vespa

Why we included this project

Vespa is a serving engine for search and machine-learned inference, not just a vector index. You can store embeddings, tensors, plain text, and structured data in one corpus, and a single query can mix dense semantic retrieval with keyword filters. At serving time it runs machine-learned ranking expressions, including ONNX models, over the candidates, then groups and aggregates results, typically returning them in well under 100 milliseconds while the data keeps changing. Because the data is spread across nodes and rebalanced automatically, it scales to the query volumes of large internet products, and you deploy it as one application package that can be updated without dropping traffic. If you are currently gluing a vector store to a separate ranking service, Vespa already has the indexing, query APIs, and failover wiring in place.