How do I create and download a backup of my Neon database to my local machine?

Use pg_dump against a direct (non-pooled) Neon connection string. The custom format (-Fc) gives you a single compressed archive that pg_restore can read selectively and in parallel. The output file lives on your local machine, so it's an off-platform copy independent of Neon. See Migrate data from Postgres with pg_dump and pg_restore for the full reference.

Create the backup

Copy the direct connection string from the Connect widget on your Project Dashboard (turn Connection pooling off so the hostname has no -pooler suffix). Then run:

pg_dump -Fc -v -d "postgresql://alex:AbC123dEf@ep-cool-darkness-a1b2c3d4.us-east-2.aws.neon.tech/neondb" -f neondb-backup.dump

The flags:

-Fc writes a custom-format archive (recommended for backups)
-v prints progress as it dumps
-d is the source connection string
-f is the output file

To produce a date-stamped backup, use a shell variable:

pg_dump -Fc -d "$NEON_URL" -f "neondb-$(date +%Y-%m-%d).dump"

Restore the backup later

pg_restore reads custom-format archives back into any Postgres database:

pg_restore -v -d "$TARGET_URL" neondb-backup.dump

Add --no-owner if you're restoring to a different role, and --clean --if-exists if you want to drop existing objects first.

When to use a local backup

Neon already keeps a change history that supports instant restore (point-in-time restore) within your project's history window: up to 6 hours on Free, up to 7 days on Launch, and up to 30 days on Scale. For most accidental-delete recoveries, restoring from history inside Neon is faster than rebuilding from a local file.

Local pg_dump backups are useful for:

Off-platform redundancy (your project's history doesn't help if you lose access to your Neon org)
Long-term archival beyond your history window
Compliance requirements that mandate an external copy
Moving data into a different Postgres instance

Use a direct connection for pg_dump

pg_dump relies on session-level SET statements, which aren't supported by Neon's transaction-mode PgBouncer pooling. Always connect through the direct hostname (no -pooler segment). Pooled dumps can fail or produce incomplete output.

For large databases

Add -j <njobs> to dump tables in parallel (set <njobs> to your CPU count) and -Z 1 for light compression. For very large dumps, scale your Neon compute up temporarily so it can serve the read load without throttling.