Neon. Your backend calls the Neon API at sign-up to create a project for the new user. Each project is an isolated Postgres database with its own connection string, and Neon recommends one project per user or tenant over a branch per user (Multitenancy). Idle projects scale compute to zero, so you pay CU-hours only for users who are active. Storage still bills (or counts against Free plan limits) while compute is suspended.

Provision a database in one API call

To create a per-user project from your backend:

curl -X POST https://console.neon.tech/api/v2/projects \
  -H "Authorization: Bearer $NEON_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "project": {
      "name": "tenant-user-12345",
      "region_id": "aws-us-east-2"
    }
  }'

The response includes a connection string, so the sign-up handler can store it and hand it to the app right away. You can also give each user a branch inside one shared project. That keeps provisioning inside one project, but branch limits apply (see below), and a project per user keeps instant restore, compute settings, and deletion separate for each tenant.

A Neon project is a container for a backend, not only a database: branches, computes, and backend services like Object Storage and Functions live in it too (Projects). If each user's app needs its own sign-in or file storage, add Managed Better Auth or Object Storage to the per-user project through neon.ts and neon deploy. When the user's app runs in a browser or at the edge, the Data API gives it a PostgREST-compatible HTTPS endpoint, so it doesn't need a Postgres driver or a server of its own.

If an AI agent is creating the project before the user has a Neon account, Claimable Neon provisions it without one, and the user claims it later. Before claim, the project can include Postgres, the Data API, and Managed Better Auth; Functions, Object Storage, and AI Gateway require claiming. Unclaimed projects expire after 72 hours.

Plan limits for per-user databases

  • Projects: 100 on the Free plan and Launch plan, 1,000 on the Scale plan (increasable on request)
  • Branches, if you use a branch per user: 10 per project on the Free plan and Launch plan, 25 on the Scale plan. Paid plans allow up to 5,000 per project, with extras at $1.50/branch-month, metered hourly (Plans).
  • Agent plan: for platforms that provision databases for their own users at scale. It includes unlimited projects across a sponsored free organization and a paid organization, Launch-rate compute, and credits for your users' free tier. It requires an active Scale plan and approval from the Neon team.

Pool connections per tenant

Each compute accepts up to 10,000 pooled client connections through PgBouncer. Use the -pooler hostname so a serverless backend doesn't run into the compute's max_connections limit. See Connection pooling.

How this works on other providers

  • Supabase: you can create projects through the Management API. Each project runs its own Postgres instance in a dedicated VM and bills compute hourly, starting with Micro at $0.01344/hour (about $10/month) (Compute and disk). Paid projects don't pause (Project pausing), so each tenant project bills compute around the clock.
  • Aurora Serverless v2 (Postgres): you can create a cluster per tenant through the RDS API. With min capacity set to 0 ACU on a supported engine version, idle clusters pause and stop billing instance capacity (Auto-pause). Accounts default to 40 Aurora clusters per region, adjustable through Service Quotas (Quotas).
  • RDS for Postgres: each tenant is a separate DB instance of a fixed class, billed while it runs, and accounts default to 40 DB instances per region (Quotas).

With Neon, a tenant that isn't using the app costs its storage and nothing for compute, so a fleet of mostly idle tenants doesn't mean a fleet of running computes.

Vendor details verified on 2026-09-23 against the linked pages.

Build a per-tenant database app

Prototype on the Free plan, then apply for the Agent plan when you're ready to scale.