turbopuffer's mission is to build the most scalable retrieval engine on earth.

Populating LLM context windows with fresh data means we're asking more from our search systems than ever before. First-hand, we've seen companies limit product ambition by cost and operational toil of incumbents.

We've designed turbopuffer from first principles:

1. Scalability, by separating compute and storage
2. Cost, by moving data between NVMe and object storage
3. Reliability, through simplicity
4. Performance, with tpuf's custom tiered storage engine

Justin and I grew up at Shopify scaling the core engine from 1K to 1M RPS. Our drive for operational simplicity is shaped by years of being on the last-resort pager for a platform powering millions of $ per minute. We are channeling all the hard-won lessons of scale into making turbopuffer the most scalable and reliable retrieval engine.

We host 35B+ documents, handle writes at 100,000 doc/s, and serve 1000+ queries/s. We are ready for far more. We hope you'll trust us with your queries.

We're backed by Lachy Groom Fund, as well as Arjun Narayan (Materialize), Erik Bernhardsson (Modal), Jessie Frazelle (Zoo & Oxide), Kenneth Auchenberg, Mikey Shulman (Suno), Richard Artoul (WarpStream), Solmaz Shahalizadeh (Shopify), Sualeh Asif (Cursor), and others.