vaultik/README.md

vaultik (ваултик)

vaultik is an incremental backup tool. It encrypts your files with a public key and uploads them to an S3-compatible object store, a local directory or an rclone remote.

The machine being backed up holds only the public key. It cannot decrypt its own backups. You keep the private key somewhere else and need it only to restore.

Why

Other backup tools, such as restic, borg and duplicity, are designed for machines that can hold a decryption key. vaultik is for machines that should not hold one.

It was written to meet these requirements together:

• open source
• no passphrase or private key on the backed-up machine
• incremental
• compressed
• encrypted
• direct upload to S3-compatible storage, with no intermediate step or tool

Features

• Encryption with age (X25519 and ChaCha20-Poly1305), to one or more public keys.
• Incremental backups. A file is read again only when it has changed.
• Deduplication. Identical content is stored once, across files and across snapshots.
• zstd compression at a level you choose.
• Small files are packed together into large objects. This keeps the number of S3 requests low.
• File contents, names, paths and metadata are encrypted at the destination.
• Three kinds of destination: S3, a local directory and any rclone remote.
• Regular files, symlinks and directories are backed up, with their permissions, owners and modification times.
• Pure Go, with no C code. It builds for Linux and macOS on amd64 and arm64.

Terms

• Snapshot: one backup of the paths listed under one name in the config, taken at one time.
• Snapshot ID: the name of a snapshot, in the form <hostname>_<snapshot-name>_<timestamp>. An example is server1_home_2025-06-01T12:00:00Z.
• Destination: the place the backups are stored.
• Chunk: a piece of a file's content. vaultik cuts each file into chunks and stores each distinct chunk once.
• Blob: one object at the destination. It holds many chunks, compressed and then encrypted.
• Local index: a SQLite database on the backed-up machine. It records what is already stored, so the next backup uploads only what changed.
• Manifest: the list of blobs one snapshot uses. It is stored at the destination without encryption.
• Remote key: a hash of the snapshot ID. It names the snapshot's directory at the destination, so the destination never shows the snapshot ID.

Install

go install sneak.berlin/go/vaultik/cmd/vaultik@latest

This needs Go 1.26.1 or later. Generating a key pair needs age-keygen, which comes with age.

Quick start

# Write a default config file. The command prints the path it wrote.
vaultik config init

# Generate a key pair. Keep this file somewhere safe, away from this machine.
age-keygen -o vaultik_backup_private_key.txt
grep 'public key' vaultik_backup_private_key.txt

# Set the public key and the destination.
vaultik config set age_recipients.0 age1YOUR_PUBLIC_KEY_HERE
vaultik config set storage_url "file:///Volumes/usbstick/mybackup"

# Create the destination directory. vaultik never creates it.
mkdir /Volumes/usbstick/mybackup

# Run the first backup.
vaultik snapshot create

# See what is stored.
vaultik snapshot list

The default config is written for macOS. It backs up your home directory and /Applications, with a list of excludes. A configured path that does not exist stops the backup. On another system, run vaultik config edit first and change the snapshots section.

On macOS, grant your terminal app Full Disk Access before the first backup. The setting is under System Settings, Privacy & Security, Full Disk Access. Without it the backup stops with a permission error on a protected directory.

Everyday use

# Check that a snapshot's blobs are all present, with the listed sizes.
vaultik snapshot verify <snapshot-id>

# Put the private key in the environment. Reading it from the file keeps it
# out of your shell history.
export VAULTIK_AGE_SECRET_KEY="$(cat vaultik_backup_private_key.txt)"

# Download, decrypt and re-hash every blob of a snapshot.
vaultik snapshot verify --deep <snapshot-id>

# Restore a snapshot into a new directory that only you can write.
vaultik snapshot restore <snapshot-id> ~/vaultik-restore

# A daily cron line: back up, then keep four weeks of snapshots.
# 0 3 * * * vaultik snapshot create --cron --prune --keep-newer-than 4w

--prune deletes data at the destination. Read Sharing a destination before you use it on a destination that more than one machine or user writes to.

