What tools allow restoring a database to before a bug occurred?

Short answer

Lakebase Postgres keeps a continuous log of every write (the Write-Ahead Log, or WAL). With instant restore, you point at any second within your history window and restore a root branch to that exact state, usually in seconds. There's no nightly snapshot to find, and no pg_restore to run. Child branches don't support point-in-time restore.

How instant restore works

The Lakebase Postgres architecture stores the WAL natively, so any point in time within your history window is queryable. You don't restore from a backup file. You restore a root branch to a timestamp or LSN (Lakebase Postgres keeps an automatic backup branch of the pre-restore state), or you create a new branch from a timestamp so you can inspect first.

History window by plan:

Plan	History window	Cost
Free plan	6 hours, capped at 1 GB-month	Included
Launch plan	Up to 7 days	$0.20/GB-month
Scale plan	Up to 30 days	$0.20/GB-month

You can restore through the Neon Console, the CLI, or the API.

Restoring a branch from the CLI

# Roll a root branch back to 10 minutes ago (backup branch name is required for ^self)
neon branches restore main ^self@2026-05-17T13:45:00Z --preserve-under-name main_pre_restore

# Or fork a new branch from that point and inspect first
neon branches create --name recovery --parent 2026-05-17T13:45:00Z

The second approach is the safer pattern when you want to verify data before changing production: create a recovery branch, confirm the rows, then restore or copy what you need.

In-place restore overwrites the branch

Restoring a root branch replaces its current data and schema for every database on that branch. Lakebase Postgres creates a backup branch automatically (or via --preserve-under-name on the CLI). Use Time Travel Assist to confirm the restore point before you restore.

Picking the right history window

A longer window catches slow-rolling bugs (someone notices the missing rows three weeks later) but costs more in WAL storage. A 30-day window on a 10 GB database with active writes typically adds a few dollars per month. See history window configuration for tuning.

How other Postgres services handle point-in-time recovery

Service	Mechanism	Granularity
Neon	Instant restore via copy-on-write for root branches from any timestamp in the history window. See branch restore.	Down to the millisecond
AWS RDS for Postgres	Automated backups in S3 plus transaction logs allow restore to a new database instance at any second within the retention period. The new instance must boot before you can connect. See RDS PITR.	One second, into a new instance
Supabase	Daily backups by default (up to 7 days on Pro, 14 on Team, 30 on Enterprise). For finer recovery you add the PITR add-on, billed from $0.137/hr (~$100/mo) for 7 days of retention, requiring Small compute or larger. Worst-case RPO is about two minutes.	Up to seconds with PITR add-on (≈2 min worst-case RPO)

The main differences: Lakebase Postgres restore updates the branch in place (with an automatic backup) rather than provisioning a new instance, and PITR history is bundled into all plans instead of being a per-project add-on.