Skip to navigation
Getting Started

Introduction

turbopuffer is a fast search engine that combines vector and full-text search using object storage.

What is turbopuffer?

turbopuffer is a fast, object-storage native search engine optimized for first-stage retrieval. It combines vector search (ANN/kNN), full-text search (BM25), and exact filtering in a single API — built to run efficiently on top of object storage like S3 or GCS.

Key features

Vector Search

Approximate nearest neighbor (ANN) and exact kNN search over high-dimensional embeddings.

Full-Text Search

BM25-based full-text search with configurable tokenization, stemming, and language support.

Hybrid Search

Combine vector and full-text queries in a single request with multi-query and RRF reranking.

Object Storage Native

Built on object storage for virtually unlimited scale at low cost — no warm-up required.

Core concepts

Namespaces are the primary organizational unit in turbopuffer. Each namespace holds a collection of documents, each with an id, an optional vector embedding, and arbitrary typed attributes.

Documents consist of:

  • id — a string, UUID, or integer identifier
  • vector — a dense float array (optional)
  • attributes — typed fields (string, int, float, bool, datetime, and array variants)

Schemas define the type and indexing configuration for each attribute — whether it’s filterable, full-text searchable, ANN-indexed, or auto-embedded.

Authentication

All API requests require a Bearer token. Set your API key as the Authorization header:

Authorization: Bearer <your-api-key>

Get your API key from the turbopuffer dashboard.

Regions

turbopuffer is available in multiple regions. Specify your region in the base URL:

https://{region}.turbopuffer.com

For example: https://gcp-us-central1.turbopuffer.com