turbopuffer
Free tierFast vector and full-text search engine built on object storage — 10x cheaper and infinitely scalable
Free tier available·Technical·API available
Key strengths
10x cheaper than traditional vector databases due to object storage architectureSub-10ms p50 latency with Memory/SSD caching layerMassive scale: 4T+ documents, 10M+ writes/s, 25k+ queries/s in productionHybrid search combining vector (ANN) and full-text (BM25) in one systemUnlimited namespaces with instant copy-on-write branching
Free tier + paid plans
San Francisco, USA
Founded 2023
No ratings yet
- RAG (Retrieval-Augmented Generation) pipelines — Store and query document embeddings at scale to ground LLM responses with relevant context.
- Semantic / ANN vector search — Perform approximate nearest neighbor search over billions of high-dimensional vectors (e.g., 1024-dim) with tunable recall.
- BM25 full-text search — Run keyword-based search with BM25 ranking over large corpora stored in turbopuffer namespaces.
- Hybrid search — Combine vector similarity scores and BM25 relevance scores with custom re-ranking for best-of-both retrieval.
- Multi-tenant SaaS search — Use unlimited namespaces to isolate per-tenant data; branch namespaces for staging/testing environments via copy-on-write.
- Recommendation systems — Query similar items by embedding distance with metadata filters to personalise results by user attributes or categories.