Glossary
Rowsafe's words in plain language, A to Z.
Each term links to its full page.
Agent
The small open-source program on your server that backs up, runs Proof and reports health. Outbound connections only. Hosts and agents.
AI agent change
A production change an AI agent makes right away as its person, with their rights, or that Rowsafe refuses with a reason. Nothing waits for approval. How AI agents act.
API key
A key (rsk_...) for the CLI, scripts, CI or AI assistants; read-only or read-write. Settings → API keys. API keys.
Apply fix
The button on a Pulse finding that fixes it for you, after you confirm. Apply fix.
Ask Rowsafe
Ask about your databases in plain words; get an answer and, when there is one, a fix button. Ask Rowsafe.
Backup
A copy of the whole database: full (Sundays) or differential (only changes since the last full, other nights). Backup types.
Bring back rows
Put lost rows back into production from a restored copy, leaving everything else as is. Bring back rows.
Bucket
Your own S3-compatible storage where backups live (R2, B2, S3, Wasabi, Spaces...). Rowsafe never has its keys. Set up backup storage.
Change log (WAL)
The database's running record of every change, copied to your bucket within about a minute so you can restore to any second. Called WAL (PostgreSQL), binary log (MySQL, MariaDB), oplog (MongoDB). How PostgreSQL backups work.
Connection pooling
A pooler (PgBouncer, ProxySQL) lets many app connections share a few database connections. Connection pooling.
Control plane
See Rowsafe service.
Copy
A Rewind copy: your database at a moment you pick, on your server, private, read-only, deleted after 24 hours unless kept. Restore a copy.
Database
One database server instance (one host and port, every database inside it), backed up and restored together. Create databases and users.
Drill
The internal name for a Proof run, seen in tasks and logs.
Encryption passphrase
The secret that encrypts backups on your server. Only you have it; without it, nothing can be restored. The encryption passphrase.
Enrollment token
A one-time code (rse_...) in the install command that joins a server to your organization. Valid one hour by default. Enrollment.
Failover
Switching to the standby when the main server dies: Promote by hand, or automatic (off by default). Automatic failover.
Find the moment
Finds the exact second rows were deleted or changed, or a table emptied or dropped. Find the moment.
Finding
One problem Pulse found, in plain words, usually with Apply fix. Pulse.
Guard
The safety net for AI coding agents: MCP tools, plugins, the rowsafe guard hook, migration previews, the GitHub Action. Guard.
Health score
0 to 100 per database. A critical problem caps it at 59, a warning at 89. Health score.
Host
A server (or the agent's Docker container) running the agent. Called Servers in the dashboard. Hosts and agents.
Logs
The database's own log in the dashboard, with values removed on your server first. Logs.
Major upgrade
A new major version (for example PostgreSQL 16 to 18), checked and rehearsed on a copy first, with undo. Debian or Ubuntu packages only. Major upgrades.
Mark
A named restore point, such as before-migration-42, you can rewind to exactly. Marks.
MCP
The standard way AI assistants use outside tools. Rowsafe's MCP server gives them your databases, as their person. MCP server.
Migration preview
Runs a migration on a fresh copy and reports locks, rewrites and time. Production is never touched. Preview a migration.
Minor update
A bug and security release of your version (for example 16.9 to 16.10): one click, a few seconds' restart. Debian or Ubuntu packages only. Minor updates.
Move in
Copies a database from a managed provider onto your server while apps keep running. Move in.
Point-in-time recovery
Restoring to any second, not just the last backup. Also called restore to any second. Backups and recovery.
Promote
Making a standby the main server. See Failover.
Proof
The weekly automatic restore test: restore the latest backup, check it, delete it. Proof.
Protected
Yes or no: could this database be restored right now? Protection status.
Pulse
Monitoring: health score, findings with fixes, alerts, slow queries, logs, tuning, recommendations, the Monday email. Pulse.
Recommendations
Not problems yet, but soon, or ways to be faster or cheaper (a missing index, an id running out). Recommendations.
Recovery window
How far back you can go: about two weeks by default. The recovery window.
Replica
See Standby.
Retention
How many full backups are kept (2 by default). More means a longer recovery window. The recovery window.
Rewind
Every change saved; restore to any second or Mark in one click. Restore a database.
Rewind in place
Rewind the whole live database, for the worst case. Undo for 7 days. Rewind the whole database.
Rowsafe service
The hosted part (control plane): dashboard, API, schedules, alerts. Never sees your data or secrets. How Rowsafe works.
Rowsafe Storage
A bucket run by Rowsafe, so you don't need your own. Still encrypted on your server. Rowsafe Storage.
Safe copy
A copy with personal data replaced by realistic fakes, for developers and AI agents. Deletes itself after a day. Safe copies.
Second copy
Your backups in a second bucket too, ideally at another provider. Keep a second copy.
Server
See Host.
Standby
A second server kept in sync (a replica), readable and promotable. One per database. Standby servers and failover.
Tuning
The settings that matter, with values for your server, changed on click, with undo. Tuning.
WAL
See Change log (WAL).