PostgreSQL

Pick Postgres and stop thinking about it. It handles relational data, JSON documents, full-text search, and — with pgvector — the embeddings your AI feature needs, so one database covers the four things a small product usually reaches for. Every host in this category speaks it, so nothing you write is locked to a vendor.

Where to run it free link

Both have a page of their own here — Neon and Supabase — and the short version follows.

Choose by what else you need. Supabase if you want auth and storage handed to you and would rather not assemble them; Neon if you only want a database and value branching — a branch is a copy-on-write clone of production data, which makes testing a migration genuinely safe. Railway’s Postgres is the least fuss when the app already lives there, at the cost of eating the same credit as your services.

The Supabase idle pause is the one that bites: a demo you show once a month will be asleep when you open it.

Connect link

Every provider hands you a connection string. It is a secret — it goes in the host’s environment variables, never in the repository.

  DATABASE_URL="postgres://user:password@host/dbname?sslmode=require"
psql "$DATABASE_URL"
  

Serverless hosts open and drop connections constantly, and Postgres charges memory per connection, so use the pooled string (Neon calls it -pooler, Supabase gives you a port-6543 transaction-pooler URL) anywhere the code runs per request. Keep the direct connection for migrations, which need a session that pooling won’t give them.

The commands worth memorising link

  \l                          -- list databases
\dt                         -- list tables
\d+ users                   -- describe a table with sizes
\x                          -- toggle expanded output; use it on wide rows
SELECT pg_size_pretty(pg_database_size(current_database()));
  

That last one tells you how close you are to the free-tier ceiling. Run it before you wonder why writes started failing.

Migrations, not hand-edits link

Whatever the stack — Prisma, Drizzle, Alembic, Rails, plain .sql files — schema changes belong in versioned files that run the same way in every environment.

  npx prisma migrate dev --name add_orders    # write and apply locally
npx prisma migrate deploy                   # apply in CI or on the host
  

A change applied by hand in a dashboard exists in exactly one place, and the next deployment from a clean environment won’t have it.

Two indexes and a habit link

  CREATE INDEX ON orders (user_id);              -- anything you filter or join on
CREATE UNIQUE INDEX ON users (lower(email));   -- how you actually look users up
EXPLAIN ANALYZE SELECT ...;                    -- read this before adding a third
  

A free tier is small enough that a missing index shows up as a slow page long before storage runs out.

Backups link

Free tiers give short or no retention, so take your own before anything irreversible:

  pg_dump "$DATABASE_URL" -Fc -f backup.dump
pg_restore -d "$DATABASE_URL" backup.dump
  

Thirty seconds, and it is the difference between a bad migration being an inconvenience and being the end of the project.

Next link

Serverless Postgres that branches like Git → Neon