Turbopuffer
Serverless vector and full-text search database built on object storage, serving AI-native companies and enterprises building RAG, semantic search, recommendation, and agent applications at sub-10ms latency and 10x lower cost.
- Company typePrivate
- Founded2023
- HeadquartersOttawa, Canada
- Headcount11–50
- GTM typeB2B
- OfferingSoftware
What Turbopuffer does
Turbopuffer is a serverless vector and full-text search database built from first principles on object storage (S3), targeting AI and machine-learning teams that need cost-efficient similarity and hybrid retrieval at scale. Founded in 2023 in Ottawa by former Shopify engineers, the company offers a single REST API product combining approximate nearest neighbor (ANN) vector search, BM25 full-text search, and hybrid retrieval via reciprocal rank fusion, with optional namespace metadata and filtering. The core technical differentiators are a proprietary SPFresh-inspired clustering ANN index that supports incremental inserts and deletes without full rebuilds, attribute indexes that are aware of the vector clustering hierarchy to enable high-recall filtered queries, a multi-tier cache hierarchy (memory, NVMe SSD, S3) that uses object storage as the source of truth, and a continuous production recall-measurement system that samples 1% of live traffic to enforce a 90-100% recall@10 SLA. The company reports production scale of 4 trillion-plus documents, 10 million-plus writes per second, and 25 thousand-plus queries per second across 12 AWS regions and 9 GCP regions.
Turbopuffer monetizes via a usage-based pricing model that bills on logical bytes stored, writes, and queries, with batch write discounts of up to 50%, copy discounts of up to 75%, and flat-rate namespace branching at $0.032 per operation. Above the self-serve usage tier, Scale and Enterprise plans add CMEK encryption (AWS KMS / GCP KMS), audit log streaming, private networking via AWS PrivateLink and GCP Private Service Connect, and dedicated support. Distribution is API-first with self-serve signup and Python/Go SDKs for developers, augmented by direct sales for enterprise customers. Its customer base spans AI-native builders (Cursor, Cognition, Harvey, Granola, Clay, Superhuman, Legora) and large enterprises (Notion, Anthropic, Atlassian, Bridgewater Associates, Ramp, Grammarly), and the company has positioned its core value proposition as a roughly 10x cost reduction versus incumbent vector databases, achieved by separating durable state onto object storage and operating stateless compute nodes.
Turbopuffer firmographics
Firmographics- Name
- Turbopuffer
- Legal name
- turbopuffer Inc.
- Website
- https://turbopuffer.com
- Company type
- Private
- Founded year
- 2023
- Operating status
- Operating
- Headcount range
- 11–50 employees
- Short description
- Serverless vector and full-text search database built on object storage, serving AI-native companies and enterprises building RAG, semantic search, recommendation, and agent applications at sub-10ms latency and 10x lower cost.
- Ownership category
- akta.pro rank
Turbopuffer industry classification
Industry- Product category
- Vector Database
- NAICS
- Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (5182), Web Search Portals and All Other Information Services (51929), Web Search Portals and All Other Information Services (519290)
- SIC
- Services-Computer Processing & Data Preparation (7374)
- akta.pro primary industry
- Retrieval-Augmented Generation (RAG), Vector Databases & Knowledge Grounding (HDAAACAE)
- akta.pro secondary industries
- Enterprise AI Data & Knowledge Platforms (Vector Databases, Knowledge Graphs) (HDAEANAH), Search, Retrieval & Semantic Ranking (BM25/vector, hybrid) (HDAAADAB)
Keywords
Where Turbopuffer is headquartered
LocationHeadquarters
- HQ city
- Ottawa
- HQ country
- Canada
- HQ region
- North America
Offices1 record
Markets served
Turbopuffer business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Technology or R&D, Infrastructure, Personnel, Operations, Marketing or Sales
Revenue model
- Usage-based Vector Database Service: Usage-based pricing model based on logical bytes stored, queries, and writes. Supports pay-as-you-go consumption with batch write discounts up to 50%. Namespace branching billed at flat rate of $0.032.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Unit Pricing | Pay-as-you-go | Namespace Branching - Flat rate per operation |
| Subscription | Annual | Scale Plan |
| Subscription | Annual | Enterprise Plan |
Go-to-market motion2 records
Distribution channels3 records
Marketing channels5 records
Turbopuffer product offering
Product offeringCore offering
Turbopuffer provides a serverless vector and full-text search database built on object storage. The service delivers approximate nearest neighbor vector search, BM25 full-text search, and hybrid search via reciprocal rank fusion, with sub-10ms p50 latency and 90–100% recall@10. It is targeted at AI applications including retrieval-augmented generation (RAG), semantic search, recommendation systems, and AI agents, and is positioned as roughly 10x cheaper than alternative vector databases.
Product overview
Turbopuffer is a unified vector and full-text search database offered as a single core product. The platform combines approximate nearest neighbor (ANN) vector search with BM25 full-text search capabilities, supporting hybrid search through reciprocal rank fusion. Key product features include namespace branching for instant copy-on-write clones, native filtering for high-recall filtered queries, and continuous recall measurement. The service is delivered via REST API with official Python and Go SDKs, supporting deployment across multiple cloud regions with optional enterprise features including CMEK encryption, private networking, and audit logging.
Differentiator
Problem solved
Functional benefit
Products and services
- Turbopuffer Fast, serverless vector and full-text search database built on object storage, providing semantic vector search (ANN/kNN), BM25 full-text search, and hybrid search via reciprocal rank fusion. Designed for AI applications such as RAG, semantic search, recommendation systems, and AI agents, with sub-10ms p50 latency, 90–100% recall@10, and usage-based pricing for developers and enterprises.
