BorgBackup

What BorgBackup actually is

BorgBackup is a deduplicating archiver: a command-line program that takes your files, chops them into content-defined chunks, compresses and encrypts those chunks, and stores them in a repository on a local disk, a USB drive, or any machine reachable over SSH. It is free software under the BSD-3-Clause license, developed in the open since 2015, and currently maintained as version 1.4.x with a large community behind it.

We group Borg with the other tools in our backup solutions category, but it is a different kind of animal. There is no service behind it, no account, no dashboard you log into. Borg is a program you run yourself. That distinction matters for everything that follows: pricing, support, and what you are actually buying (spoiler: nothing, you are buying nothing).

The core idea is content-defined chunking. Each file is split into chunks at content-dependent boundaries, and every chunk is stored exactly once per repository, no matter how many archives, machines, or years of history reference it. In practice this means a nightly full backup costs roughly what yesterday's changes cost. Our own view after years of running Borg across fleet machines: the deduplication ratios are real and they are the reason the tool exists. A repository holding thirty daily snapshots of a document-heavy server often ends up only modestly larger than the source data itself.

On top of dedup, Borg layers authenticated encryption and compression. The security model is spelled out in the internals documentation and it is unusually honest: the client environment is trusted, the repository server is not. The server only ever sees ciphertext, and an attacker with full read/write access to the repository cannot modify, rename, remove, or add an archive without the client detecting it, nor recover plaintext. Compression is a per-repository choice between lz4 (speed), zstd (balance), zlib, and lzma (maximum squeeze), or none at all. Archives can be mounted as read-only filesystems via FUSE, which turns restoring an old version into browsing a folder.

Two operational properties win Borg a lot of fans in the sysadmin world. First, backups are one command: there is no daemon, no agent, no always-on service to babysit. Second, repositories support append-only mode and an enforced retention/compaction model, so a compromised client can be prevented from deleting existing history — the single most common way ransomware destroys backups.

Pricing: what you really pay

The software is free. BSD-3-Clause license, installable from distribution packages, single-file binaries, or pip. There is no paid tier of Borg itself, no per-seat licensing, no usage metering. If your repository lives on storage you already own, your marginal cost is zero.

Where money enters the picture is repository hosting, and the established option is BorgBase, a service built specifically around Borg (and Restic) repositories. BorgBase advertises hosting "from $2.50/month, 10 GB free forever", with a Small plan at $30/year for 250 GB (additional space $0.015/GB/month), Medium at $110/year for 1 TB (additional space $12/TB/month), and Large at $210/year or $20/month for 2 TB (additional space $10/TB/month). Those plans include EU and US regions, append-only mode, 2FA, and email support, and BorgBase charges only for stored data after compression and dedup. If you want commercial support for the software itself, Thomas Waldmann's company Waldmann EDV offers Borg consulting and development.

We think that split is healthy. You can run Borg for a decade without paying anyone, and if you want someone else to hold the encrypted repository, you pay a small, comprehensible amount to a company whose entire product is "a safe place for your Borg repo". Compare that with the pricing pages of the major backup suites in our benchmarks and the simplicity stands out.

Performance & infrastructure

Borg's performance profile is dominated by two phases: hashing/chunking on the client, and transfer to the repository. On modern hardware the chunking runs at disk-throughput speeds for most workloads; compression cost follows exactly the codec you picked. zstd is the setting we reach for by default; lz4 when the machine is old or the link is fast; lzma only for cold, tiny archives where CPU time is free.

The important infrastructure fact is that Borg has no infrastructure. Repositories are filesystem-based transactional key-value stores — segment files, an index, hints for compaction, a locking roster. Anything that looks like a POSIX filesystem works: ext4, ZFS, Btrfs, NFS, SSHFS, an external drive in a drawer. There is no S3-native repository mode in 1.4.x; people do reach object storage through rclone mounts or similar, but that is a workaround with real caveats around locking and latency, and we would not build a production backup scheme on it. If object storage is a hard requirement, pick a tool designed for it.

That "bring your own disk" model is also why the usual hosting specs in our comparison tables are not applicable here. Datacenter counts, SLA uptime figures, DDoS protection, IPv6 support, GPU availability — Borg is the client, not the facility. When you host with BorgBase you get EU and US regions and the service's own operational guarantees; when you host on a Hetzner box or a NAS, the specs are those of that box. We have left those fields empty rather than invent numbers for a program that has none.

On the reliability side, Borg's track record is long and public. The project has been actively maintained since 2015, ships through distribution packages and PyPI, documents its on-disk format in detail, and publishes a security page covering its attack model and known issues (including a documented pre-1.0.9 manifest authentication flaw, CVE-2016-10099, fixed with the TAM mechanism). Backup tools live or die by restore reliability, and the community's accumulated restore experience over a decade counts for more than any marketing claim.

Where it falls short

The interface is a command line. There is no official GUI. Graphical front ends exist — Vorta for the desktop, Borgmatic for configuration-driven runs — but they are separate projects with their own release cadence and their own bugs. For a one-person shop without terminal habits, the learning curve is real: keyfiles, repo keys, prune policies, and compaction all need to be understood, because getting retention wrong is how people discover they kept ninety full snapshots or zero.

Borg does not back up live databases well on its own. It is file-based; consistent snapshots of PostgreSQL or VM images require pre/post hooks or filesystem-level snapshotting. Enterprise agent features — bare-metal restore images, centralized job scheduling across thousands of endpoints, web consoles — are out of scope. Borgmatic softens the scheduling and configuration problem but it is still the same self-hosted model.

The multi-client caveat in the security documentation deserves emphasis: several independent clients writing to the same repository at the same time weaken the confidentiality guarantees, and there is no server-side access control in Borg itself — access control is whatever SSH gives you. And while the community support is genuinely good (mailing list, issue tracker, docs), there is no guaranteed response time unless you pay Waldmann EDV or your hosting provider.

Finally, the dedup index must be managed. Repositories grow until compacted, the local cache needs care, and a first-time user who points Borg at a multi-terabyte dataset over a slow link should budget for a long initial run.

Who should pick this

We reach for Borg when the user is comfortable on a command line and wants the storage efficiency of deduplication without paying per gigabyte to a backup vendor. It is our pick for self-hosters, homelab users, and small teams backing up Linux servers, workstations, and NAS boxes to their own storage or to a plain hosted repository. If you run a fleet and want encrypted, append-only, versioned backups with retention policies you control end to end, Borg is still the tool we compare everything else against.

Look elsewhere if you need a managed product with a web console, compliance certifications, or phone support, or if your recovery-time target demands bare-metal re-imaging rather than file restore. And if nobody on the team can read a man page under pressure, the tool's power will not save you.

Key takeaways

• BSD-licensed, free, and actively maintained since 2015 — the software itself costs nothing; you pay only for storage.
• Content-defined chunking plus optional lz4/zstd/lzma compression makes repeated full backups dramatically cheaper in space than tarballs or rsync mirrors.
• Authenticated encryption with a documented "untrusted server" model, plus append-only repositories that resist ransomware-style deletion.
• Repositories work on any filesystem, including plain SSH targets; there is no native S3 mode in 1.4.x — plan accordingly.
• Command-line first: GUIs exist as third-party projects, and there is no bundled monitoring, scheduling, or enterprise agent set.
• Our pick in backup solutions for technical users who value storage efficiency and control over hand-holding.

Sources

• https://www.borgbackup.org/
• https://pypi.org/project/borgbackup/
• https://borgbackup.readthedocs.io/en/stable/internals/data-structures.html
• https://borgbackup.readthedocs.io/en/stable/internals/security.html
• https://www.borgbase.com/