Features
- Serverless architecture with no infrastructure to manage
- BM25 and vector search in one database
- Storage-based pricing (no compute charges)
- Multi-tenant namespaces for data isolation
Pros
- Predictable, storage-based pricing model
- Zero operational overhead
- Hybrid search with BM25 built in
Cons
- Newer service with less track record
- Smaller community and fewer integrations
- Limited documentation compared to established options
Overview
Turbopuffer is a serverless vector database that combines simplicity with transparent, storage-based pricing. Unlike most vector databases that charge based on compute or queries, Turbopuffer charges primarily based on storage, making costs predictable and often lower for many workloads.
The database supports both vector similarity search and BM25 keyword search natively, enabling hybrid retrieval strategies without an additional search engine. Multi-tenant namespaces allow data isolation between customers or use cases within a single account.
Turbopuffer’s serverless architecture means there is no infrastructure to provision, scale, or maintain. The database scales automatically based on data volume and query load.
When to Use
Choose Turbopuffer when you want a simple, cost-effective serverless vector database with predictable pricing. It is ideal for multi-tenant applications and teams that want to avoid managing vector database infrastructure.
Getting Started
pip install turbopufferimport turbopuffer as tpuf
ns = tpuf.Namespace("my-namespace")
ns.upsert(
ids=[1, 2],
vectors=[[0.1, 0.2], [0.3, 0.4]],
attributes={"text": ["hello", "world"]}
)
results = ns.query(vector=[0.1, 0.2], top_k=5)