Tech USP: Backups run continuously by default, you don’t turn anything on. What matters is how far back you can restore.
Buyer’s view: “Our last backup ran three weeks ago and nobody tested if the restore actually works.” We hold a recoverable state of every minute of the last 30 days, and we practice the restore on a schedule.
Three layers, all running by default
Layer 1: Continuous capture. Every change to your database is written to a separate, durable backup stream as it happens, not “every night at 3am”. If something goes wrong at 14:23, you can recover the state from 14:22.
Layer 2: Daily full snapshot. Once a day, the full database plus all attached files (uploads, generated PDFs, audit log) are bundled into a single recovery point and shipped off-site. Retention: 7 days, 4 weeks, 3 months.
Layer 3: Practiced restore. Once a quarter, we restore the latest snapshot into a throwaway environment and verify the system actually comes up with your data intact.
What you can ask for
| Scenario | What you get |
|---|---|
| “Restore yesterday’s state of this one record” | Single-record rollback from the audit trail, in seconds |
| “Restore the database to 14:23 last Tuesday” | Point-in-time recovery, typically minutes for mid-market databases |
| “GDPR access request: give me everything we have on customer X” | One-click export, typically a few minutes |
| “GDPR erasure: delete everything about customer X” | Soft-delete first (audit trail keeps the fact that data existed), hard-delete on retention rollover |
| “Disaster: the whole instance is gone” | New instance from off-site snapshot, typically under an hour for mid-market data |
Restore times scale with database size. The numbers above are realistic for typical DACH mid-market apps (low-millions of records). Larger databases take longer. We’ll tell you upfront.
Off-site means EU-off-site
Backup storage is EU-only: same compliance posture as the live database.
For self-hosted and on-prem customers, the same architecture runs on your own storage.
Why this matters
Most “we have backups” stories collapse on three questions:
- When was the last successful restore actually tested? Usually “never” or “I think Bernd did it once in 2022”.
- How far back can you recover? Usually “last night” or “no idea”.
- Can you restore one customer’s data without nuking everyone else? Usually a long silence.
Kumiko’s answer to all three: tested quarterly, 30 days, yes.