turbopuffer is a proprietary, managed search database for first-stage retrieval. It supports approximate nearest-neighbor vector search, BM25 full-text search, hybrid search, metadata filtering, regular-expression search, and copy-on-write namespace branching. The DBMS uses object storage (e.g., Amazon S3) as the system of record, and relies on multi-level caching using memory and NVMe SSD.
- Website
- https://turbopuffer.com[01]
- Tech Docs
- https://turbopuffer.com/docs[02]
- Developer
- Country of Origin
- CA
- Start Year
- 2023 [06]
- Project Type
- Commercial
- Written in
- Rust
- Operating System
- Hosted
- License
- Proprietary
- @turbopuffer[04]
turbopuffer is a proprietary, managed search database for first-stage retrieval. It supports approximate nearest-neighbor vector search, BM25 full-text search, hybrid search, metadata filtering, regular-expression search, and copy-on-write namespace branching. The DBMS uses object storage (e.g., Amazon S3) as the system of record, and relies on multi-level caching using memory and NVMe SSD.
History[05]
turbopuffer Inc. began developing turbopuffer in 2023. A July 2024 company post described the system as having been in production for several months and documented its object-storage architecture and vector-search design. Subsequent milestones listed by the company include continuous recall measurement in September 2024, native filtering for vector search in January 2025, and a second version of its full-text-search implementation in December 2025.
Citations
6 sources- turbopuffer - fast search engine built on object storage turbopuffer.com
- Introduction turbopuffer.com
- turbopuffer blog turbopuffer.com
- https://twitter.com/turbopuffer twitter.com
- https://www.theglobeandmail.com/business/article-turbopuffer-fastest-growing-canadian-ai-startup-never-heard-of theglobeandmail.com
- https://www.linkedin.com/posts/sirupsen_turbopuffer-o-activity-7115396121253818370-bGPc linkedin.com