Search vs Vector Search runs the same queries through Meilisearch and Qdrant side by side, which is a faster way to decide than reading feature lists.

Analytics, A/B testing and merchandising

Here is where Algolia is straightforwardly ahead, and I would rather say it plainly than bury it.

Algolia ships search analytics with the product: top queries, queries with no results, click and conversion tracking tied back to individual records. Retention is 7 days on Free, 30 days on Grow, and 90 days on Grow Plus and Elevate, as of September 2026. A/B testing is listed on every plan, including Free, so you can run a ranking experiment and read a result without upgrading. Query Rules give you pinning, boosting and banner injection: 10 rules per index on Free and Grow, 10,000 on Grow Plus and Elevate. Merchandising Studio, the interface a non-engineer uses to do all of that, is Grow Plus and above.

Meilisearch has none of this in the engine. The open-source binary gives you search, not a record of what people searched for. Meilisearch Cloud adds analytics and monitoring dashboards, so the gap narrows if you are on their hosted product, but there is still no merchandising UI and no built-in experimentation framework.

For a documentation site or a SaaS app's internal search, that absence costs you nothing. For an e-commerce team whose job is measured in conversion rate, it is the whole reason Algolia is priced the way it is. If a merchandiser is going to log in and pin a product for a seasonal campaign, buy Algolia and stop reading.

Indexing model and hard limits

Both index asynchronously and both return a task you can poll. Meilisearch is noticeably faster on bulk ingest for a given amount of hardware, mostly because you are writing to one machine over your own network rather than to a distributed service over the internet, and because there is no per-operation meter making you batch carefully.

Algolia publishes hard limits, and they matter when you size a catalog. From the service limits page, as of September 2026: maximum record size is 10 KB to 100 KB depending on plan (10 KB on Free), maximum index size is 100 GB (1 GB on Free), indices per application are 10 on Free, 50 on Grow and 1,000 on Premium, and indexing is rate-limited at 10,000 operations per Unit. Long product descriptions and HTML-heavy documents hit the record size ceiling first, and the usual fix is to strip the record down to what you search and display, then fetch the rest from your database.

Meilisearch's limits are your machine's. There is no document count cap, no index count cap, and no per-operation rate limit; the constraints are RAM during indexing and disk for the index itself. That is more freedom and more rope. An index that outgrows its box degrades in ways a managed meter would have warned you about.

SDKs and the InstantSearch question

Algolia's client ecosystem is the most complete in the category, and InstantSearch (React, Vue, Angular, vanilla, iOS, Android) is a genuinely good set of components. That is the single biggest reason teams stay after deciding to leave.

The thing that makes a swap tractable is that InstantSearch is not Algolia-only. The @meilisearch/instant-meilisearch adapter (0.31.4 at the time of writing) translates InstantSearch's requests into Meilisearch's API, so if your frontend is built on InstantSearch widgets, the change is the client you hand to <InstantSearch searchClient={...}> rather than a rebuild:

import { instantMeiliSearch } from '@meilisearch/instant-meilisearch'

const { searchClient } = instantMeiliSearch(
  process.env.NEXT_PUBLIC_MEILISEARCH_HOST!,
  process.env.NEXT_PUBLIC_MEILISEARCH_SEARCH_KEY!,
)

It is not perfect parity. Facet behaviour and some widget options differ, and anything backed by Query Rules or Personalization has nothing to translate into. But it moves the frontend from "rewrite" to "swap and fix the edges," which changes the arithmetic on a migration considerably.

Server-side, both have first-party clients for JavaScript, Python, PHP, Ruby, Go, Rust, Java and more. Algolia's are more mature and better documented. Meilisearch's are fine.

Hosting and licensing

Algolia is SaaS only. There is no self-hosted Algolia, no on-prem build, and no way to run the engine yourself. What you get in exchange is their Distributed Search Network: your index is replicated to points of presence around the world and queries are answered near the user. If you have customers on four continents and search latency is a conversion metric, this is a real engineering capability that Meilisearch does not sell an equivalent of outside its own Cloud.

