Duplicacy

Last reviewed Sep 13, 2026 · 897 words

Duplicacy is not open source, and that is the first thing to settle. The CLI is source-available on GitHub under a licence that allows free personal use and requires a paid licence for commercial use; the Web Edition GUI is a separate paid product, about 20 USD for the first year per personal licence at last check, with cheaper renewals. That is why the catalogue lists it as Proprietary next to a 5,600-star repo. If you require an OSI licence on principle, stop here and use restic or Kopia. If you can live with source-available, Duplicacy does one thing better than either: many computers backing up into one storage bucket, deduplicating against each other, with no lock file and no coordination.

Lock-free deduplication is a real advantage, not marketing

Every chunk-based backup tool splits files into content-defined chunks and stores each chunk once. The hard problem is pruning: when two machines share a storage and one deletes old snapshots, it must not remove a chunk the other still references. Restic and Borg solve this with exclusive locks, which is why a Borg repository is per-machine and a shared restic repo needs care. Duplicacy names chunks by their hash, uploads them as immutable files, and uses a two-step "fossil collection" during prune that only deletes a chunk after every other repository has been seen to complete a newer backup without it. No lock, no repository-wide index to rebuild, and a laptop losing Wi-Fi mid-backup leaves nothing to repair.

The practical payoff is a household or small office: 4 laptops and a server all pointing at one B2 bucket, with the operating system files and shared documents stored once. Kopia offers the same shared-repository model with an Apache-2.0 licence, and it is the closest competitor; restic can do it with rclone and discipline, and Borg cannot.

The CLI is four commands

Initialise a repository from the directory you want to back up, then back up, prune and check on a schedule:

cd ~/Documents
duplicacy init -e laptop-docs b2://my-backups     # -e sets a storage password
duplicacy backup -threads 4
duplicacy prune -keep 0:365 -keep 30:90 -keep 7:30 -keep 1:7
duplicacy check

init writes a .duplicacy/ folder into the directory with a preferences file and the repository ID (laptop-docs above), which is the name other machines will see when deduplicating and pruning. The -keep grammar reads as "keep one every N days for snapshots older than M days," so the line above keeps daily for a week, weekly for a month, monthly for a year and nothing older. Run check weekly; it verifies every referenced chunk exists, and check -chunks actually downloads and hashes them, which costs egress but is the only real proof.

Backends cover S3 and compatibles, Backblaze B2, Google Drive, OneDrive, Dropbox, Google Cloud Storage, Azure, Wasabi, Storj, SFTP, WebDAV and local disk. The consumer cloud-drive support is the other reason people pick it over restic, which needs rclone in between for those.

The Web Edition is worth paying for on a desktop

The CLI has no scheduler, so on a laptop you are writing cron or Task Scheduler entries yourself. The Web Edition wraps the same engine in a local web UI with schedules, email notifications, restore browsing and per-storage settings, and it is the version most people actually run day to day on Windows and macOS. On a Linux server, saspus/duplicacy-web is the widely used community container; map /config, /cache, /logs and the source directories, and the UI listens on port 3875. It is a licensed product, so the container prompts for the licence on first launch.

Restores work from either edition: duplicacy restore -r 42 -stats in a fresh directory pulls snapshot revision 42, and the Web Edition lets you browse a revision and pick files. Test one restore before you trust it, the same as every backup tool in the backup category.

What it does not do

There is no server component and no push-from-client agent model, so central management of 50 machines means 50 Web Edition licences and 50 configs. There is no built-in block-level image backup, no bare-metal restore, and no snapshot browsing over FUSE the way restic and Borg offer mounts. Metadata such as extended attributes is preserved on macOS and Linux but Windows ACLs are not, so it is a file backup, not a system backup. And the storage password is the encryption key: lose it and the chunks are noise.

What I'd do

For one Linux server backing up to one bucket, use restic; it is free in every sense, the tooling around it is larger, and the 3-2-1 pattern is easier with a tool every guide already covers. For a household of Mac and Windows laptops plus a server that should all deduplicate into one B2 bucket with a GUI the family can use, buy Web Edition licences, put the server on the CLI with a cron, and run check weekly. If the licence is a dealbreaker but the shared-repository model is the draw, Kopia is the answer and it is the one I would point a new self-hoster to first.