zfsvault behaves like a remote host with ZFS. You log in over SSH as your account name and run the usual zfs commands. There is no shell and no agent to install. In the examples, replace ACCOUNT with the account name you chose at sign-up. It is your SSH user and the top-level name of your datasets.
1. Create an account. On the machine that will send data, print its public key (run ssh-keygen -t ed25519 first if it has none, with no passphrase if a scheduler runs the sends). In the dashboard, open SSH keys and paste it in. Use one key per machine.
$ cat ~/.ssh/id_ed25519.pub
2. The first time you connect, ssh asks whether to trust the host. Answer yes only if the fingerprint it shows matches this one:
SHA256:5lOcaGaCEAgN9ygZQbuHkjgSjAotzl5EIMQ2g6OsOgg
3. Send a snapshot. The first send is full, later ones are incremental. -s keeps progress if the connection drops. Encrypted dataset? Add -w to every send below. Not encrypted? Run them exactly as shown.
$ zfs snapshot tank/data@monday $ zfs send tank/data@monday | ssh ACCOUNT@ssh.zfsvault.com zfs receive -s ACCOUNT/data $ zfs snapshot tank/data@tuesday $ zfs send -i @monday tank/data@tuesday | ssh ACCOUNT@ssh.zfsvault.com zfs receive -s ACCOUNT/data
Not sure? zfs get encryption tank/data says anything but off for an encrypted dataset. Sent without -w, an encrypted dataset leaves your machine decrypted. More on encrypted data.
Every dataset lives under your account name: ACCOUNT/anything/you/like. You do not need to create it first. The first receive creates it. These send options work: -c (compressed), -L (large blocks), -e (embedded data), -p (properties) and -R (replication stream: you can send it to us but not back out).
syncoid takes care of snapshots, incrementals and resuming. Run it from cron or a systemd timer:
$ syncoid --recursive --no-privilege-elevation \
--sshport=22 --no-sync-snap \
tank/data ACCOUNT@ssh.zfsvault.com:ACCOUNT/dataRun it as root on your machine. --no-privilege-elevation tells syncoid there is no sudo on our side, and --no-sync-snap stops it creating its own snapshot, which zfsvault cannot do. To restore, swap source and target and keep both flags.
sanoid takes and thins snapshots on your machine. syncoid ships them. Keep sanoid on the source and let syncoid send what it makes (--no-sync-snap uses sanoid's snapshots instead of its own). Retention on our side is auto-prune, so the sending key does not need delete rights.
# /etc/sanoid/sanoid.conf
[tank/data]
use_template = production
[template_production]
hourly = 24
daily = 30
autosnap = yes
autoprune = yes
$ sanoid --cron
$ syncoid --no-privilege-elevation --sshport=22 --no-sync-snap tank/data ACCOUNT@ssh.zfsvault.com:ACCOUNT/dataTrueNAS replicates with zettarepl. The steps below set it up from the web interface.
ssh.zfsvault.com, port 22, username ACCOUNT, that keypair, cipher Standard. Discover the remote host key and check it matches the fingerprint on the Connect page.ACCOUNT/name. For encrypted datasets keep Include Dataset Properties ticked. That setting makes TrueNAS send them raw, so we hold only ciphertext. Without it, TrueNAS sends them readable. Do not enable Hold Pending Snapshots.pve-zsync has no port, key or user option for the connection, so give the Proxmox host an ssh alias for us and pass the account with --dest-user. Create the destination dataset once (ssh -p PORT ACCOUNT@HOST zfs create ACCOUNT/pve) and install cstream on the Proxmox host.
# /root/.ssh/config
Host zfsvault
HostName ssh.zfsvault.com
Port 22
IdentityFile ~/.ssh/zfsvault_ed25519
IdentitiesOnly yes
$ pve-zsync sync --source 100 --dest zfsvault:ACCOUNT/pve --dest-user ACCOUNT --verbose --maxsnap 7 --name nightly--maxsnap makes pve-zsync destroy old snapshots on the destination, so its key needs permission to destroy. Give it Full access with Applies to set to pve, not the whole account. Schedule the same arguments with pve-zsync create. Proxmox Backup Server is a different product and is not a target here.
znapzend has no port option either, so give it an ssh alias with the user and key, create the destination dataset once (znapzend does not), and point the plan at ACCOUNT@alias:ACCOUNT/data:
# ~/.ssh/config
Host zfsvault
HostName ssh.zfsvault.com
Port 22
User ACCOUNT
IdentityFile ~/.ssh/zfsvault_ed25519
IdentitiesOnly yes
$ ssh zfsvault zfs create ACCOUNT/data
$ znapzendzetup create SRC '7d=>1h' tank/data DST:a '30d=>1d' ACCOUNT@zfsvault:ACCOUNT/data
$ znapzend --runonce=tank/dataznapzend applies its retention on the destination too, which is a destroy there: give its key Full access with Applies to set to data, not the whole account.
Encryption is optional, and we recommend it. Send an encrypted dataset raw, with -w, and the data stays encrypted on the wire and at rest. We never see a key, a file name or any file content. Dataset and snapshot names and sizes stay visible to us. Data sent plain works the same way, with the same features. It is stored as sent, and anyone who can read our storage can read it. Plain is fine for data you would not mind us holding. Use -w only for datasets that are encrypted.
