# turbopuffer

> Serverless vector and full-text search database built on object storage — fast, 10x cheaper than alternatives, and extremely scalable for AI applications.

turbopuffer is a serverless vector and full-text search database built from first principles on object storage, combining speed, cost-efficiency, and massive scalability. It uses a tiered storage engine with NVMe SSD and memory cache for hot data while keeping the rest in low-cost object storage (S3), delivering sub-10ms p50 warm query latency. Designed for AI applications, semantic search, and recommendation systems, turbopuffer handles billions of documents at a fraction of the cost of traditional vector databases. It is trusted in production at scale, handling 2.5T+ documents, 10M+ writes/s, and 10k+ queries/s.

- **Serverless Architecture**: *Sign up and start indexing immediately — no infrastructure to provision or manage, with automatic horizontal scaling.*
- **Vector Search**: *Perform approximate nearest neighbor (ANN) search with 90–100% recall@10, supporting up to 500M documents per namespace.*
- **Full-Text Search**: *Built-in full-text search capabilities alongside vector search for hybrid retrieval pipelines.*
- **Hybrid Search**: *Combine vector and full-text search in a single query to maximize retrieval quality for AI and LLM applications.*
- **Metadata Filtering**: *Filter search results by arbitrary document attributes at query time without sacrificing performance.*
- **Extreme Cost Efficiency**: *Usage-based pricing with storage, writes, and queries billed separately — up to 10x cheaper than traditional vector databases.*
- **Multi-Tenancy & Namespaces**: *Organize data into isolated namespaces with support for 100M+ namespaces in production.*
- **Security & Compliance**: *SOC2 report, GDPR-ready DPA, HIPAA-ready BAA (Scale+), SSO, CMEK, and private networking (Enterprise).*
- **Simple API**: *Interact via a straightforward HTTP API with client libraries; get started with the quickstart guide in the docs.*
- **Benchmarked Performance**: *Cold queries for 1M vectors at p90=444ms; warm queries at p50=8ms — as fast as in-memory engines when cached.*

## Features
- Serverless vector search
- Full-text search
- Hybrid search
- Metadata filtering
- Object storage backend (S3)
- NVMe SSD + memory cache
- Automatic horizontal scaling
- Multi-tenancy with namespaces
- SOC2 compliance
- GDPR-ready DPA
- HIPAA-ready BAA
- Single Sign-On (SSO)
- CMEK per namespace
- Private networking
- Usage-based pricing
- Sub-10ms warm query latency
- Billions of vectors supported
- Unlimited namespaces

## Integrations
Amazon S3, Slack

## Platforms
WINDOWS, WEB, API, JETBRAINS_PLUGIN

## Pricing
Paid

## Links
- Website: https://turbopuffer.com
- Documentation: https://turbopuffer.com/docs
- Repository: https://github.com/turbopuffer
- EveryDev.ai: https://www.everydev.ai/tools/turbopuffer