cloudparts

Plakar

A backup engine in Go that takes deduplicated, encrypted, immutable snapshots of files, databases and cloud sources into a store called a Kloset, which you can browse, diff, verify and sync without restoring.

What it is
Backup & restore
Type
Open source
Made by
Plakar Korp
License
ISC
Facts checked
24 October 2026
Website Source code

What it is

Plakar is a single command-line program, plakar, with a built-in web UI. You create a store, a Kloset, on a local disk, an SFTP server or an S3-compatible bucket, then run plakar backup against it. Data is split into chunks with content-defined chunking, so an insertion in a file only changes the chunks around it; each chunk is compressed, encrypted on the client and stored once, by content hash. A snapshot is a signed, hashed tree of those chunks plus metadata. Nothing already written is ever modified, and the storage backend only sees opaque encrypted bytes; the encryption design was reviewed by the cryptographer Jean-Philippe Aumasson, after which Plakar moved to Argon2id, AES-256-GCM-SIV and keyed BLAKE3.

Snapshots can be listed, browsed, diffed and checked from the CLI or the UI without restoring them, and a single file or folder can be restored by adding its path to the snapshot ID. plakar sync copies snapshots between stores, for example from a local disk to S3, which is how you build a 3-2-1 setup. plakar ptar exports snapshots to a .ptar file, a self-contained, deduplicated and encrypted archive. Integrations installed with plakar pkg add add sources and destinations such as PostgreSQL, MySQL, MongoDB, etcd, Kubernetes, Proxmox, VMware and OpenStack.

Plakar doesn’t run as a daemon; you schedule plakar backup with cron or a systemd timer. The company behind it, Plakar Korp, sells Plakar Control Plane, a self-hosted management layer with policies, schedules and an inventory, on top of the ISC-licensed engine.

Why teams pick it

  • Content-defined chunking deduplicates across snapshots and machines, and keeps working when data shifts inside a file
  • Client-side encryption of data and metadata, with an audited design and every snapshot verifiable without a restore
  • Snapshots are browsable and diffable from the CLI or the web UI, and single files restore without touching the rest
  • One tool for files, databases, Kubernetes and object stores, through installable integrations, with sync between stores for off-site copies

Watch out for

  • Young: 1.0 shipped in May 2025, and the store format and integrations are still changing quickly
  • No built-in scheduler or central server in the open-source engine; fleet management and policies are the commercial Control Plane
  • Encryption is always on, so losing the passphrase means losing the backups

Alternatives

What all backup & restore have in common →Compare side by side →
DuplicatiOpen sourceA backup client for Windows, macOS and Linux that splits files into blocks, stores each block once, and uploads compressed, encrypted volumes to cloud storage or a file server, driven from a web UI or the command line.

How it fits with other technologies

Create a store, back up /etc, restore one folder, copy the store to S3 · bash

plakar at /var/backups create
plakar at /var/backups backup /etc
plakar at /var/backups ls

# restore only /etc/ssh from one snapshot
plakar at /var/backups restore -to /tmp/restore <snapshot-id>:/etc/ssh

# keep an off-site copy in a bucket configured as @s3
plakar at /var/backups sync to @s3

# browse, diff and preview snapshots in the browser
plakar at /var/backups ui

See it in a real stack

No architecture uses Plakar yet. Drop it into the playground and wire it up yourself.Open playground →
Where it came from2022–2026

Gilles Chehade is by far the largest contributor to Plakar’s code on GitHub, and its first alpha releases appeared there in April 2022. In the 2025 beta announcement he wrote that it was “born out of dissatisfaction with both open-source and commercial alternatives”, and that the aim was “to provide the most advanced backup features in the easiest form possible”.

The company, Plakar Korp, was incorporated in September 2024 to support its development. A public beta followed in February 2025, with a cryptography audit published the same month, and v1.0 was announced on May 1, 2025 together with a $3 million pre-seed round. Over the rest of 2025 the team described Kloset, its storage engine, introduced the .ptar archive format, and published its chunking library, go-cdc-chunkers.

Plakar joined the Linux Foundation and the CNCF as a member in January 2026. Version 1.1, which the team called its largest release, removed the background agent, rewrote mounting and sped up restores; in June 2026 the company opened a free plan for Plakar Control Plane, its first commercial product.

Why it’s built this way

Content-defined chunking
Fixed-size chunks all shift when bytes are inserted near the start of a file. Cutting chunks where the content says so keeps the delta between snapshots as small as possible.
An immutable store that never trusts the backend
Data and metadata are encrypted before they leave the client and never rewritten, so a storage connector only needs to list, write and read ranges of opaque bytes, and a compromised bucket reveals nothing.
Snapshots you can query without restoring
Kloset keeps each snapshot as a structured, self-describing tree, so you can browse, diff, search and verify backups directly instead of treating them as opaque archives.

Timeline

  1. 2022

    First alpha releases on GitHub in April

  2. 2024

    Plakar Korp incorporated in September

  3. 2025

    Public beta and cryptography audit in February; v1.0 and $3M pre-seed round in May

  4. 2026

    Joins the Linux Foundation and CNCF as a member; v1.1; a free plan for Plakar Control Plane

Sources7 · facts checked 24 October 2026

Improve this page on GitHub