SafeGrd backs up your databases, files and mail, encrypts and locks each copy, and restores it on a schedule to prove it works. An agent with your production keys can drop the database, but not the backups.
14 days free, no card. A full refund of any payment within 30 days. About 2 minutes from install to the first backup. Plans from $15 a month, unlimited databases.
shop restored in full.
Restored into a sandbox at 13:54 UTC, then signed and added to the attestation record.
Customer-managed key. Only you can decrypt these backups.
The daemon runs beside your database and backs up straight to a bucket, yours or ours. SafeGrd receives a record of each backup, not the data. If you have no server, SafeGrd runs the backup on a machine it starts for the job.
One command on a server that can reach the database, or the Docker image beside it. On Supabase, Neon or RDS, paste the connection string in the console.
Name a database, a directory or a mailbox in the console. It backs up on schedule, hourly on paid plans. A failed or late backup sends an alert.
SafeGrd restores the newest backup into a throwaway database, counts every table and row, and signs the result. We call this a Fire Drill, and it runs weekly or daily depending on your plan.
curl -fsSL https://safegrd.dev/install.sh | sh
Then safegrd try postgres://… backs up your database and restores it to check, before you sign up. safegrd signup keeps the backup and the host (how it works).
Also a GitHub Action and a Helm chart. Details in the install guide and the docs.
$ safegrd verify --snapshot <newest> --sandbox-target postgres://…/drill_sandbox
Warning: the value of --sandbox-target is visible in `ps` and shell history. Pass env:VAR or file:/path instead.
Verifying snap-20261031-135407-16c6d3 (epoch e202610-b3036e43, shop-db): checking the packs, restoring it into postgres://postgres:xxxxx@localhost:62309/drill_sandbox?sslmode=disable.
[PASS] DecryptionIntegrity (Expected: the run, its indexes and trees open with the identity, Actual: they do)
[PASS] RepositoryIntegrity (Expected: every pack present and agreeing with its index, Actual: 3 packs checked)
[PASS] DigestIntegrity (Expected: f9eef402a29aca7aac7575ff9de386b4e31978b155fc35b5affd11e1923133b1, Actual: f9eef402a29aca7aac7575ff9de386b4e31978b155fc35b5affd11e1923133b1)
[PASS] TableCountMatch (Expected: 4 tables, Actual: 4 tables)
[PASS] RowCountMatch (Expected: 45404 rows, Actual: 45404 rows)
[PASS] Row Count: public.customers (Expected: 1840 rows, Actual: 1840 rows)
[PASS] Row Count: public.order_items (Expected: 31000 rows, Actual: 31000 rows)
[PASS] Row Count: public.orders (Expected: 12500 rows, Actual: 12500 rows)
[PASS] Row Count: public.products (Expected: 64 rows, Actual: 64 rows)
[PASS] Newest Row: public.customers (Expected: created_at 2026-10-31T13:53:57.447761Z, Actual: created_at 2026-10-31T13:53:57.447761Z)
[PASS] ExtensionBootCheck (Expected: [], Actual: [])
[PASS] Schema Captured By pg_dump (Expected: schema from pg_dump, Actual: pg_dump 18.6)
[PASS] Data Check: amcheck (Expected: every heap and B-tree index readable, Actual: 4 tables, 3 B-tree indexes, no corruption)
Fire Drill Passed
Verification ID: verif-cc68f9a0
Duration: 236ms
Tables: 4
Rows: 45,404
Certificate: cert_sg_d05260a52f24625bb27f4687d67b67781d72da6b13b2003c33128037d1f88ecf
safegrd verify: the newest backup restored into an empty database and checked.Any S3-compatible bucket works, and SafeGrd does not meter what you store in it. Setup for each one.
Every backup is encrypted and locked under one key, and the newest is restored on a schedule.
shop restored in full.
Restored into a sandbox at 13:54 UTC, then signed and added to the attestation record.
The restore and the certificate are from a drill of SafeGrd's demo database on 31 October 2026.
Databases, directories and mailboxes, each with its own drill record.
Green when the last drill passed, amber when one runs late, red when one fails.
SafeGrdverified 3h agoConnect an agent over MCP and it can check backups, take one before a risky migration and test a restore. Its token can add backups but cannot remove one.
What your agent can do with it.
No token can delete a backup or shorten retention. Every call is logged. Revoke the token and the agent stops.
An agent can restore only into an empty directory, or a database SafeGrd confirms is empty.
safegrd guard -- <command> takes a locked snapshot, then runs the command. Hooks for Claude Code, Cursor and Codex call it before every destructive shell command. If the backup fails, the command does not run. Hook recipes.
safegrd doctor --agent-proof runs five checks that the backups survive an agent holding this host’s credentials. A failing check prints its fix.
SafeGrd writes each backup under S3 Object Lock, and nothing can delete it until the lock ends. Your plan or your retention settings set that date.
SafeGrd keeps your key sealed and releases it only to your enrolled hosts, so you can restore even after losing a host. Or manage the key yourself, and only you can decrypt the backups.
About 2 minutes from install to the first backup, and under 15 minutes to a first tested restore. Click Drill now in the console to run the first Fire Drill straight away.
A cron job copies the database, and that is all it does. SafeGrd also encrypts and locks each copy, alerts you when a backup fails, and restores it on a schedule. To do that by hand, see how to test that a backup restores.
In your own bucket, they stay where they are. The daemon keeps backing up on its own schedule, and the source-available CLI restores without an account. Each enrolled host also keeps a copy of its key. safegrd export copies hosted snapshots to your own bucket at any time.
From $15 a month, with unlimited surfaces and team members. Plans differ in how often SafeGrd tests your backups. See Pricing, or compare with AWS Backup, SimpleBackups and Ottomatik.
Any machine that can reach the database works: a server, a container, a Kubernetes cluster or a scheduled GitHub Actions job. A database with no server beside it can be backed up and drilled on SafeGrd’s own machines. Install and enroll covers each option.
Read access only. On PostgreSQL 14 and newer that is a role granted pg_read_all_data, and safegrd doctor warns if the role can write. Files need read access, a mailbox its IMAP login. Give SafeGrd a read-only role has the grants.
Provider snapshots sit in the same account as the database, so whoever can drop the database can often delete them, and nothing tests them. SafeGrd keeps a separate, locked copy and tests it.
Not on a bucket with S3 Object Lock in compliance mode. The storage provider refuses to delete a locked backup before its lock ends, even for root credentials. An agent’s token can check status, take a backup or ask for a drill, and nothing else. AI agents (MCP).
The signed record of each Fire Drill, which they can verify offline with SafeGrd’s public key. On Scale, the Evidence Pack collects them with your retention and lock settings. Backup controls maps records to controls.
Install the CLI, add a database and run safegrd verify. The first backup takes about 2 minutes. 14 days free, no card.