Most cloud storage articles mention rclone in a single paragraph, somewhere in a list of “tools people use,” right next to restic and Duplicati, and move on. That undersells it. rclone is the actual workhorse that moves the data in the cloud cold storage comparison and the 3-2-1 backup strategy on this site, and it has enough sharp edges and genuinely useful features that it deserves its own walkthrough instead of a footnote. This is that walkthrough: how the commands actually differ, how to encrypt before anything leaves your network, and the one mistake that turns “backing up” into “deleting.”
What rclone actually is
rclone is a command-line tool that speaks an enormous number of storage backends through one consistent interface: local disk, SFTP, Backblaze B2, Wasabi, S3 and S3-compatible endpoints, Google Drive, OneDrive, and dozens more. You configure each one as a named “remote” with rclone config, and from then on every command looks the same regardless of what’s actually on the other end. Syncing to Backblaze and syncing to a friend’s NAS over SFTP use the identical command structure, just a different remote name. That consistency is the actual value. You learn the tool once and it works everywhere you’d plausibly want to send a data hoard.
It’s not a backup tool in the restic or Borg sense. It doesn’t do deduplication, it doesn’t version files automatically, and by default it doesn’t encrypt anything. It’s closer to a very capable, remote-aware version of rsync that happens to understand object storage APIs. Treat it as the transport layer, and decide separately how you want versioning and encryption handled on top of it.
copy, sync, and the command that can delete your remote
This is the part that actually matters more than any flag tuning below it. rclone has three core commands that look similar and behave very differently:
rclone copy source: dest:copies files from source to destination. It adds and updates, but never deletes anything on the destination. Safe by default.rclone sync source: dest:makes the destination match the source exactly, which means it will delete files on the destination that don’t exist in the source anymore. This is what you want for a true mirror, and it is also how people have wiped out a cloud bucket of years of backups in one command, because they ran it with the source and destination reversed, or ran it against a bucket that had drifted for a legitimate reason they’d forgotten about.rclone moveis like copy, but deletes the source files after a successful transfer. Useful for staging-to-archive workflows, dangerous for the same reason as sync if pointed the wrong way.
The discipline that actually prevents a bad day: run every sync and move command with --dry-run first, read the output, and only drop the flag once you’ve confirmed the list of adds/deletes matches what you expected. It costs ten seconds and it is the single highest-value habit in this entire article. Pair it with --backup-dir on sync jobs, which moves anything that would be deleted into a separate folder instead of deleting it outright, giving you a grace period to notice a mistake before it’s unrecoverable.
Encrypting before it leaves your network
rclone has a backend called crypt that sits on top of any other remote and encrypts file names and contents before they’re uploaded, using the destination remote purely as dumb storage for ciphertext. You set it up once in rclone config: create the crypt remote, point it at an existing remote (say, your Backblaze B2 bucket), set a password (and optionally a second “salt” password), and from then on every file you copy into the crypt remote is encrypted locally before it ever hits the network.
This matters for the same reason client-side encryption matters anywhere else: the provider is storing something they can’t read even if their systems were breached or a staff account was compromised. The tradeoff is real too. Encrypted filenames mean you can’t browse the bucket contents through the provider’s own web console and recognize anything, and losing the crypt password means the data is gone just as thoroughly as if the bucket itself had been deleted, with no recovery path from the provider’s side. Write that password down somewhere durable and offline, the same way you’d treat a LUKS passphrase, not just in a note on the same machine doing the backups.
If you’re already using restic or Duplicati for a different part of your backup strategy, you don’t need rclone’s crypt layer on top of those, since both already encrypt client-side. Crypt remotes earn their place specifically when you’re using plain rclone sync/copy as the whole pipeline and want encryption without adding a second tool.
Keeping a home connection usable during a multi-terabyte push
The default behavior is to go as fast as the link allows, which is exactly wrong on a home connection where someone else is trying to video call while the first full sync of a media library chews through upload bandwidth. A few flags that actually get used in practice:
--bwlimit 10Mcaps combined transfer speed, and--bwlimit "08:00,512k 23:00,50M"sets a schedule, throttled during the day and wide open overnight.--transfers 4controls how many files move in parallel. Higher isn’t always better: against B2 or S3-compatible endpoints, 4-8 is a reasonable starting point before diminishing returns set in, and cranking it too high just means more concurrent connections fighting for the same capped bandwidth.--checkers 8controls how many file comparisons run in parallel when deciding what needs transferring, separate from the actual transfer count, and matters more on a library with hundreds of thousands of small files than on one with a few huge ones.--multi-thread-streams 4splits a single large file across multiple streams, which is the flag that actually matters for a handful of 50GB+ files rather than a pile of small ones, where the bottleneck often isn’t bandwidth but per-connection throughput caps on the provider’s side.
Verifying the data actually arrived intact
A sync that reports success isn’t the same as data that’s actually correct on the other end. rclone check source: dest: compares file sizes and hashes between both sides and reports anything that differs, without transferring anything. Run it after an initial full sync and periodically afterward, especially against providers where you’re paying for egress and don’t want to find out about a problem by restoring a corrupted file. Some remotes don’t support every hash type rclone prefers, in which case it falls back to size-only comparison and says so in the output. Read that output; a size-only check on a collection of similarly-sized video files is a much weaker guarantee than a real hash match.
Mounting versus syncing, and why they’re not interchangeable
rclone mount exposes a remote as a local filesystem via FUSE, which looks tempting as a way to treat cloud storage like just another drive. It works for occasional browsing and pulling back a single file, but it is a poor substitute for the actual sync workflow as a backup target. Mounted remotes introduce real latency per file operation, caching behavior that can surprise you (writes may not be durable the instant a program thinks they’ve completed), and nothing about a mount enforces the kind of scheduled, logged, verified transfer that an actual backup needs. Use mount for convenience access when you need it, and keep the real backup pipeline on sync/copy run from a script or scheduler, the same separation covered from the local-disk angle in the NAS OS comparison on this site between a share you browse and a job that actually protects your data.
Putting it together
A workable setup for most home data hoards looks like: a crypt remote layered over whichever cloud cold storage tier you landed on, a scheduled sync job run from cron or a systemd timer with --dry-run reviewed once when the job is first written and --backup-dir left on permanently as a safety net, bandwidth limits that match your actual connection and household usage patterns, and a periodic rclone check run separately from the sync job itself so a transfer problem doesn’t go unnoticed until the day you actually need the data back. None of this requires a second tool layered on top. It’s one program, configured deliberately once, and then left to run unattended, which is the entire point of automating an off-site copy in the first place.