Turbopuffer
UnclaimedServerless vector and full-text search engine built on object storage for AI applications.
Visit WebsitePaidVisit Website
TL;DR - Turbopuffer
- Serverless vector and full-text search database.
- Built on object storage for 10x cost savings and extreme scalability.
- Achieves sub-10ms p50 latency for billions of vectors.
Pricing: Paid only
Best for: Enterprises & pros
Pros & Cons
Pros
- Significantly reduces operational costs compared to traditional vector databases.
- Handles massive scale with 2.5T+ documents and 10M+ writes/s in production.
- Provides extremely low latency search results, even for large datasets.
- Offers robust security and compliance features including SOC2, GDPR, and HIPAA readiness.
Cons
- Requires a minimum monthly commitment for all pricing tiers.
- Specific details on data migration tools from other databases are not explicitly detailed.
- The 'Launch' plan does not include HIPAA readiness or SSO.
Preview
Key Features
Serverless architectureAutomatic scalingLow latency (sub-10ms p50)Support for billions of vectorsFull-text searchHybrid searchMetadata filteringCost-effective (10x cheaper)
Pricing Plans
launch
$64/month
- All database features
- Multi-Tenancy
- SOC2 report, GDPR-ready DPA
- Community Slack & Email
scale
$256/month
- All database features
- Multi-Tenancy
- SOC2 report, GDPR-ready DPA
- HIPAA-ready BAA
- Single Sign-On (SSO)
- Community Slack & Email
- Private Slack Channel
- Support Hours 8-5
enterprise
Contact us
- All database features
- Multi-Tenancy
- Single-Tenancy
- BYOC
- SOC2 report, GDPR-ready DPA
- HIPAA-ready BAA
- Single Sign-On (SSO)
- CMEK (Per Namespace)
- Private Networking
- Community Slack & Email
- Private Slack Channel
- Support Hours 24/7
- Support SLA
- Uptime SLA 99.95%
What is Turbopuffer?
Turbopuffer is a serverless vector and full-text search database designed for high-performance, cost-effective, and scalable search capabilities. It is built from first principles on object storage, allowing it to handle massive datasets and high query loads efficiently. The platform is engineered to separate compute and storage, leveraging NVMe and object storage for optimal cost and performance.
This product is ideal for AI applications, semantic search, recommendation systems, and any use case requiring high-performance similarity search. It serves companies that need to connect AI with large amounts of fresh data without incurring the high costs and operational complexity of traditional vector databases. Turbopuffer is trusted by leading companies for production workloads, handling trillions of documents and millions of writes per second.
Key benefits include significant cost savings (up to 10x cheaper than alternatives), automatic scaling to support billions of vectors, and low-latency query responses. Its architecture ensures reliability and performance, making it suitable for critical production systems.
Reviews
Be the first to review Turbopuffer
Your take helps the next buyer. Verified LinkedIn reviewers get a badge.
Write a reviewBest Turbopuffer Alternatives
Top alternatives based on features, pricing, and user needs.
Explore More
Turbopuffer FAQ
How does Turbopuffer achieve 10x cost savings compared to other vector databases?
Turbopuffer achieves significant cost savings by being built from first principles on object storage. It separates compute and storage, intelligently moving data between NVMe and object storage, which optimizes resource utilization and reduces the overall infrastructure cost compared to solutions that rely on more expensive, always-on compute and storage configurations.
What is the difference between logical and physical bytes for storage billing in Turbopuffer?
Turbopuffer's pricing is based on logical bytes for vector storage. Logical storage size for vectors is calculated as the number of vectors multiplied by the vector dimension and the size of the data type (e.g., 4 bytes for float32). Full-text search attributes, vector attributes, and other attributes are billed based on their compressed logical size.
Can Turbopuffer handle hybrid search queries that combine vector similarity with full-text search and metadata filtering?
Yes, Turbopuffer supports hybrid search capabilities. This allows users to combine vector similarity search with full-text search and metadata filtering within a single query, enabling more precise and relevant search results for complex AI applications and recommendation systems.
What are the production limits for a single namespace in Turbopuffer, specifically regarding documents and write throughput?
For a single namespace, Turbopuffer supports up to 500 million documents, totaling approximately 2TB of data. The maximum write throughput for a single namespace is 10,000 writes per second, with data ingestion rates up to 32 MB per second.
How does Turbopuffer ensure data security and compliance for its enterprise customers?
For enterprise customers, Turbopuffer provides a comprehensive suite of security and compliance features. This includes a SOC2 report, GDPR-ready Data Processing Agreement (DPA), HIPAA-ready Business Associate Agreement (BAA), Single Sign-On (SSO), Customer Managed Encryption Keys (CMEK) per namespace, and Private Networking. These features ensure data protection and adherence to industry-specific regulations.
What is the typical latency for vector search queries on a warm versus cold namespace?
For a warm namespace, Turbopuffer achieves a p50 latency of 8ms and a p99 latency of 35ms for vector search queries (e.g., 768 dimensions, 1M documents). For a cold namespace, the p50 latency is 343ms and p99 latency is 554ms, demonstrating its performance even when data needs to be retrieved from underlying storage.
Source: turbopuffer.com