The AI database for AI · MCP built in

Postgres for AI apps.
The same database your agents query.

Dedicated PostgreSQL with pgvector and pgvectorscale, plus a hosted MCP server, so Claude, Cursor and your own agents can query, inspect and manage it. Always on, local NVMe, PITR included, and a flat monthly price.

7 days free, then $29/mo · card required

Or start at $9/mo, Hobby (EU)

agent accessHosted MCP
storageLocal NVMe
recoveryPITR included
query & vector fees$0
Built on foundations you knowPostgreSQLpgvectorpgvectorscaleMCPTerraform
01 · Less to managededicated node

One database for the app and the agent.

Vectors, application data, filters, and joins. All the flexibility of Postgres, with pgvector ready to go.

01.1

Your whole app, in one database.

Keep vectors close to the rows that give them context. Build semantic search, recommendations, and RAG with the SQL you already know.

01.2

Dedicated. Always on.

Your own compute and local NVMe. Compute stays active, so your first query never waits for a database to wake up.

01.3

The operations are on us.

Automated backups, recovery to any point in time, and monitoring. Spend your time on what you're building.

01.4

A predictable bill.

Fixed monthly pricing per node, with no charges for queries, egress, or vector count. Resizing or adding a node changes the bill; query traffic does not.

02 · Built for AI agents

Give your agent the database.

Rivestack runs a hosted MCP server at mcp.rivestack.io/mcp. Add it to Claude Code, Cursor or any MCP client that supports remote servers, and your agent can list databases, read schemas, run SQL and search vectors on your Rivestack Postgres.

  • Read-only by defaultSQL runs inside a READ ONLY transaction. Tools that write appear only when you set read_only=false.
  • Vector tools built inCreate pgvector tables, build HNSW or DiskANN indexes, run similarity search and read index stats.
  • Provision with a previewcreate_database returns the price first. Nothing runs until the agent confirms.
  • Same handlers as the dashboardMigrations and index builds land in the cluster activity log, next to backups and restores.
// Cursor · .cursor/mcp.json
{
  "mcpServers": {
    "rivestack": {
      "url": "https://mcp.rivestack.io/mcp?read_only=true"
    }
  }
}

# Claude Code
claude mcp add --transport http rivestack \
  "https://mcp.rivestack.io/mcp?read_only=true"
oauth sign-in or rsk_ api keyread-only by default
› "Which docs are closest to 'refund policy'?"

→ list_databases                       ok
→ describe_table                       ok
    { table: "documents" }
→ similarity_search                    ok
    { table: "documents",
      query: "refund policy", limit: 3 }

← 1  Refunds and cancellations
  2  Billing FAQ
  3  Changing your plan

› read_only=true · nothing was written
real tool namesexample session · example data
-- nearest neighbours, filtered like any SQL
SELECT title, embedding <=> $1 AS distance
FROM   documents
WHERE  category = 'guide'
ORDER BY distance
LIMIT  3;
titledistance
Getting started with RAG0.042
Choosing an embedding model0.068
Building your search pipeline0.110
3 rowsexample data
# copy it from the Connection tab in the dashboard
psql "postgresql://username:password@\
rs-abc12345.eu.db.rivestack.io:6432/mydb\
?sslmode=verify-full"

# dedicated plans: port 6432, sslmode=verify-full
# Hobby: port 5432, sslmode=require
# any Postgres driver works: Python, Node.js, Go, psql
tls onlyexample host
resource "rivestack_cluster" "production" {
  name               = "production-cluster"
  region             = "eu-central"
  server_type        = "growth"
  node_count         = 2
  postgresql_version = 18
}

resource "rivestack_cluster_database" "analytics" {
  cluster_id = rivestack_cluster.production.id
  name       = "analytics"
  owner      = rivestack_cluster.production.db_user
}
source = "rivestack/rivestack"from the docs
03 · Measured on real workloadshnsw · nvme

QPS with recall. Not marketing.

Measured pgvector throughput, latency, and recall by node and workload. Reproduce the results with the open source pgvector-bench CLI.

3.8ms

p50 on Solo, 4 clients, recall@10 0.93, ef_search=80, 250k × 1536.

planclientsrecall@10QPSp50p95
Solo40.939833.8 ms6.6 ms
Starter40.931,1853.2 ms4.8 ms
Growth160.942,9495.1 ms8.7 ms
Scale160.954,4653.4 ms5.3 ms

250k × 1536 vectors · ef_search=80 · HNSW m=16, ef_construction=64 · cosine · same-region client via TLS · EU Falkenstein · PostgreSQL 17.10 · pgvector 0.8.2 · synthetic clustered vectors. Full table and method on the benchmarks page.

04 · How it worksthree steps

From sign-up to connection string.

step 1

Sign up.

Create your account with email, GitHub or Google.

app.rivestack.io/signup
step 2

Create a database.

Pick a name, a plan and a region. A dedicated Solo VM provisions in about 3 minutes; HA clusters typically take 7–10.

Solo · eu-central · 7 days free
step 3

Connect.

Copy the connection string from the Connection tab and use it from any Postgres client.

postgresql://…:6432/mydb?sslmode=verify-full
05 · Simple by designeu central · usd

Dedicated Postgres. A predictable bill.

