Restic
restic project · backups · reviewed by Stan K. · Source: data snapshot · data as of 2026-11-12
The engine pick: parallel multi-client repos, S3/B2/SFTP/rclone backends, and rest-server --append-only — a compromised client cannot wipe the repo.
#2 of 12 in Backups
ToolScore breakdown — why 88/100
10 criteria × 10 points. Low pillars are visible on purpose — that's the point of a score you can audit. Methodology →
What it does well
- ▸Parallel multi-client repos — backup takes a non-exclusive lock; prune/check go exclusive
- ▸rest-server --append-only: new backups yes, deletion and modification no — a compromised client cannot wipe the repo
- ▸S3/B2/SFTP/rclone backends — the BYO-storage champion of this board
- ▸Permissions and ownership always saved; zstd compression since repo format v2 (restic ≥0.14, levels from 0.19)
Where it costs you
- ▸Unix birth time is not saved
- ▸Hardlink semantics on restore are undocumented
- ▸Append-only is a deployment choice — you must run rest-server with --append-only to get it
Hidden traps — the stuff the pricing page doesn't say
The repo is ransomware-immune only if you deploy rest-server with --append-only. Run the plain server and a compromised client can prune its own history like any other change.
Unix birth time is not saved, and hardlink semantics on restore are undocumented. Neither touches file contents — but know both before a restore contract depends on either.
Know the limits before you pay: read the traps above, then decide with eyes open.
Visit official siteAffiliate link where available — never affects the score.
- official Restic docs and repo (official, checked 2026-11-12) — v0.19.1 (2026-07-05) verified
- official rest-server — append-only mode (official GitHub) verified
Data as of 2026-11-12. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
Object storage at $6.95/TB/mo that any engine can write to — Object Lock WORM defense with no extra
$99/yr for one computer with unlimited data — the least thinking per dollar in the industry, as long
The compression workhorse (zstd/lzma) over an SSH-only transport, with repo-side append-only immunit
Restic review
In-depth analysis · last updated 2026-11-12
The engine the hub rides
Restic is the engine pick of the backups board, and the hub's own recommended stack is built on it: B2 at $6.95/TB/mo (or e2 at $6/TB/mo) plus Restic gives you parallel clients, append-only rest-server, and Object Lock underneath for less than a coffee per TB. The license is BSD-2-Clause; the current release is v0.19.1 (2026-07-05).
Parallel where Borg is serial
The structural fact: backup takes a non-exclusive lock, so multiple clients write to one repo concurrently — prune and check are the operations that go exclusive. Backends are S3, B2, SFTP and rclone, which means every paid storage row on this board is a valid target. Permissions and ownership are always saved, and zstd compression has been in the repo format since v2 (restic ≥0.14, with levels exposed from 0.19).
The ransomware line
The defense that earns the score: rest-server --append-only. New backups are accepted; deletion and modification are refused — a compromised client cannot wipe the repo. That is the same class of protection as Object Lock on B2/e2, and the two stack: append-only at the protocol layer, Object Lock at the storage layer.
How the pillars land
Price/Value 10 — the engine is $0; the bill is electricity plus a bucket. Retention & versioning 10 — retention is fully yours (prune policy), with the append-only wall available server-side. Speed limits 9 — parallel multi-client writes and zstd compression are the throughput facts on record. Encryption model sits at the neutral 5 — this snapshot records no cipher facts; what the docs do record is metadata fidelity (permissions and ownership always saved). BYO storage (B2/S3) 10 — S3/B2/SFTP/rclone is the widest backend list among the engines alongside Kopia. Feature set 9.5 — everything above, docked half a point for the two documented gaps (birth time not saved; hardlink-on-restore semantics undocumented). Track record 9 — v0.19.1 (2026-07-05) with a documented format lineage (zstd since ≥0.14, levels from 0.19). Ease of use 7.5 — a CLI engine: powerful and scriptable, but there is no GUI in this snapshot's facts (that is Vorta-on-Borg's lane). Pricing transparency 10 — the project documents its own gaps in release notes; nothing here is marketing. Support & docs 7 — docs and the rest-server repo are verified sources; no commercial support exists to score.
Verdict — OUR ENGINE PICK, 88/100
If you are pairing an engine with a bucket, this is the default answer: parallel multi-client repos, a backend for every storage row on the board, and an append-only wall at the server. Deploy rest-server with --append-only from day one, and read the two fidelity gaps before your first restore.