Restore points

A restore point gives a moment a name. before-drop-legacy-orders is much easier to find at 3 a.m. than a timestamp. If a change goes wrong, you can restore to exactly that point: everything committed before it is there, nothing after it.

Create one right before a migration, a backfill, a bulk delete, or anything an AI agent is about to run:

rowsafe mark app before-migration-42

Rowsafe waits until the point is safely in your bucket before it returns, usually within a few seconds.

Names

• 1 to 63 characters: lowercase letters, digits, - and _, starting with a letter or digit.
• Unique per database.
• Without a name, rowsafe mark app uses manual-<UTC time>, for example manual-20260924-140500. Restore points created by AI agents are named agent-<UTC time> by default.

In a project with a .rowsafe.json, you can leave out the database: rowsafe mark before-drop.

List them

rowsafe marks app

NAME                 STATUS    CREATED              RESTORE FROM BACKUP  LSN         BY
before-migration-42  archived  2026-09-24 14:05:12  20260921-010002F     0/5A000090  key:key_... (CLI · rowsafe CLI on laptop)

Restore from backup is the newest backup that finished before the point was created. You need it to restore to the point: see Restore to a restore point.

Good to know

• Restore points are cheap: two SQL calls and one WAL segment on PostgreSQL.
• They need an active database (archiving proven).
• They don't pause or wait behind a running backup or drill.
• Creating one is recorded in the audit log with who created it.
• A restore point doesn't restore anything by itself. It marks a moment you can restore to.