Meilisearch 1.12 shifts the search engine market

Meilisearch 1.12 provides the hybrid search capability that solves the semantic failures seen in pure vector implementations. When users query specific technical identifiers like a PX-9000-v2 error code, vector search alone fails by returning semantically similar but irrelevant results because semantic distance remains too great from the specific technical description. In the high-dimensional vector space, a query for a specific model might be pulled toward a cluster of unrelated products. This lack of lexical precision makes hybrid search a necessity for enterprise applications. Hybrid search combines dense vector embeddings with keyword-based lexical retrieval. This method uses the Reciprocal Rank Fusion (RRF) algorithm to merge results from both approaches. The RRF algorithm ignores raw scores and focuses on rankings, using a constant k=60 to prevent a single top-ranked result from dominating the final list. This constant ensures the engine prioritizes consensus between semantic and lexical matches. In the high-dimensional space, the RRF formula calculates the score as the sum of 1 divided by the sum of k and the rank of the chunk. Using hybrid search adds compute overhead and increases retrieval latency by 10% to 30% compared to pure vector search, but it prevents the loss of precision for specialized jargon. For production RAG applications, the cost of a fast but incorrect answer from a missed keyword is far higher than the compute cost of adding a keyword index to the retrieval pipeline.

Teams choose Meilisearch when they require open-source flexibility and transparent pricing that Algolia cannot match. Algolia charges $0.50 per 1,000 searches on its Grow plan and $1.75 per 1,000 searches on its Grow Plus tier. Algolia also limits users to 10,000 searches and 50,000 records on its free plan. For users exceeding these limits, Algolia charges overages of $0.40 per 1,000 records per month. Algolia defines a search request as a group of search operations in one exchange, which simplifies pricing for federated search. In contrast, Meilisearch is unlimited in searches and documents via its open-source version. You should evaluate your scale before committing to a provider. Typesense also provides an open-source alternative that uses a RAM-first approach to return results fast. Both Algolia and Typesense store the entire index in memory to allow for search-as-you-type experiences. Algolia uses 128GB RAM clusters and recommends keeping indexes below 100GB. Typesense supports all languages except logographic ones like Chinese, Japanese, and Korean. While Algolia is advanced in AI ranking through its Elevate tier, its closed-source nature and escalating costs at scale drive many developers toward Meilisearch.

Meilisearch delivers results in under 50ms because it uses a Rust-native architecture. This speed remains consistent regardless of the dataset size. The 1.12 release includes AI Embedders that allow connections to OpenAI, Voyage AI, Cohere, Mistral, Gemini, or local Ollama instances. This automation removes the need for a separate, manual embedding pipeline. Developers can also point to a custom REST embedder to manage their own vectors. Meilisearch manages the generation of embeddings automatically during the indexing process. While this automation simplifies the workflow, can developers fully trust automated embedding processes for highly specialized domain vocabularies? Meilisearch is typo tolerant and has faceted search. The engine also supports non-Latin scripts, though it historically struggled with logographic languages. To deploy Meilisearch, developers often use the Sealos App Store for one-click deployment. Configuration requires setting the MEILI_ENV variable to either production or development. In production mode, Meilisearch requires a master key of at least 16 bytes to protect all routes except the health endpoint. The engine maintains over 40,000 stars on GitHub.