Restoring on another machine

You can restore on a machine that never ran the backup, such as a replacement for a lost one. That machine needs three things:

• the vaultik binary
• the age private key
• the destination, and credentials that can list and read it

It does not need the original local index, config file or hostname.

# Create a config and point it at the original destination.
vaultik config init
vaultik config set storage_url "s3://bucket/prefix?endpoint=https://s3.example.com"
vaultik config set s3.access_key_id "..."
vaultik config set s3.secret_access_key "..."

# See what the destination holds. This needs no private key.
vaultik snapshot list

On this machine the local index is empty, so the list cannot show snapshot IDs. Each snapshot appears as <remote only:...>, with the first 12 characters of its remote key. The snapshot ID is stored only in the local index and in the snapshot's encrypted database.

Use that remote key in place of the snapshot ID:

export VAULTIK_AGE_SECRET_KEY="$(cat vaultik_backup_private_key.txt)"

# Restore everything, then check the chunk hashes of every restored file.
vaultik snapshot restore --verify <remote-key> ~/vaultik-restore

# Optionally, also check every blob of the snapshot at the destination.
vaultik snapshot verify --deep <remote-key>

Any leading part of the remote key works, as long as it matches one snapshot. snapshot list --json gives the full 64-character key as remote_key.

A machine that only restores does not need age_recipients.

Commands

vaultik config init
vaultik config edit
vaultik config get <key>
vaultik config set <key> <value>
vaultik snapshot create [snapshot-names...] [--cron] [--prune] [--keep-newer-than <duration>]
vaultik snapshot list [--json]
vaultik snapshot verify <snapshot-id> [--deep] [--json]
vaultik snapshot purge (--keep-latest | --older-than <duration>) [--snapshot <name>...] [--force]
vaultik snapshot remove <snapshot-id> [--dry-run] [--force] [--local-only] [--json]
vaultik snapshot restore <snapshot-id> <target-dir> [paths...] [--verify]
vaultik prune [--force] [--json]
vaultik info
vaultik remote info [--json]
vaultik remote nuke --force
vaultik database delete [--force]
vaultik completion <bash|zsh|fish|powershell>
vaultik version

snapshot list has the alias snapshot ls. snapshot remove has the alias snapshot rm.

What every command needs

config init, completion and version need nothing.

config edit, config get and config set need a config file that exists. It does not have to be complete.

Every other command needs a config file that loads. See Configuration for the checks.

Every other command except database delete also needs these:

• Write access to the local index. The command creates the file, and its directory, when they do not exist.
• For an rclone:// destination, an rclone config file that defines the remote.

Each command below lists what it needs beyond this.

No command needs root. A backup can read only what its user can read. A restore sets file owners only when it runs as root.

Global flags

• --config <path>: the config file to use. See Where the config file is.
• --verbose, -v: write informational log records to stderr.
• --debug: write debug log records to stderr.
• --quiet, -q: print only warnings, errors and the output a script reads. See Output.
• --skip-errors: skip a file that cannot be read during a backup, or cannot be restored during a restore. Without it the command stops at the first such file.

--skip-errors applies to snapshot create and snapshot restore only. A command that skipped anything still finishes, and then exits non-zero.

Durations

--keep-newer-than, --older-than and prune_grace_period take a duration.

• d: days.
• w: weeks.
• mo: months of 30 days.
• y: years of 365 days.
• h, m and s: hours, minutes and seconds, as in 36h or 90m.

m means minutes and mo means months. Days, weeks, months and years take whole numbers and can be combined, as in 2w3d.

prune_grace_period accepts only hours, minutes and seconds.

config init

Writes a default config file, with a comment that explains each setting it lists.

• It writes to the path from --config, or else from VAULTIK_CONFIG.
• Otherwise it writes to config.yml in the platform config directory. That is ~/.config/vaultik/ on Linux and ~/Library/Application Support/vaultik/ on macOS.
• As root it writes to /etc/vaultik/config.yml instead.
• It creates the file with mode 0600, because the file can hold credentials.
• It refuses to overwrite a file that exists.

config edit

Opens the config file in $EDITOR. It uses vi when EDITOR is not set.

config get <key>

Prints one value from the config file. The key is a dotted path, as in vaultik config get storage_url or vaultik config get age_recipients.0. A list or a map prints as YAML.

config set <key> <value>

Sets one value in the config file. The key is a dotted path, as in vaultik config set compression_level 9.