Fixed monthly pricing per node. No fees for queries or vector count. EU Central prices shown.

Hobby
shared · EU

One database on a shared Postgres host.

$9/mo
  • EU Central
  • pgvector and pgvectorscale
  • vector 0.8.7 · vectorscale 0.9.0
  • Shared CPU · 1 database
  • 5 GB storage
  • 20 connections
  • Daily backups (14-day retention)
  • PITR
  • HA
Choose Hobby
Start here
Solo
dedicated

Entry plan. Always-on single node, no failover.

$29/mo
  • 1 vCPU · 2 GB RAM
  • 30 GB usable NVMe (40 GB disk)
  • Daily backups + 7d PITR
  • 7 days free · Card required
  • Founder assisted pilot available
Try Solo free
Starter
dedicated · HA-ready

Production tier. HA-ready: add a second node for automatic failover.

$49/node/mo
  • 2 vCPU · 4 GB RAM
  • 55 GB usable NVMe (80 GB disk)
  • Daily backups + 14d PITR
  • Add nodes for automatic failover
  • Monitoring + Terraform
Choose Starter
Growth
dedicated · HA-ready

More compute and memory for growing workloads.

$85/node/mo
  • 4 vCPU · 8 GB RAM
  • 135 GB usable NVMe (160 GB disk)
  • Daily backups + 14d PITR
  • Add nodes for automatic failover
  • Priority support
Choose Growth
Scale
dedicated · HA-ready

The capacity for larger datasets and busy applications.

from $159/node/mo
  • 8 vCPU · 16 GB RAM
  • 295 GB usable NVMe (320 GB disk)
  • Daily backups + 14d PITR
  • Add nodes for automatic failover
  • Priority support
Choose Scale

Solo trial: 7 days free for new customers, then $29/month. Card required. Cancel anytime. US East checkout is Solo $29, Starter $59, Growth $119, Scale $229, and US East Scale disk is 240 GB. Singapore is Solo at $29.

06 · Regionsyou choose where it runs

Close to your users. Built in France.

Rivestack is a French company. Selecting EU Central keeps provisioned database nodes in Germany; US East and Singapore are separate choices.

eu-central

EU Central

Falkenstein, Germany · lowest prices
HobbySoloStarterGrowthScale
us-east

US East

Virginia
HobbySoloStarterGrowthScale
singapore

Singapore

Asia Pacific · Solo only
HobbySoloStarterGrowthScale
07 · A human when you need one

Considering a move? Let's find the right fit.

Share what you're building. An engineer will help you work through capacity, cost, and a realistic migration plan.

  • A plan for your data and traffic
  • An honest cost comparison
  • A clear path to migration
08 · Good questions

A few things worth knowing.

Anything else? Email support@rivestack.io and an engineer who runs the platform replies.

Will Rivestack actually be cheaper than Supabase for pgvector?

Rivestack Starter is $49/month in EU Central for a 2 vCPU / 4 GB node. Supabase prices compute, storage, and egress separately and includes a broader platform, so compare against the features you actually use and send us the workload if you want an honest cost check.

Can my AI agent use Rivestack?

Yes. Rivestack runs a hosted MCP server at https://mcp.rivestack.io/mcp. Add it to Claude Code, Cursor or any MCP client that supports remote servers, then sign in with OAuth or send an rsk_ API key as a Bearer token. Connections are read-only by default: SQL runs inside a READ ONLY transaction, and tools that write or spend money, such as apply_migration and create_database, only appear with read_only=false. create_database returns a price preview and needs a second, confirmed call.

Does Rivestack avoid cold starts?

Yes. Dedicated Rivestack plans keep PostgreSQL running with local NVMe. Compute stays active, so your first query never waits for a database to wake up.

Can pgvector really replace Pinecone?

For many workloads, yes: HNSW on tuned PostgreSQL keeps vectors beside relational data and supports SQL filters, joins, and transactions. The right answer depends on vector count, dimensions, filters, QPS, and latency targets, which is why the workload review is free.

How does migration work?

The dashboard migration workflow uses PostgreSQL dump and restore. Duration and write downtime depend on database size, network throughput, extensions, and write activity. Test the migration first and plan a maintenance window; Rivestack does not promise a cutover without downtime when the workload cannot support one.

Where is my data hosted?

Rivestack is a French company. Selecting EU Central keeps provisioned database nodes in Germany; US East and Singapore are separate choices. Data residency is not itself a compliance certification, so contact us for a specific control review.

Is the price fixed?

Yes. Dedicated plans are a fixed monthly price per node, with no charges for queries, egress, or vector count. Resizing or adding a node changes the bill; query traffic does not.

Workload reviewfree

Tell us the workload. We'll tell you if it fits.

An engineer reads what you send. If Rivestack is the wrong tool, we say so.

Get a free workload plan

No call required. Four fields, then an engineer reviews it.

By submitting, you ask Rivestack to reply about this workload. We do not add this address to a marketing list.

Your next idea starts here

Make something worth querying.

Start your free trial

7 days free, then $29/mo · card required