A serverless vector database
built from first principles on object storage: 10-100x cheaper, usage-based pricing, and massive scalability
╔═turbopuffer══════════════════════════════╗
╔════════════╗ ║ ║░
║ ║░ ║ ┏━━━━━━━━━━━━━━━┓ ┏━━━━━━━━━━━━━━┓ ║░
║ client ║░───API──▶║ ┃ Memory/ ┃────▶┃ Object ┃ ║░
║ ║░ ║ ┃ SSD Cache ┃ ┃ Storage (S3) ┃ ║░
╚════════════╝░ ║ ┗━━━━━━━━━━━━━━━┛ ┗━━━━━━━━━━━━━━┛ ║░
░░░░░░░░░░░░░░ ║ ║░
╚══════════════════════════════════════════╝░
░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
╔════════════╗
║ ║░
┌──║ client ║░
│ ║ ║░
API ╚════════════╝░
│ ░░░░░░░░░░░░░░
└─────────┐
│
▼
╔═turbopuffer══════════════════╗
║ ║░
║ ┏━━━━━━━━━━━━━━━━━━━━━━━━┓ ║░
║ ┃ Memory/ ┃ ║░
║ ┃ SSD Cache ┃ ║░
║ ┗━━━━━━━━━━━━━━━━━━━━━━━━┛ ║░
║ │ ║░
║ ▼ ║░
║ ┏━━━━━━━━━━━━━━━━━━━┓ ║░
║ ┃Object storage (S3)┃ ║░
║ ┗━━━━━━━━━━━━━━━━━━━┛ ║░
║ ║░
╚══════════════════════════════╝░
░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░Warm query: 1m vectors
Percentile
Latency
Warm queries have all their data in cache.
Search for 100 random vectors from the dataset with top k = 10, when dataset is fully in cache. Warming the cache for an index takes about ~10s (for 1m vectors) after the first cold query, and typically stays in cache for a few hours.
Hear from our customers
Aman Sanger
Co-founder
billions
documents
1.5M+
namespaces
10×
Justin Watts
Distinguished Engineer
Moving to turbopuffer felt less like an upgrade and more like discovering a new paradigm. We didn’t just save costs; we turned once-prohibitive features into standard tools in our arsenal. Vectorize all the things and say goodbye to sharding
MTL 🇨🇦
| Metric | Max seen in production | Production limits (current) | Production limits (soon) |
|---|---|---|---|
| Max documents (global) | 8B+ | ||
| Max documents (per namespace) | 100M+ | ||
| Number of namespaces | 3.5M+ | Unlimited | |
| Max dimensions | 10,752 | ||
| Max inactive time in cache | Contact us if you need custom | ||
| Write rate (global) | 10,000 | ||
| Write rate (per namespace) | 10,000 | 10,000 doc/s | 100,000 doc/s |
| QPS (global) | >200 QPS | ||
| Max QPS (per namespace) | ~20 QPS | 100+ QPS | 10,000 QPS |
| ~90-95% | ~90-95% | Configurable | |
| See full list and workloads that are a great fit for turbopuffer 🐡 |
FAQ
