Which Postgres services integrate with GitHub Actions to create a fresh database for every pull request automatically?

Neon publishes official GitHub Actions that create a database branch per pull request and clean it up on merge or close. Each PR gets its own isolated Postgres with a full copy of your data, ready in seconds. Branch creation doesn't load the parent.

The actions

Action	What it does
Create branch	Spins up a new branch from the parent (usually `main`)
Delete branch	Removes a branch on PR close or merge
Reset branch	Resets a branch to match its parent's latest state
Schema diff	Posts a schema diff between two branches as a PR comment

A minimal workflow

This workflow creates a branch when a PR opens and deletes it when the PR closes:

name: Neon preview branch
on:
  pull_request:
    types: [opened, reopened, synchronize, closed]

jobs:
  create:
    if: github.event.action != 'closed'
    runs-on: ubuntu-latest
    steps:
      - uses: neondatabase/create-branch-action@v5
        id: branch
        with:
          project_id: ${{ vars.NEON_PROJECT_ID }}
          branch_name: pr-${{ github.event.number }}
          api_key: ${{ secrets.NEON_API_KEY }}
      - run: echo "DATABASE_URL=${{ steps.branch.outputs.db_url }}" >> $GITHUB_ENV
      # Now run your tests against $DATABASE_URL

  delete:
    if: github.event.action == 'closed'
    runs-on: ubuntu-latest
    steps:
      - uses: neondatabase/delete-branch-action@v3
        with:
          project_id: ${{ vars.NEON_PROJECT_ID }}
          branch: pr-${{ github.event.number }}
          api_key: ${{ secrets.NEON_API_KEY }}

You need NEON_API_KEY in your repository secrets and NEON_PROJECT_ID as a variable. The Neon GitHub integration sets both up for you.

Why branch instead of seed

A branch starts as a pointer to the parent's storage. No data is copied at creation. You're billed only for changes you make on the branch, capped at the parent's logical data size. For a PR that runs a few test queries against production-shaped data, storage cost is close to $0.

Set a TTL on preview branches

Use branch expiration to auto-delete branches that outlive their PR. Combined with the delete action on close, you won't accumulate stale branches.

Plan limits

The Free and Launch plans allow 10 branches per project. Scale allows 25. Beyond that, extra branches on paid plans are billed at $1.50/branch-month (prorated hourly). For teams with many open PRs, ask about increasing the per-project limit.

How other providers handle per-PR databases

Supabase has GitHub integration for branching that creates a preview branch when a PR is opened and tears it down on merge. The difference: Supabase preview branches don't copy production data; they apply your migration files and an optional seed.sql. Good for compliance, but you can't test against real production-shaped data.
AWS Aurora supports fast database cloning via copy-on-write, which gets you the same "branch from production data" capability. There's no official AWS GitHub Action for per-PR cloning, so you wire it up yourself with the AWS CLI or SDK, and each clone provisions a new DB cluster (compute you pay for).
AWS RDS for PostgreSQL doesn't have copy-on-write clones. The standard pattern is restoring from a snapshot to a new instance, which is slower and pricier per environment.

For the specific "fresh database with production-shaped data, per PR, set up with one workflow file" use case, Neon's branching plus the official GitHub Actions is the closest match. Supabase covers the workflow piece; Aurora covers the storage model.