turbopuffer
Vector and full-text search database built on object storage
Linked from turbopuffer.com · checked 29 Oct
- Tool type
- Vector and full-text search database
- Made for
- Developers and engineering teams building search
- Runs on
- Hosted API on object storage
- Deployment modes include multi-tenant, single-tenant, or BYOC per the docs.
From @turbopuffer
Recent important posts · features
- New feature
turbopuffer v3 adopts fully vectorized query execution
“turbopuffer v3 will be the first search engine to adopt fully vectorized query execution, a design more often used by OLAP engines modern CPUs reward dumb, serial work. tpuf v3 makes the best use of the hardware for better performance on more queries”
@turbopuffer · Oct 26View
What it does
turbopuffer is a search database for AI applications that combines vector search and full-text search over documents. It stores data in object storage such as S3, with NVMe SSD and memory caching on compute, and is described as fast when cached and cheaper to run than in-memory engines.
Documents can carry vector embeddings and text attributes, and queries can combine ranking with metadata filters. Features named in its documentation include BM25 full-text search, hybrid search, regex search with trigram indexes, copy-on-write branching of namespaces, and encryption with your own key. Documented tradeoffs are higher write latency and occasional cold queries for uncached or unpinned data.
It suits engineering teams building semantic search or retrieval for AI products who can accept those latency tradeoffs. Access is through an API with a quickstart in the docs and account sign-up. Plans carry monthly minimum usage, and a cost calculator on the pricing page estimates storage, writes, and queries.
What it can do
- Vector search with 90%+ recall, combinable with filters, dense and sparse
- Full-text search with ranking and boosting
- Hybrid search combining vector and text
- Metadata filtering against an inverted index
- Regex search with trigram indexes
- Branching through copy-on-write namespaces
Before you choose
- Writes go directly to object storage, so write latency is higher (p90 about 248ms for 512KB upserts per the docs).
- Uncached or unpinned data can produce occasional cold queries (p90 about 1214ms on 1M documents per the docs).
- Per-namespace limits apply, such as 10k writes/s and 4 embedded attributes per namespace.
Listed under
Get started
Quickstart guide is in the documentation.
- Create an account via the sign-up page.
- Follow the quickstart guide in the documentation to make first API calls.
- Use the cost calculator on the pricing page to estimate storage, writes, and queries before choosing a plan.
Plans and prices
Prices in USDUsage-based billing with a monthly minimum per plan, and a cost calculator on the pricing page for estimates. Enterprise adds a 35% usage premium. Whether a free trial or permanent free tier exists should be verified on the pricing page.
- Includes all database features
- Community Slack & Email support
- HIPAA-ready BAA
- Single Sign-On (SSO)
- Audit Logs
- IP Allowlisting
- Private Slack Channel
- Single-Tenancy
- BYOC
- Private Networking
- CMEK (Per Namespace)
- 24/7 support
- 99.95% uptime SLA
Source: turbopuffer.com/pricing, checked 29 Oct 2026. Prices can change; check before you buy.
Sources for this overview
Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.
- Introduction ↗turbopuffer.com
- turbopuffer pricing ↗turbopuffer.com
- turbopuffer - fast search engine built on object storage ↗turbopuffer.com
Ratings and reviews
No ratings for turbopuffer yet.
What’s changed
Free accountSignals
Free accountData and security
Free accountCompare with similar tools
Free accountWhere this comes from
How we research- Company website
- turbopuffer.com: product, pricing and policy pages
- YC directory mirror
- Batch and company name
- Social accounts
- Accounts the company links, and its posts on X
- Human review
- Checked by a person
- knowledge_base
- manual_review
Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.