Rivestack

Rivestack is a French-based managed PostgreSQL service purpose-built for AI and vector search workloads. It ships pgvector pre-enabled and tuned with HNSW indexing on NVMe storage, targeting teams that need fast, predictable vector search without the overhead of general-purpose database platforms. The service is available in EU (France) and US-East regions, with a free tier that requires no credit card.

What It Is

Rivestack provides dedicated PostgreSQL nodes with pgvector configured out of the box, positioned as a focused alternative to broader platforms like Supabase and Neon, and to standalone vector databases like Pinecone. Rather than bundling auth, storage, edge functions, or realtime features, Rivestack concentrates on one job: running Postgres and pgvector fast and cheaply on NVMe hardware. The service is standard PostgreSQL under the hood, so any existing Postgres driver — Python's psycopg2, Go, Node.js, Java, Rust, Ruby, .NET — connects without modification.

Performance and Infrastructure

Rivestack publishes benchmark figures on its homepage: the site claims 2,150 QPS and 2.8ms p50 latency on a base node for a 1M × 1536-dimension dataset with ef_search=40 and 16 clients, compared to approximately 410 QPS and 18ms on an equivalent Supabase configuration. Storage uses NVMe rather than cloud SSD, which the company attributes as the primary driver of the latency difference. Full benchmark methodology is documented in a linked blog post. Each cluster runs Patroni for high availability, pgBackRest for backups, and ships with Prometheus and Grafana monitoring. PostgreSQL 18 is available on all clusters, and pgvector 0.8.x is the current extension version.

Deployment and Operations

Clusters are provisioned and managed via a Terraform provider, allowing infrastructure-as-code workflows without dashboard interaction. High availability is achieved by adding nodes to a cluster rather than through a separate HA product tier. Backups are daily with 14-day point-in-time recovery (PITR) on paid plans. The service is hosted entirely within the EU for the EU region, and the company states it is a French entity with GDPR data processing agreements available on request.

Migration Path: Vector Rescue

Rivestack offers a free "Vector Rescue" service for teams migrating from Supabase, Neon, Pinecone, or self-hosted pgvector. Users submit their row counts, vector dimensions, QPS targets, and current pain points via email; the company responds within 48 hours with a plan recommendation, expected performance figures, a side-by-side cost comparison, and a migration path (pg_dump, logical replication, or cutover plan). The company states that if Rivestack is not a better fit, it will say so in the same reply.

Update: PostgreSQL 18 and SQL Editor

The homepage footer changelog notes two recent additions: PostgreSQL 18 is now available on all clusters, and a SQL Editor with embeddings search has been added to the product. These represent the most recent visible product updates at the time of this writing.

Why It Matters for AI Workloads

The core proposition is that vector workloads — particularly RAG pipelines and semantic search — generate disproportionate cost and latency on platforms priced for general-purpose use. Rivestack's fixed per-node monthly pricing eliminates per-query billing, egress surcharges, and vector-count overages, making the monthly bill predictable from day one. The live demo at ask.rivestack.io runs semantic search over 30 days of Hacker News data on a real Rivestack cluster, with the site claiming sub-50ms query times at no additional infrastructure cost.