Quantifiable outcome
- 10x cost reduction compared to traditional vector databases
- +3 more outcomes
Companies that use Turbopuffer
Customer profileNamed customers14 records
Segments2 records
Ideal customer profiles2 records
Turbopuffer technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
Integration9 records
AI capability5 records
Feature7 records
Turbopuffer partnerships and signals
Strategic signalScale indicators7 records
Recent moves3 records
Expansion highlights6 records
Turbopuffer competitors and assessment
Company assessmentDirect peers
- Pinecone: Managed vector database purpose-built for production AI retrieval, offering ANN search, metadata filtering, and namespaces. Closest direct competitor to Turbopuffer in the serverless managed vector store category targeting AI/ML builders.
- Qdrant: Open-source and managed vector database with filtering, payload indexing, and Rust-based engine. Directly comparable to Turbopuffer on the core vector search product, with both an open-source and a hosted offering that compete for the same AI workloads.
- Weaviate: Open-source vector database with hybrid sparse-dense retrieval and modules for generative search. Competes with Turbopuffer for the same AI retrieval and RAG workloads, with overlapping enterprise and developer mindshare.
- Milvus: Open-source vector database built for trillion-scale similarity search under the LF AI & Data umbrella. Comparable architecture goals (scale, filtered recall) and audience (AI engineers, enterprise platform teams) as Turbopuffer.
- Chroma: Vector database popular with Python and LLM developers, originally focused on embeddings storage and now expanding into managed cloud. Targets the same AI/ML builder segment as Turbopuffer, especially for RAG prototypes.
Emerging players
- LanceDB: Open-source columnar vector database built on the Lance format for fast random access over large embedding datasets. An emerging alternative in the same AI retrieval category, often evaluated alongside Turbopuffer for cost and scale.
Broad incumbents
- Vespa.ai: Yahoo-spun platform combining vector search, lexical search, and ranking at large scale. An established incumbent in the same hybrid retrieval space Turbopuffer targets with ANN plus BM25 plus RRF.
- Elastic (Elasticsearch / OpenSearch): General-purpose search and analytics engine with native vector search and hybrid retrieval. Competes with Turbopuffer when enterprises consolidate retrieval into existing search platforms rather than adopting a dedicated vector store.
- Databricks (Mosaic AI Vector Search): Databricks offers a managed vector search product integrated with its lakehouse and ML platform. An incumbent that competes with Turbopuffer for enterprise AI retrieval, especially where data already lives in Databricks.
- Redis (RedisVL / vector search): Redis has expanded from an in-memory data store to include vector search capabilities and modules for AI retrieval. An incumbent that competes with Turbopuffer when customers prefer to extend an existing Redis deployment rather than adopt a dedicated vector database.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat6 records
Key risks6 records
Key highlights7 records
Customer concentration
Turbopuffer social profiles
Digital presenceTurbopuffer financial estimates
Financial estimateRevenue estimate
Valuation estimate
Turbopuffer leadership team
Management profileNumber of profiles
Profiles5 records
Turbopuffer funding detail
Funding detailFunding overview
Funding rounds1 record
Investors1 record
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Turbopuffer M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Turbopuffer
What does Turbopuffer do?
Turbopuffer provides a serverless vector and full-text search database built on object storage. The service delivers approximate nearest neighbor vector search, BM25 full-text search, and hybrid search via reciprocal rank fusion, with sub-10ms p50 latency and 90–100% recall@10. It is targeted at AI applications including retrieval-augmented generation (RAG), semantic search, recommendation systems, and AI agents, and is positioned as roughly 10x cheaper than alternative vector databases.
Is Turbopuffer a public or private company?
Turbopuffer is a private company. It is classified as venture growth investor backed and is currently operating.
When was Turbopuffer founded?
Turbopuffer was founded in 2023. It employs 11 to 50 people.
Where is Turbopuffer based?
Turbopuffer is headquartered in Ottawa, Canada, in the North America region.
How does Turbopuffer make money?
One revenue line is on record: usage-based Vector Database Service.
Who are Turbopuffer's main competitors?
Direct peers on record are Pinecone, Qdrant, Weaviate, Milvus and Chroma. LanceDB is listed as an emerging player. Broad incumbents are Vespa.ai, Elastic (Elasticsearch / OpenSearch), Databricks (Mosaic AI Vector Search) and Redis (RedisVL / vector search).
Does Turbopuffer have an API?
Yes. REST API with v1 and v2 endpoints. Uses Bearer token authentication with API keys formatted as standard Bearer tokens passed in the Authorization header. JSON encoding for request and response payloads. Supports HTTP compression headers, though official clients disable compression by default. OpenAPI specification publicly available at https://github.com/turbopuffer/turbopuffer-openapi. Official client libraries available in Python and Go. Supports asynchronous requests for long-running operations. Developer documentation is at turbopuffer.com/docs/api-overview.
What industry is Turbopuffer in?
Turbopuffer's product category is Vector Database. Its primary akta.pro industry code is HDAAACAE, Retrieval-Augmented Generation (RAG), Vector Databases & Knowledge Grounding, with a secondary code of HDAEANAH, Enterprise AI Data & Knowledge Platforms (Vector Databases, Knowledge Graphs). Its NAICS code is 5182 and its SIC code is 7374.