Vespa

Vespa, built and maintained by Vespa.ai AS, is a distributed serving engine unifying retrieval, ranking, ML inference, and real-time serving for AI applications — handling vector, text, and structured search across billions of items at sub-100ms latency and thousands of queries per second. It supports hybrid search combining vector and text retrieval, distributed machine-learned ranking, a "streaming search" mode that's up to 20x cheaper for personal/private data than full indexing, continuous zero-downtime deployment, and integrated ML inference for real-time model evaluation. It's open source and self-hostable, or available as a managed service through Vespa Cloud. Production users include Spotify, Yahoo, and Perplexity — real, checkable evidence of scale — though as infrastructure for large-scale search and ranking, it requires genuine engineering investment to operate; it's not a plug-and-play tool for a small project. Exact managed-cloud pricing isn't published on the homepage. Good fit for a team building search or recommendation systems at genuinely large scale, where a lighter-weight vector database isn't sufficient.