turbopuffer is a serverless search database for AI applications that stores vector and text indexes on cheap object storage (S3/GCS/Azure Blob) instead of keeping everything in RAM. When a query arrives, it pulls the relevant data into NVMe or RAM cache and answers in single-to-low-double-digit milliseconds. When a namespace goes idle, it costs nearly nothing. This architecture is roughly 10–100x cheaper than traditional in-memory vector databases like Pinecone, and scales to tens of millions of isolated namespaces without operators managing clusters. Cursor uses it to index over a trillion code chunks across 80 million namespaces. Notion migrated its entire 10B+ vector workload to turbopuffer in 2024 and subsequently dropped per-user AI pricing. Other customers include Anthropic, Atlassian, and Linear. The product combines vector search, full-text search, and hybrid search in one API. It was founded in 2023 by two ex-Shopify engineers — Simon Eskildsen (CEO) and Justine Li (CTO) — and is headquartered in Ottawa, Canada.