turbopuffer
turbopuffer.com- Category
- Developer Tools
- Rank
- No. 593Tools index
Previous survey · No. 588 ·
- Platform
- web
- Type
- TOOL
- Builder
- @turbopuffer
- Date
About
turbopuffer is a vector and full-text search database built directly on object storage (like S3), offering sub-10ms query latencies while being roughly 10x cheaper than traditional in-memory vector databases. It supports hybrid vector plus BM25 full-text search with metadata filtering, and is used in production by companies like Cursor, Notion, Anthropic, and Linear at scales exceeding 4 trillion documents and 10 million writes per second.
What it does
This is a database built to search huge collections of documents by meaning or by keyword, storing everything on cheaper cloud storage instead of memory. It combines similarity-based lookups with traditional keyword matching, adds filters to narrow results, and aims to keep working reliably as stored data grows very large.
Stated on the product site
- Architecture
- The vendor describes the product as a search engine built on object storage rather than in-memory infrastructure.
- Search capabilities
- The site lists separate vector, full-text (BM25), and hybrid search modes as core guides.
- Interfaces
- Documentation is organized around API operations for writing and querying documents.
- Data organization
- Data is organized into namespaces, and a single namespace can be sharded to scale further.
- Stated capacity limits
- The vendor publishes a documented ceiling of 128 billion documents per namespace at up to 256TB.
Not stated on the site
- The page does not specify which programming languages or client SDKs are officially supported.
- The page does not state pricing amounts or how costs are calculated beyond a general cost comparison claim.
- The page does not describe which cloud regions or geographic locations the service is deployed in.
Written from the product site at turbopuffer.com.
What it can do
Perform vector similarity search
Vector embeddings/query → Ranked search results
Perform full-text search using BM25 ranking
Text query → Ranked search results
Combine vector and full-text search into hybrid queries
Vector and text query → Combined ranked search results
Filter search results by metadata
Metadata filter criteria → Filtered search results
Intel on turbopuffer
Tags
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.