Cloud & Architecture · Paid

turbopuffer

turbopuffer is a serverless vector and full-text search database built on object storage. Instead of keeping every index in memory, it stores data in cheap object storage and caches the parts that are actually queried. That design keeps storage costs low, which suits workloads with very large numbers of documents or many separate namespaces, such as per-user memory for AI agents and multi-tenant search products. The database supports vector search, full-text search and filtering on attributes, and organizes data into namespaces. Teams can run it as multi-tenant cloud service, and higher plans add single-tenant deployments and Bring Your Own Cloud. Security features on paid tiers include a SOC 2 report, a GDPR-ready DPA, a HIPAA-ready BAA, single sign-on, audit logs, IP allowlisting, private networking and per-namespace customer-managed encryption keys on Enterprise. Pricing as of Nov 4, 2026, from turbopuffer's official pricing page: there is no free tier, because the company has chosen a commercial-only model. The Launch plan has a $16 monthly minimum, and usage above that is billed on logical bytes. Storage costs up to $0.33 per GB per month, writes up to $2.00 per GB and queries up to $1.00 per PB of data queried, plus a charge for data returned. Enterprise starts at $4,096 a month with a 35% usage premium, 24/7 support and a 99.95% uptime SLA. Two billing rules matter for agent builders. Every write is billed at a minimum of 10 KB, so many tiny writes cost more than their size suggests; batched writes earn a discount that reaches 50% at about 3.2 MB per batch. Queries are billed on the size of the namespace queried, with a floor of 1.28 GB, and large namespaces receive steep marginal discounts above 32 GB and 128 GB. Attributes marked as not filterable are billed at half price. turbopuffer's own calculator estimates about $7 of usage for 1 million 1,024-dimension documents with 1 million writes and 1 million queries a month, below the Launch minimum. Metarix covers turbopuffer in its comparisons of vector databases for AI agents.