Database backups

A database backup you can query, verify, and erase from.

A compressed dump is usually a dead weight: opaque bytes you can only read by restoring the whole thing. AT-1 compresses a PostgreSQL or MySQL backup well below gzip and zstd, then keeps it queryable in place, proves it restores byte-for-byte, and lets you delete one subject for POPIA/GDPR without a rewrite.

From a production deployment

158.8 MB
raw PostgreSQL 17 dump
5.46 MB
AT-1 (29.1x), verified lossless
6.20 MB
zstd-19 on the same dump
19.0 MB
gzip-9 on the same dump

A property-management SaaS backs up its live PostgreSQL with AT-1. The dump above came down to 5.46 MB (29.1x), smaller than both gzip-9 and zstd-19, then passed a full disaster-recovery drill: decompressed, restored into a fresh PostgreSQL, and reconciled table by table against live production. It matched to the cent. For a business where a lost database means lost records, that provable restore is worth more than the disk saving.

For a structured backupSmaller*Query in
place
Verify
byte-exact
Per-subject
erasable
gzip -9
ubiquitous, but the largest here and opaque: you must restore to read anything
zstd -19
fast and strong, 6.20 MB on this dump, still opaque and no per-record delete
Restic / Borg
excellent deduplicating, encrypted, integrity-checked backup repos; contents stay opaque and a single subject can't be deleted without a restore + rewrite
dedup
AT-1
5.46 MB on this dump (29.1x), query it in place, verify byte-for-byte, erase one subject without a rewrite

* “Smaller” is on a single structured dump, where AT-1 measured below gzip-9 and zstd-19. Restic and Borg deduplicate across many snapshots, so on a long backup history their total footprint can be the smallest; they are not queryable in place or per-subject erasable. Pick the tool for the job.

What a backup gains with AT-1

Queryable in place

Filter, project, and pull one record with a range request, without restoring the whole dump first. The backup stays useful, not just stored.

Provable restore

A SHA-256 manifest binds every archive, so a restore is byte-for-byte the original or verification fails. No silent corruption, no hoping the backup is good.

Erase one subject

A right-to-erasure request destroys one subject's key; their rows become unrecoverable while the rest of the backup stays intact and queryable. No restore-and-rewrite.

Try it on your own dump

The CLI is public. Point it at a real backup and it reports the ratio and a byte-exact lossless verdict, no account needed to read.

npm i -g @tinyfiles/cli
pg_dump mydb > mydb.sql
at1 audit ./mydb.sql          # ratio + byte-exact lossless verdict

Database-backup questions, answered

How much smaller is a PostgreSQL backup with AT-1?
On one real production dump, a 158.8 MB PostgreSQL 17 backup compressed to 5.46 MB with AT-1, a 29.1x reduction. On the same dump gzip-9 produced 19.0 MB and zstd-19 produced 6.20 MB, so AT-1 was 3.48x smaller than gzip-9 and 1.13x smaller than zstd-19. Your ratio depends on your schema and data; the point is that structure-aware compression beats a general byte compressor on a structured dump.
Is the compressed backup actually lossless?
Yes, and it is enforced, not asserted. Before any AT-1 file is written the encoder decodes it and compares the result byte-for-byte against the original; a single mismatched byte rejects the file, so it never ships. The production dump above was also put through a full disaster-recovery drill: decompressed, restored into a fresh PostgreSQL, and every table and financial total reconciled against live production. It matched to the cent.
Can I query the backup without restoring it?
Yes. An AT-1 backup is queryable in place: you can run filters and projections, and pull specific records with an HTTP range request, without decompressing the whole archive or loading it into a database first. Plain pg_dump + zstd, Restic and Borg all require a full restore before you can read anything inside.
How is this different from Restic or Borg?
Restic and Borg are excellent at what they do: deduplicating, encrypted, integrity-checked backup repositories, and across many snapshots their deduplication can win on total size. They are not trying to be queryable or per-subject erasable, and their repos are opaque: to read or verify a record you restore first. AT-1 is for when you need the backup itself to stay searchable and provable, be smaller on a single structured dump, and let you delete one person for POPIA/GDPR without restoring and rewriting. Use Borg/Restic for broad file-system snapshots; use AT-1 where the backup is structured data you need to query, verify, or erase from.
Do I have to change my backup pipeline?
No migration. Point AT-1 at the dump your existing job already produces (pg_dump, mysqldump, a CSV/Parquet export) as the compression and verification step, in place of gzip or zstd. Nothing is uploaded: AT-1 runs where your backup runs, and the open Apache-2.0 decoder reads any AT-1 file forever, so you are never locked in.
How do I verify a restore is correct?
Every AT-1 archive carries a SHA-256 manifest, so a restore either reproduces the exact original bytes or verification fails, there is no silent corruption. You can confirm it yourself with the public CLI: run the audit on your own dump and it reports the ratio and a byte-exact lossless verdict, no account needed.

Bring one real backup and we'll show you the ratio, the restore proof, and the query, on your data.

Get in touch