Meilisearch is open core, and the licence is worth reading carefully rather than repeating from memory. The repository is marked MIT AND BUSL-1.1: the Community Edition is MIT, and the Enterprise Edition parts (sharding for horizontal scale is the headline one) are under the Business Source License 1.1, which permits use, testing and development but not production use without a commercial agreement. Verified as of September 2026 from the Meilisearch LICENSE file. So "Meilisearch is MIT" is true of the engine nearly everyone runs and false as a blanket statement. If your plan is to shard a single index across machines yourself, that path is licensed, not free.

For running it, the options are your own server, Meilisearch Cloud, or a third-party host. Meilisearch Cloud's own pricing page, as of September 2026, advertises "starting at $20/month" with two billing shapes: a usage plan at $30 a month for 100K documents and 50K searches, and resource-based instances beginning at an XS at about $23 a month ($18 for 0.5 vCPU and 1 GB RAM, plus $5 for disk). There is a 14-day trial with no card and no free plan. Prices for the larger instance sizes are behind a "see all server sizes" link I could not read today, so I have left them out rather than guess.

What it costs at three sizes

The comparison people actually want. Algolia's Grow plan, as of September 2026, includes 10K search requests and 100K records a month, then charges $0.50 per additional 1,000 search requests and $0.40 per additional 1,000 records. Grow Plus keeps the record rate and raises searches to $1.75 per 1,000, which buys the AI feature set. Figures from algolia.com/pricing.

Three sizes, each with a stated search volume, all on Algolia Grow:

Two caveats so the table is not read as more precise than it is. Algolia's record meter counts records in the index, and replica indexes for extra sort orders count too, so a four-sort catalog meters closer to four times its record count than one. And the $855 row is the base search product: personalization, Recommend (10,000 requests a month included on Grow, then $0.60 per 1,000) and the crawler (10,000 a month, then $0.80 per 1,000) are separate line items on top.

The shape is what matters more than any cell. Algolia's number is a straight line through your traffic with no ceiling, so the month a campaign lands is the month you find out. The other two columns are flat and you find out in advance.

Where Layerbase fits

We host Meilisearch on Layerbase Cloud, and the specific thing we sell is the absence of a meter. New instances run Meilisearch 1.52 (upstream's latest is 1.53.2, released 2026-09-07; we move engine versions deliberately and never underneath a running database). You get the standard HTTPS API with a master key, so meilisearch-js, the other official SDKs, @meilisearch/instant-meilisearch, and scoped search keys from the /keys endpoint all work by changing a host and a key.

Meilisearch is a Pro-tier engine here, and Pro is $15/month: up to 10 databases across the 18 engines on Layerbase Cloud, a 1.5 GB always-on pool with 25 GB of storage, 30-day rolling backups, and a 7-day trial. It is not on the Free plan, so "try it for nothing" means the trial rather than the free tier. What the plan does buy that a search-only vendor cannot is the rest of the stack: the Postgres your documents come from and the Valkey in front of it are two more of those ten databases, on one bill. A search index is derived data, and keeping it next to its source is the boring fix for the stale-index problem.

Meilisearch also branches here, which neither Algolia nor Meilisearch Cloud offers: a copy-on-write clone of the index for a preview environment or a settings experiment, ready in seconds and deleted when the PR closes. And the Layerbase CLI runs the same binary locally with lbase create app-search -e meilisearch --start, so you can develop against a real Meilisearch without pointing at production.

Sizing, plainly. A 10K or 100K document index sits comfortably in the Pro pool alongside a database. A million wide documents with several filterable attributes does not: that is $10/month pool blocks (each adds 1 GB of RAM, 1 vCPU and 25 GB of storage) or a dedicated server, which starts at $35/month for 2 vCPU and 4 GB and goes to $65 and $120 for the larger presets. Pricing has the full ladder, and Meilisearch Cloud alternatives goes deeper on exactly where our pool ends.

Where we are the wrong answer: you need multi-region search, you need a contractual SLA, you need managed embedders rather than bringing your own key, or search analytics is something you look at weekly. Those are Algolia's or Meilisearch Cloud's, not ours.

