Algolia alternatives now the free plan stops at 50,000 records: pick by where the index lives

The useful way to compare algolia alternatives is not by feature list but by where the index physically runs, because that decides both your bill and your weekend. Algolia's free plan covers 50,000 records and 10,000 search requests a month as of October 2026. Below that line we would usually stay put. Above it, the choice is a managed cloud, your own server, or a search bundle shipped with your static site.

What Algolia's free plan includes now, and what it costs to outgrow it

Algolia's free plan lists 50,000 records, 10,000 search requests a month, 5,000 recommendation requests and 5,000 crawls a month. Its FAQ says the free plan can run in production and names small online stores as a use case, so this is not a trial that expires.

That allowance used to be far larger. The pricing-change tracker PricingSaaS logged the change in late August 2026: the Build plan was renamed Free and its included records were cut from 1 million to 50,000, with recommendations and crawls also reduced to 5,000 a month. We could not find an Algolia announcement of that cut, so treat the date as PricingSaaS's observation and the current numbers as Algolia's own.

Fifty thousand records goes quickly. A documentation site indexed per heading, a 5,000-product store with variants, or a blog with 20,000 posts will all pass it. What happens next is the real question, and Algolia's pricing page answers it: on the Grow plan, additional usage runs $0.50 per 1,000 search requests and $0.40 per 1,000 records, while Grow Plus charges $1.75 per 1,000 search requests with the same record rate. A site serving 500,000 searches a month on Grow is paying for 490,000 of them.

In our view that structure is the reason people search for alternatives rather than a complaint about the free tier. Search traffic grows with your traffic, and per-search metering turns a successful launch into a variable cost you cannot cap.

The best algolia alternatives split three ways by deployment model

Every credible replacement sits in one of three places. Managed clouds keep the operational burden with the vendor and charge for it. Self-hosted engines move the cost to a box you already pay for, and the work to you. Static search skips the server entirely and runs in the reader's browser.

We would pick the model first and the product second. It is the same method we used for choosing a Pipedream replacement by what your workflows actually do: match the architecture to the job, then compare the shortlist that survives.

Meilisearch: MIT for the community edition, with an enterprise licence line to read

Meilisearch's community edition is, according to its own pricing page, fully open source under the MIT licence and free to use commercially, with hybrid semantic and keyword search among its core features. Self-hosting it carries no usage limits.

The managed option starts at $20 a month. Meilisearch prices it two ways on that page: a usage-based estimate of $30 a month for a base plan of 100,000 documents and 50,000 searches, or a resource-based estimate of $23 a month for an XS instance of 0.5 vCPU and 1 GB RAM plus $5 of disk. Cloud includes a 14-day trial with no credit card.

The catch is the split edition. Meilisearch's enterprise edition is governed by a commercial licence or the Business Source License 1.1, and its page states it is not allowed in production without a commercial agreement; non-production testing and evaluation are permitted. Sharding, S3-streaming snapshots and search rule previews sit on that side of the line. If your plan depends on sharding a large index, you are buying a licence, not self-hosting for free.

Skip Meilisearch if you need those enterprise features on your own hardware without a contract.

Typesense: closest to Algolia in feel, with a GPL-3.0 server and an always-on cloud bill

Typesense describes itself in its GitHub README as an open source Algolia alternative and an easier-to-use Elasticsearch alternative, with typo correction out of the box. The same README gives memory figures worth planning around: a fresh server uses about 30 MB, and indexing 1 million Hacker News titles with their points used 165 MB of RAM against 88 MB of JSON on disk. Search indexes live in memory, so size your box by index, not by disk.

The licence detail matters more than most listicles admit. The README states the Typesense server is GPL-3.0 licensed while the client libraries are Apache licensed. Running it as a service for your own site is ordinary use, but if you redistribute a modified server inside a product you ship, GPL-3.0 obligations apply and that is a question for your counsel, not a blog post.

