Vespa

Vespa is a serving engine for computing over large datasets at request time, combining full-text search, approximate vector search, structured filtering, and machine-learned model inference in one horizontally scalable system. Rather than gluing a vector database to a separate ranking service, applications express retrieval and ranking together: Vespa selects candidate data, evaluates tensors and ML models over it, and returns organized results typically within 100 milliseconds, all while absorbing continuous real-time writes. The engine is written in C++ and Java, originated at Yahoo where it serves hundreds of thousands of queries per second, and is used by large services for search, recommendation, RAG, and personalization at billion-document scale. It is open source under Apache 2.0 with frequent releases from master, deployable self-hosted or on the managed Vespa Cloud. Teams building serious search or recommendation infrastructure choose it when they outgrow single-purpose vector stores and need ranking logic to run next to the data.