Picking one

• You sell things and someone merchandises the results. Algolia. Query Rules, Merchandising Studio and A/B testing are the product, and nothing in the Meilisearch world replaces them.
• You need search answered close to users on several continents. Algolia. The Distributed Search Network is not something you rebuild on a single binary.
• Search is a feature of your app and the bill is what brought you here. Meilisearch. You keep typo tolerance, faceting, synonyms, custom ranking and instant results, and the number stops tracking your traffic.
• You want semantic search without an annual contract. Meilisearch. Hybrid search with your own embedder is available on every plan; Algolia's NeuralSearch is Elevate.
• You have compliance or data-residency requirements that need the engine on your own hardware. Meilisearch, because Algolia cannot be self-hosted at all.
• You have years of tuned ranking rules and personalization. Stay on Algolia. Porting the tuning costs more engineering time than the invoice costs money, and nothing in this post changes that.

FAQ

Is Meilisearch a drop-in replacement for Algolia?

For the data and most of the query surface, close to it: both index JSON documents with a primary key, both do typo tolerance, faceting, synonyms and custom ranking, and if your frontend uses InstantSearch the @meilisearch/instant-meilisearch adapter means you swap the search client rather than rewrite the UI. What has no equivalent is Query Rules, Personalization, Merchandising Studio and A/B testing. Result ordering will also differ slightly no matter how carefully you map the settings, because the two ranking models tie-break in a different order.

Is Meilisearch cheaper than Algolia?

At any meaningful traffic, yes, because the two price different things. Algolia meters search requests and records, so the bill climbs with usage forever. Meilisearch is priced for a box: the same instance answers 1,000 searches a month or 10 million. Worked out on September 2026 prices, a 100K-record index serving 250,000 searches a month is $120 on Algolia Grow, roughly $23 and up on a Meilisearch Cloud instance, and $15 on the Layerbase Pro plan. The gap widens as traffic grows and closes to nothing if your traffic is tiny.

Is Meilisearch open source?

Mostly, and the detail matters. The repository is licensed MIT AND BUSL-1.1: the Community Edition, which is what almost everyone runs, is MIT, while the Enterprise Edition parts (sharding is the main one) are under the Business Source License 1.1 and are not licensed for production without a commercial agreement. Algolia is closed source and cannot be self-hosted at all.

Does Meilisearch do semantic or vector search?

Yes. Hybrid search is built into the engine: you configure an embedder on the index against OpenAI, Cohere, Mistral, Gemini, Hugging Face, AWS Bedrock or any REST endpoint that returns vectors, and a semanticRatio on the search call blends keyword and semantic results. Algolia's equivalent is NeuralSearch, which is on the Elevate plan and therefore behind an annual contract.

What does Algolia do better than Meilisearch?

Four things, and they are not small. Search analytics with click and conversion attribution. Built-in A/B testing on every plan. Query Rules plus Merchandising Studio, so a non-engineer can pin, boost and inject banners. And a distributed search network that answers queries near the user worldwide. Personalization and Recommend belong on that list too if you use them.

How many records fit on a Layerbase Pro plan?

Tens of thousands to the low hundreds of thousands of documents fit comfortably in the 1.5 GB always-on pool, shared with whatever else you run there. A million wide documents with several filterable attributes is past it: that is $10/month pool blocks or a dedicated server starting at $35/month. Meilisearch itself memory-maps its index, so it is more forgiving of a small RAM allowance than an in-memory engine, but the resident set still grows with the index.

Which Meilisearch version should I expect?

Upstream released 1.53.2 on 2026-09-07. New databases on Layerbase run 1.52, and an existing database stays on the version it was created with until you choose to move it. Meilisearch Cloud tracks the current line on their own schedule.

Where to start

If you are leaning Algolia, the deciding question is whether anyone on your team will use the merchandising and experimentation tools. If the answer is yes, the bill is buying something real.

If you are leaning Meilisearch, Getting Started with Meilisearch gets an index running and searching in one TypeScript file, and create a managed Meilisearch when you are ready to point production at it.