ankane/searchkick
Searchkick
Searchkick is an integration library and wrapper that connects application models to search engines such as Elasticsearch and OpenSearch. It functions as a search index synchronizer, automatically mirroring database records to a search server to enable full-text and vector retrieval.
The project provides a high-level interface for implementing keyword search, semantic vector search, and hybrid search. It distinguishes itself through the ability to combine traditional keyword matching with vector embeddings using reranking and fusion techniques to improve precision.
The library covers the end-to-end search lifecycle, including data synchronization via background queues, parallel reindexing, and advanced query execution. Its retrieval capabilities include geospatial filtering, search result boosting, and data aggregation. User interface enhancements are supported through autocomplete suggestions, term highlighting, and automated spelling corrections.
Features
- Elasticsearch Client Integrations - Acts as a programmatic library for connecting application models to Elasticsearch clusters for indexing and retrieval.
- Semantic Vector Search - Enables semantic vector search and nearest neighbor lookup using embeddings within the search engine.
- OpenSearch Integrations - Provides a dedicated integration library for syncing application models to OpenSearch.
- Full-Text Search Integrations - Provides a high-level library for syncing application models to search engines to enable intelligent full-text searching.
- Full Text Search - Provides a high-level interface for keyword search, boosting, and aggregations without writing complex raw queries.
- Hybrid Vector and Full-Text Search - Orchestrates hybrid search by combining traditional keyword matching with semantic vector embeddings using reranking and fusion.
- Search Index Synchronizers - Automatically maintains consistency between primary databases and external search indices using callbacks and queues.
- Cross-Model Search - Executes a single query that retrieves results from several different data models simultaneously.
- Parallel Reindexing - Speeds up the reindexing of large datasets by distributing the workload across multiple background jobs.
- Attribute Mapping - Allows developers to define exactly which model attributes and associations are sent to the search index.
- Geospatial Search - Implements geospatial filtering and sorting based on proximity, bounding boxes, and map polygons.
- Search Result Aggregators - Produces summarized counts, histograms, and range-based aggregations from search results.
- Search Result Filtering - Restricts results using equality, ranges, regular expressions, and logical operators for precise scoping.
- Search Suggestions - Supports real-time predictive search suggestions and automated spelling corrections as users type.
- Autocomplete Search Fields - Provides real-time predictive search suggestions as users type by indexing fields for rapid lookup.
- Similarity Search - Retrieves records most similar to a document using vector-based nearest neighbor search or shared fields.
- Search Result Management - Improves search quality through field boosting, synonym management, and custom ranking logic.
- Relevance Score Boosting - Increases document relevance scores and ranks based on field importance, recency, or proximity.
- Gestión de datos - Provides search functionality.
- Search - Intelligent search library for Rails.
- Search Engines - Simplified search implementation using Elasticsearch.
