Typesense

Typesense combines typo-tolerant site search with vector and semantic search in a self-hosted engine. It's for developers building searchable apps, product catalogs or AI search over their own data. The C++ engine uses an in-memory architecture for low-latency results as users type.

Keyword search includes autocomplete, synonyms and adjustable ranking. Filters and facets help users narrow results, while geographic search supports location-based listings. Catalog teams can pin products, sort results dynamically and group similar variants. Federated search brings multiple collections into one search box.

For semantic search, Typesense can index your own embeddings or generate them with built-in models such as S-BERT and E-5. Hybrid search combines meaning with keyword matching. Built-in RAG produces conversational answers based on indexed data, and natural language search turns free-form requests into filters and sorting rules. CLIP supports image search by text or similarity; Whisper transcribes voice queries.

You can run Typesense on Linux or macOS, including through Docker, under the GPL-3.0 open-source license. Self-hosting keeps the search engine and index on infrastructure you control. Typesense Cloud provides managed hosting instead. Optional external embedding services, including OpenAI, receive the queries and indexed text they process.

Scoped API keys restrict access to records in shared applications, and replication supports high availability. Typesense has a REST API, JavaScript and Python clients, and integrations with WordPress, WooCommerce, Laravel and InstantSearch.