Self-hosted alternatives to hybrid-search-implementation

H

hybrid-search-implementation

Combine vector and keyword search for improved retrieval

hybrid-search-implementation is an agent skill from the wshobson/agents collection for combining vector search with keyword search to improve retrieval. Vector search finds passages that mean the same thing as a query even when the words differ, while keyword search finds exact matches for names, codes and specific terms. Each approach misses things the other catches, so the skill covers architectures that run both and merge the results. It applies when implementing RAG systems that need better recall, building search engines, combining semantic understanding with exact matching, handling queries that contain specific terms such as product names or identifiers, improving search for domain-specific vocabulary that embeddings may not capture well, and cases where pure vector search misses matches that a keyword search would find. The skill starts with the core architecture of hybrid search, which shows how the two result lists are produced and combined, and it is aimed at engineers building retrieval for chatbots, document search and knowledge bases. The guidance is a pattern to adapt to your own index and data rather than a finished search engine, so the team still needs to test it on real queries. The skill is installed with the skills CLI from the wshobson/agents repository. Its skills.sh page showed about 10.5K installs.