What is turbopuffer?
turbopuffer is a serverless vector and full-text search engine designed from first principles on object storage. It provides a fast, scalable and cost effective solution, making it significantly cheaper compared to traditional systems.
The system architecture integrates a Memory/SSD Cache with Object Storage (S3), enabling efficient handling of large datasets. It is capable of managing 125B+ documents and processing 6K+ queries per second, and offers both vector search and full-text search functionalities.
Features
- Vector Search: Supports high-dimensional vector search (768 dims, 1M docs).
- Full-Text Search: Implements BM25 for efficient full-text search (1M docs).
- Query Latency: Offers low latency for both warm (p50: 16ms) and cold (p50: 402ms) namespaces.
- Scalability: Capable of handling over 125 billion documents and more than 6,000 queries per second.
- Cost Efficiency: Reduces search costs by up to 10 times compared to traditional solutions.
- High Write Rate: Supports a global write rate of 200,000 documents per second.
- High Dimentions: Supports upto 10,752 dimentions.
- Namespaces: Can manage over 5 million namespaces.
Use Cases
- Large-scale document retrieval
- Semantic search applications
- Real-time data analysis
- Content recommendation systems
- E-commerce product search
FAQs
-
Is turbopuffer production quality?
Yes, turbopuffer is production quality. We have powered production applications since November '23 at 99.99% uptime. We host billions of production vectors at thousands of writes per second. We maintain SOC2 Type 2 certification and HIPAA compliance.
