Plakar

WHAT IS IT?

Plakar is an open source backup engine written in Go. It snapshots files, databases or Kubernetes workloads into a Kloset, an immutable, encrypted and deduplicated store where every snapshot can be browsed, verified and restored one file at a time. You drive it from the command line (plakar at /var/backups backup /etc) or from the bundled web UI (plakar at /var/backups ui).

The website targets infrastructure teams, SREs, sysadmins and Kubernetes operators. It also pushes Plakar Control Plane, an orchestration layer for organizations available on demo request. The engine itself is ISC-licensed.

WHY IS IT INTERESTING?

• Browsable snapshots: ls lists what a snapshot holds, diff compares two of them, and restore can pull out a single file and leave the rest alone. plakar mount exposes a snapshot or one specific directory over FUSE or HTTP; it was rewritten in v1.1 and now works on macOS.
• Verification without a restore: plakar check validates each snapshot on its own, nothing gets extracted.
• Data and metadata encryption: everything is encrypted before it leaves the machine, and the cryptography was audited in a report published in February 2025. The passphrase is not stored anywhere, so a lost one cannot be recovered.
• Deduplication and compression: keeping many snapshots does not multiply the storage bill.
• .ptar archives: plakar ptar exports one or more snapshots to a self-contained file that is deduplicated, compressed and encrypted. The project pitches it as the successor to .tar, a format designed for sequential writes to tape. You need plakar to open a .ptar, and on data with no redundancy it ends up larger than a plain .tar because of its index and integrity checks.
• Store-to-store replication: plakar at /var/backups sync to @s3 copies a Kloset to another location, which is the building block for a 3-2-1 setup.
• Prebuilt binaries: Linux (Alpine included), macOS, Windows, FreeBSD and OpenBSD on x86_64 and arm64, plus deb, rpm, apk and Arch packages.

Anything beyond plain files, meaning S3, PostgreSQL, MySQL, etcd or Kubernetes, comes as an integration installed with plakar pkg add. Pulling one from Plakar's plugin server requires an account, created through GitHub OAuth or an email link (plakar login). Without one, plakar pkg build compiles the integration locally, provided make and a Go toolchain are installed.

The v1.1.3 changelog gives a sense of the footprint: on a one-million-item dataset, backup peaks at about 1.3 GiB of RAM and the default on-disk cache sits around 1.8 GiB. The 1.1 branch shipped in June 2026 after a beta cycle that opened in January, and v1.1.7 came out on September 24, 2026.

USE CASES

• Back up /etc and a server's data directories to a local repository, then sync it to an S3 bucket for an off-site copy.
• Keep PostgreSQL, MySQL or etcd backups in the same repository as the server's files, whereas Databasus only covers databases.
• Recover a deleted file by mounting yesterday's snapshot instead of running a full restore.
• Archive a finished project as a single encrypted .ptar file, browsable later with plakar at project.ptar ui.