AI search at scale
Engine combining vector search, keyword search and ranking over billions of documents in one system.
Vespa is the search and AI platform built for scale at Yahoo: a single engine for lexical search, vector search and machine-learned ranking, storing documents with their tensors and running ranking expressions at query time over billions of items. It is the search engine behind large-scale recommendation and discovery systems.
Why founders use it: the ambitious search product needs ranking that combines keywords, semantics and business signals in one query, and Vespa's model lets you express that as configuration rather than a pipeline of separate services. Real-time indexing means inventory, pricing or freshness changes are searchable immediately, which matters for marketplaces and feeds.
A realistic use: a marketplace search where relevance blends text match, embedding similarity and conversion signals, with per-tenant ranking profiles. A recommendation surface reusing the same document store and ranking framework.
Trade-offs: operational weight is real, Vespa is a cluster to run and tune, and the learning curve is steep compared with managed search APIs. Compare with Algolia for hosted simplicity, Elasticsearch for the familiar incumbent, and Qdrant when only vector retrieval is needed.
Who it's for
Engineering teams building search or recommendation backends that outgrow a standalone vector database.
Official links
Keep exploring
Concepts in the glossary
3 in catalog
Vespa features
2 in catalog
Vespa is an alternative to
1 in catalog