turbopuffer is a search engine and database for vector search and full-text search, made by turbopuffer Inc. Vector search finds the stored items whose embeddings are closest to the embedding of a query; an embedding is a list of numbers that a machine learning model computes to represent meaning. Full-text search finds documents that contain the query's words and ranks them with BM25.
turbopuffer keeps all of its data in object storage such as Amazon S3 and uses NVMe SSDs and memory only as a cache. Cached queries are as fast as in-memory search engines, at a much lower running cost. The trade-offs are higher write latency, because writes go directly to object storage, and occasional slow queries on uncached data. An application writes documents into namespaces and queries them through an API, combining vector search with attribute filters, full-text, hybrid or regular-expression search. A namespace can be branched: a copy-on-write clone created in constant time.
turbopuffer is a hosted service billed by usage, with a monthly minimum that depends on the plan. Deployment is multi-tenant by default, or single-tenant, or in the customer's own cloud (BYOC).