Meilisearch Documentation

Meilisearch and Typesense are both open-source search engines focused on speed and developer experience, but they differ in architecture, language support, and scalability.

Key Architectural Differences

• Storage & Scalability: Meilisearch uses disk-based storage with memory mapping, allowing datasets to grow beyond available RAM. Typesense stores indexes entirely in RAM, which can make large datasets costly or impractical. Meilisearch Enterprise Edition supports horizontal scaling via sharding, whereas Typesense requires every node to hold the entire dataset in RAM.
• Language Support: Meilisearch provides optimized tokenization for complex languages (Chinese, Japanese, Korean, Thai, Hebrew, Arabic) and features automatic language detection. Typesense uses Unicode-based tokenization which may struggle with languages lacking conventional word spacing.
• AI & Embeddings: Meilisearch offers high flexibility by supporting local embedding generation via Candle and any provider via HTTP API (OpenAI, Mistral, Cohere, Ollama, etc.). Typesense is limited to built-in ONNX models and specific cloud providers.
• Licensing: Meilisearch Community Edition is MIT licensed (permissive). Typesense is GPL-3 (copyleft).

When choosing between Meilisearch and Qdrant, the decision depends on whether your primary requirement is hybrid search (combining keyword and semantic search) or pure vector database operations.

Key Differences

When to choose Meilisearch

• Strong Full-Text Search & Typo Tolerance: Meilisearch is a dedicated search engine with mature typo tolerance, prefix matching, and sophisticated relevancy ranking. Qdrant relies on sparse vectors for keyword capabilities, which lacks the same depth.
• Unified Hybrid Search: Meilisearch provides a single API to combine keyword and semantic search with an adjustable balance. In Qdrant, you must implement hybrid logic yourself.
• Flexible Embedding Generation: Meilisearch integrates with OpenAI, HuggingFace, Ollama, and any REST-compatible provider for automatic embedding generation. Self-hosted Qdrant typically requires external embedding management.
• Relevancy Tuning: Meilisearch offers out-of-the-box configurable ranking rules and custom ranking attributes.
• Site/App Search Use Cases: Ideal for e-commerce, documentation, or general site search where keyword matching is critical.

When to choose Qdrant

• Pure Vector Search: Best for AI applications where vector similarity is the primary requirement.
• Advanced Vector Operations: If you need quantization, custom distance metrics, or complex metadata filtering integrated into the search process.
• External Embedding Management: If your team manages embeddings externally and needs a dedicated vector store.
• Recommendation Systems: Ideal for systems based primarily on similarity matching.

Use the Kestra Meilisearch plugin to orchestrate data synchronization from various sources into Meilisearch using declarative YAML workflows. This approach turns indexing into an observable, retryable, and scheduled job.

Core Pipeline Concepts

• Backfill vs. Incremental Sync: Pipelines typically start with a one-shot backfill (initial load) followed by an incremental sync to keep the index current.
• ION Format: Kestra extractors write data in ION format to internal storage, which the Meilisearch DocumentAdd task consumes directly. This eliminates the need for custom glue code.
• Idempotent Upserts: The DocumentAdd task performs add-or-replace operations. This makes overlapping sync windows or accidental re-runs safe.
• Handling Deletes: The DocumentAdd task does not remove documents. You must implement a delete strategy based on your source, such as using documents/delete-batch for soft deletes, handling tombstone events, or performing an index alias swap.

Security Best Practice

Always store credentials (like your Meilisearch URL and Admin API key) as Kestra secrets using the {{ secret('NAME') }} syntax. Never use plaintext in your YAML workflows.

Meilisearch offers a wide range of search and data management features. Key capabilities include:

• Full-text search: Fast, typo-tolerant keyword search using multi-criteria ranking.
• Hybrid and semantic search: Combines keyword matching with AI-powered vector search to match both terms and meaning.
• Geo search: Filtering and sorting results by geographic location using radius, bounding box, or polygon.
• Agentic search: Enables natural language questions and answers grounded in your data.
• Multi-search: Querying multiple indexes in a single request with separate or federated results.
• Filtering, sorting, and faceting: Narrowing, ordering, and categorizing results.
• Search rules: Curating results by pinning documents based on query or time-based conditions.
• Personalization: Re-ranking results based on user context and behavior.
• Analytics: Tracking searches, clicks, and conversions.
• Security and tenant tokens: Controlling access via API keys and tenant tokens for multi-tenant apps.
• Teams: Managing collaborators and roles (Meilisearch Cloud).
• Indexing: Managing documents via async task processing and foreign keys.