Typesense Cloud is billed differently from Algolia. Its billing documentation describes a fixed per-hour fee for the cluster configuration you pick, plus per-GB bandwidth of usually about 9 to 12 cents per GB depending on region, and states that the hourly fee applies regardless of whether the cluster is receiving traffic, because each cluster runs on dedicated hardware. Critical production support is included in that price.

That is the inverse of Algolia's model: predictable when traffic spikes, wasteful when traffic is near zero.

Skip Typesense Cloud if your search volume is tiny and bursty - you will pay for idle hours. Self-host it instead.

OpenSearch: one Apache 2.0 engine for search and the logs next to it

OpenSearch's about page says the project was released under the Apache License, Version 2.0, and puts it bluntly: "It's all Apache 2.0." For anyone who has been burned by a licence change under a dependency, that single sentence is the reason OpenSearch is on this list.

It also asks the most of you operationally. Typesense's README argues that Elasticsearch-class engines take significant effort to set up, administer and tune, against a single lightweight binary - that is a competitor's characterization, but it matches the operational reality of running a cluster, and the same shape of work applies to OpenSearch. You get analytics, dashboards and log search in the bargain, which is why we would choose it when search is one workload among several.

Upfynd also lists ElasticSearch in the same category for teams at that scale. If you compare the two, compare the licence terms as carefully as the benchmarks.

Skip OpenSearch if site search is your only requirement and nobody on the team wants to own a cluster.

Pagefind: static site search without a server, for docs and content sites

Pagefind takes the index out of the equation. Its project site describes a fully static search library that works without hosting any infrastructure: it indexes your built files, splits the index into chunks so a search loads only a small subset, and exposes a JavaScript search API that runs on any static HTML output. The stated goal is a 10,000-page site running full-text search with a total network payload under 300 kB including the library, and closer to 100 kB for most sites.

For a documentation site, a changelog or a marketing site, we think this is the cheapest correct answer: there is no index to host, no per-search meter, and nothing to rotate credentials for. It is an independent open source project rather than a company with a pricing page, which is both the appeal and the risk.

Skip Pagefind if you need server-side ranking rules, per-user permissions, personalization or search analytics. Those need a search service.

When staying on Algolia is still the right answer

The strongest case against migrating is that Algolia's free plan now covers more small sites than its old 1 million record allowance suggests, because most small sites were never near either number. If your index is under 50,000 records and your traffic is under 10,000 searches a month, you are getting a hosted search API, a crawler and its InstantSearch front-end widgets for nothing, and a migration buys you a maintenance job.

Migrating pays when one of three things is true: your record count is growing faster than your revenue, your search volume is large enough that per-search rates dominate the bill, or a licence or data-residency requirement means the index has to live on infrastructure you control.

Run the arithmetic before the rewrite. Algolia's pricing page puts 10,000 search requests and 100,000 records inside the Grow plan, so take a store with 200,000 records serving 1 million searches a month and the published Grow rates give roughly $40 of record overage and about $495 of search overage - around $535 a month, dominated almost entirely by the searches. The same index on Meilisearch's resource-based estimate or a Typesense cluster sized to hold it in RAM is a fixed instance cost that does not move when a post goes viral. On our own reading of these pricing pages, the crossover arrives sooner on search volume than on index size.

What we would check before committing to a migration

1. Index size in memory, not on disk - Typesense's own figures show an index can be roughly double the JSON it came from.
2. Whether anything you rely on sits behind a paid licence tier, as Meilisearch's sharding does.
3. Who owns upgrades, backups and the on-call page if self-hosted search goes down during a sale.
4. How much of your front end is tied to Algolia-specific UI libraries, since that code is usually the slowest part of a move.
5. Whether you need search analytics and ranking rules at all, because a static index removes a whole class of work if you do not.

Where to look next

As of October 2026, Upfynd lists 99 products in Search & Discovery, alongside 247 in Databases & Infrastructure and 211 in Open Source - the three categories this decision usually spans. Each listing carries pricing and similar products so you can shortlist in one place, and every claim above should still be checked on the vendor's own pricing page before you commit, because these numbers move.

If you build a search tool and it is not here yet, you can submit it for a free listing.