Meilisearch

Last updated: Apr 14, 2026

---

Rationale

We use Meilisearch Cloud as a unified search engine for keyword and semantic search across Fluid Attacks' documentation, replacing the need to maintain separate providers for each search type.

Meilisearch allows us to achieve the following:

• It supports hybrid search that auto-balances keyword and vector results per query.
• It generates embeddings automatically via auto-embedding, removing the need to manage embedding infrastructure.
• It supports index replacement by clearing and re-adding all documents on each reindex, with support for atomic swap if needed in the future.
• It highlights and crops search results natively, with per-field control over snippets.
• It supports filtering and faceting, which we use for access control (public vs. private documents).
• It provides a full CRUD API for document management, enabling auditing and enumeration of indexed content.
• It exposes a single RESTful API that integrates with both the docs site and MCP-based consumers like our MCP server.
• It costs approximately $30 per month on the Build plan, with AI and semantic search included.
• It is licensed under the MIT license.

Alternatives

Below are alternatives we have used in the past that were replaced by Meilisearch.

Algolia

Algolia is an industry-standard search-as-a-service provider.

• It does not include semantic search in its standard plans. NeuralSearch requires the Elevate tier at $50,000+ per year. Without it, a second provider is required for semantic search.
• It does not generate embeddings, as semantic search is only available on the Elevate tier.
• It supports atomic index replacement via replaceAllObjects in a single call.
• It provides native per-field highlighting.
• It provides powerful faceting and configurable searchable attributes.
• It provides a full CRUD API for document management.
• It exposes a RESTful API with excellent SDKs, but only covers keyword search, requiring a separate provider for MCP-based semantic consumers.
• It costs approximately $50 per month for keyword-only search at our usage level.
• It is licensed under a proprietary license.

AWS Kendra

AWS Kendra is an enterprise search service by AWS.

• It supports both keyword search via QueryCommand and semantic search via RetrieveCommand, but does not combine them in a single hybrid query.
• It does not support configurable embedding models; it relies on its own built-in retrieval model.
• It does not support atomic index replacement. Replacing index contents requires a data source rotation workaround.
• It does not support highlighting on semantic search results.
• It supports limited filtering via AttributeFilter on document attributes, but does not provide faceting.
• It does not provide a document listing API, making it impossible to enumerate or audit what is in the index.
• It exposes an AWS SDK-based API, but keyword and semantic search are separate operations, requiring two different API calls for each use case.
• It costs approximately $300 per month at our usage level.
• It is licensed under a proprietary license.

Typesense Cloud

Typesense is an open-source search engine with a cloud offering.

• It supports hybrid keyword plus vector search.
• It generates embeddings automatically with auto-embedding and optional GPU acceleration.
• It supports atomic index replacement via collection aliases, but this requires a manual three-step process.
• It provides native per-field highlighting.
• It supports dynamic filtering and faceting at query time.
• It provides a full CRUD API for document management.
• It exposes a RESTful API with Algolia-compatible clients, capable of serving both keyword and semantic consumers.
• It costs approximately $30 per month, but requires manual cluster sizing decisions.
• It is licensed under GPL-3.0.

Usage

We use Meilisearch Cloud as the centralized search engine for all documentation search at Fluid Attacks. It powers both the search feature on docs.fluidattacks.com and semantic retrieval for the MCP server.