Turbopuffer
Serverless vector database on object storage, priced for cold data.
About
Turbopuffer is a serverless vector database that separates storage from compute: vector partitions live in object storage and are paged in on demand, so inactive data costs near-zero instead of RAM residency. It became known when Cursor adopted it as its vector DB and reported its first invoice 95% lower than the vendor it replaced; Notion and Anthropic are also customers.
Pricing starts at a $64/month Launch minimum with usage on top (queries billed per PB of data scanned); there is no free tier and no open-source version.
Honest limits: cold queries add 300-800ms while partitions load from object storage, so it is the wrong tool for interactive chat RAG that needs sub-100ms p95 — it shines on large corpora with moderate query rates.
What it can do
- Serverless vector search on object storage
- Attribute filtering
- Namespaces for multi-tenant isolation
- Pay-per-query pricing model
- Python and TypeScript SDKs
Setup
- Sign up at turbopuffer.com and create an API key.
- Install the SDK: npm install @turbopuffer/turbopuffer or pip install turbopuffer.
- Create a namespace: tpuf.namespace("docs").
- Upsert vectors with attributes, then query with filters and top_k.
Prerequisites
Turbopuffer account (paid, $64/mo minimum); embeddings from any model.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Cursor's vector DB — first invoice 95% below prior vendor; No free tier; $64/mo Launch minimum; Cold-query latency 300-800ms — wrong tool for sub-100ms chat RAG. License: Proprietary, closed source.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 2.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Cursor's vector DB — first invoice 95% below prior vendor
- · No free tier; $64/mo Launch minimum
- · Cold-query latency 300-800ms — wrong tool for sub-100ms chat RAG
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.