There are three ways to get there:
zfs get encryption tank/data, then send it raw.# migrate (or create the replica) once
$ zfs snapshot -r tank/data@migrate
$ zfs send -R tank/data@migrate | zfs receive -o encryption=on \
-o keyformat=passphrase -o keylocation=prompt tank/data-enc
# then send the encrypted copy raw, first in full and afterwards incrementally
$ zfs send -w tank/data-enc@migrate | ssh ACCOUNT@ssh.zfsvault.com zfs receive -s ACCOUNT/data
$ zfs send -w -i @migrate tank/data-enc@next | ssh ACCOUNT@ssh.zfsvault.com zfs receive -s ACCOUNT/dataRaw datasets can only be sent back raw. You load the key on the machine that receives them. The file browser cannot open them because we cannot read them.
Old snapshots pile up, and pruning them normally needs the right to destroy. Give your backup tool a Backup only key instead and set rules under Auto-prune in the dashboard. A rule has a name, the datasets it covers, the snapshots it looks at (both by name pattern, with optional exceptions) and how many to keep: the newest N, or N days, in each dataset.
Home daily datasets ACCOUNT/backups/* except ACCOUNT/backups/scratch
snapshots daily-*, autosnap_*_daily
keep the newest 30, in each datasetDatasets you add later are covered when their names match, so there is nothing to set up per dataset. A dry run shows which snapshots a rule would remove before anything happens. You can switch a rule off without deleting it. You can have up to 20 rules.
A snapshot is kept if any rule that matches it keeps it. Snapshots no rule matches, and the newest snapshot of a dataset, are never touched. What a rule removes goes to Recently destroyed and can be restored for your retention period (7 days by default). A rule waits an hour after it is saved or switched on, and a rush of matching snapshots pauses it. Rules are set on the website, not over SSH.
Settings has a Monitoring link for your own monitor (Healthchecks, Uptime Kuma, UptimeRobot, a cron job). It answers 200 with one line while your newest backup is recent enough and 503 when it is not, so you are alerted by phone or chat when a backup stops.
$ curl --fail https://ssh.zfsvault.com/ping/<token> ok: newest backup 3 hours ago (ACCOUNT/data@daily-12)
"Recent enough" is 26 hours. Add ?max=50 for 50 hours, or ?ds=ACCOUNT/data* to count only matching datasets. Anyone with the link sees how old your newest backup is, and nothing else.
List what you have, then send any snapshot back, in full or incrementally from an earlier one:
$ ssh ACCOUNT@ssh.zfsvault.com zfs list -r -t snapshot ACCOUNT/data $ ssh ACCOUNT@ssh.zfsvault.com zfs send ACCOUNT/data@tuesday | zfs receive tank/restored $ ssh ACCOUNT@ssh.zfsvault.com zfs send -i @monday ACCOUNT/data@friday | zfs receive tank/restored
Restore traffic (zfs send and web downloads) is metered (see pricing). Receiving data is free.
With zfs receive -s, an interrupted transfer keeps what arrived. Resume it with the token on the receiving dataset:
$ TOKEN=$(ssh ACCOUNT@ssh.zfsvault.com zfs get -H -o value receive_resume_token ACCOUNT/data) $ zfs send -t "$TOKEN" | ssh ACCOUNT@ssh.zfsvault.com zfs receive -s ACCOUNT/data # or give up on it $ ssh ACCOUNT@ssh.zfsvault.com zfs receive -A ACCOUNT/data
$ ssh ACCOUNT@ssh.zfsvault.com zfs list -t all -r ACCOUNT $ ssh ACCOUNT@ssh.zfsvault.com zfs destroy ACCOUNT/data@monday $ ssh ACCOUNT@ssh.zfsvault.com zfs destroy ACCOUNT/data@monday%wednesday # a range $ ssh ACCOUNT@ssh.zfsvault.com zfs rename ACCOUNT/data@a ACCOUNT/data@b $ ssh ACCOUNT@ssh.zfsvault.com zfs rollback -r ACCOUNT/data@tuesday
In the dashboard you can also browse snapshots and sizes, look inside a snapshot and download single files, and copy the restore commands. File browsing is not available for datasets sent raw (encrypted): we never hold their key.
A snapshot or dataset you destroy leaves your listings at once, but we keep what it held for 7 days by default (0 to 30 days, set in Settings or per dataset). It sits under Recently destroyed and Restore puts it back with its name and place in the history. It is stored, so it is billed, until the date shown.
Delete now removes it for good, ending the restore window. It asks for your password and a two-factor code, and it cannot be undone. We email you when you use it. Data stored for less than 30 days stays on your bill until day 30.
A lock turns Delete now off. Set it on your account in Settings, or on a single dataset. While locked, nothing destroyed can be deleted early, the time it is kept can only be made longer, and the account cannot be deleted. Malware on your machine, or someone with your login, can then destroy snapshots but cannot erase your history.
Locking takes effect at once. Unlocking takes effect 7 days after you ask. We email you straight away, and you can cancel. New accounts start unlocked. We recommend locking your backups.
-i), all intermediates (-I), -c -L -e -w -p, and -R when sending to us. All of them can resume.zfs list get set inherit receive send destroy rollback rename create version, and zpool list get statuszfs send -R out of zfsvault. Interactive shells and port forwarding are disabled.-w) sends are stored as ciphertext and we hold no keys for them. Datasets sent without -w are stored as sent, so we can read them.