Migrating from Algolia to Meilisearch in 2026

The cost of scaling

Meilisearch recently closed a $15 million Series A round led by Felicis to expand its marketing and sales teams. Co-founders Quentin de Quelen, Clement Renault, and Thomas Payet built the project in 2018 after working at Veepee and Louis Vuitton. I see teams abandoning Algolia because the per-query pricing punishes growth. Algolia’s free tier gives 10,000 search requests per month, but its usage-based model scales linearly with success. Algolia charges $0.50 per 1,000 requests beyond the Grow plan limits. Meilisearch provides a flat alternative through Meilisearch Cloud, which costs $30 per month for 100,000 documents and 50,000 searches. You can also run a resource-based XS instance for approximately $23 per month. More than 10,000 apps today rely on Meilisearch. The project has grown to over 10 million downloads. The company plans to expand its 25-person headcount to 30 by the end of the year. The Community Edition uses an MIT license, while the Enterprise Edition uses a Business Source License 1.1. The decision to move from Algolia to Meilisearch usually follows a period where usage-based pricing becomes a burden, as Algolia charges $0.50 per 1,000 requests beyond the Grow plan limits and forces teams to negotiate with sales.

Hardware and performance

Meilisearch runs on Rust and uses LMDB as its core key-value store. I find the developer experience polished, with indexing taking one API call and returning results in under 50ms for datasets under 10 million documents. The Meilisearch binary takes up only 35 MB when installed on Ubuntu or MacOS. It provides six default ranking rules including typo, words, proximity, attribute, words position, and exactness. However, Meilisearch struggles with massive datasets. In a benchmark using 116 million Hacker News documents, Meilisearch failed to load the data even after two days of processing. Manticore Search loaded the same dataset in 2 hours and 9 minutes using a single CPU core.

Meilisearch imposes a hard-coded limit of 200 indices and only indexes the first 1,000 words of any attribute per index. It also consumes 69 GB of disk space for 10 million NGINX logs, whereas Manticore uses only 4.4 GB. One user found that a 96-core machine did not improve Meilisearch indexing speed because the engine did not use the cores efficiently. The engine processes document updates in a single queue, which creates massive bottlenecks for large datasets that require frequent updates. I observed that the engine frequently runs single-threaded, a behavior that impacts indexing performance. Can Meilisearch solve the single-threaded write bottleneck with its planned rewrite?

Vector search and migration

The gap between managed services and open-source tools has narrowed. Meilisearch provides hybrid search by allowing you to configure an embedder like OpenAI, Cohere, or Mistral. Users bring their own model and key, and the engine handles the plumbing. This hybrid approach dials between keyword and vector results by using a semantic ratio to ensure that the search results remain highly relevant to the user. Embedding latency ranges from about 10ms for locally-run models up to 800ms when calling out to a hosted provider. Meilisearch supports all languages that use whitespace to separate words, and it includes support for Chinese, Japanese, and Hebrew. For those using Algolia’s InstantSearch, the @meilisearch/instant-meilisearchadapter translates requests. This adapter allows teams to swap the client in code to avoid a total rebuild. Teams must prepare to re-tune ranking rules because Meilisearch leads with how many query terms matched rather than typo count. Algolia’s typo criterion runs first in the ranking, while Meilisearch runs words first.