• Comments and formatting in the file are kept.
• A key that does not exist yet is created. A missing parent becomes a list when the next part of the path is a number, and a map otherwise.
• In a list, the index one past the end appends an entry.
• The command prints the key it set, and never the value.
• It sets the file's mode to 0600 when group or others could read it.

The command refuses a change that would stop the config from loading. An example is vaultik config set compression_level 99. The file is then left unchanged.

A config that does not load yet is checked less strictly. An example is the file config init writes, which has no storage_url. There the command refuses only a value of the wrong type, such as a word where a number belongs.

snapshot create

Backs up the configured snapshots. With no names it backs up all of them. Each name gets its own snapshot.

It needs:

• At least one public key in age_recipients.
• Read access to the configured paths.
• Permission to write objects at the destination. See Destination permissions.
• Free temporary space. See Temporary files.
• The lock.

It does not need the private key.

A file is read again when any of these has changed since the last backup:

• its size
• its modification time
• its mode
• its owner or group
• its link target

Every other file is recorded in the new snapshot without being read.

Flags:

• --cron: print nothing when the backup succeeds. Warnings and errors are still printed.
• --prune: delete old snapshots and unused data after the backup. See below.
• --keep-newer-than <duration>: with --prune, keep every snapshot newer than this.

The backup stops before it writes anything in these cases:

• A file:// destination directory does not exist.
• The destination has no marker file, although the local index holds snapshots. See The marker file.
• The destination in the config is not the one the local index was used with. See Changing the destination.

The backup also stops when a configured path is a symlink whose target does not exist. It stops when a configured path does not exist, unless --skip-errors is given.

With --skip-errors, a file or directory that cannot be read is skipped. The snapshot is kept and counts each skipped entry as failed, and the command exits non-zero. An error while packing or uploading still stops the backup.

A file that is removed during the backup is counted as failed. It does not change the exit status.

--prune

After a successful backup, --prune does two things:

• It removes the older snapshots of each name it just backed up. It keeps only the newest one. With --keep-newer-than, it keeps every snapshot newer than that duration instead.
• It then deletes at the destination what prune deletes.

--prune also needs permission to list, read and delete at the destination.

--keep-newer-than without --prune is an error. So is a value that is not a duration. Both are reported before the backup starts.

The prune does not run after a backup that may have missed data. An older snapshot may then hold the only copy of what was missed. This applies when:

• --skip-errors skipped a file or directory that could not be read.
• A configured path held nothing, as the mount point of an unmounted volume does.
• An earlier snapshot of the same name found a mount point that this backup did not find.

In each case every older snapshot is kept. The command says why, names the path and exits non-zero. It does so on every backup until the cause is gone.

A mount point here is a directory at or below a configured path that is on a different filesystem from its parent. It counts only when the backup finds an entry in it that is not excluded.

A backup does not find a mount point in these cases:

• The volume is not mounted.
• The directory holds no entry that the backup reads.
• The directory is now excluded.
• The directory is no longer under a configured path.

If the cause is intended, such as a retired volume, remove the older snapshots yourself with snapshot purge. Take a path that is empty on purpose out of the config.

snapshot list

Lists the snapshots in the local index and the snapshots at the destination, as one table. Before the table, the command prints the snapshot names and paths in the config.

It needs permission to list and read at the destination. It needs no private key.

The table has these columns:

• SNAPSHOT ID
• TIMESTAMP: in UTC.
• COMPRESSED SIZE: the total size of the blobs the snapshot uses.
• UNCOMPRESSED SIZE: the total size of its distinct chunks before compression.
• NEW CHUNK SIZE: the size of the chunks this snapshot stored that were not stored before.

A snapshot at the destination that is not in the local index is shown as <remote only:...>. The dots stand for the first 12 characters of its remote key. Its last two columns read <remote only>, because those sizes come from the local index. Its timestamp is the time its backup finished.

The command reports these below the table:

• Snapshots in the local index that are missing at the destination. Run vaultik prune to drop those records.
• Remote-only snapshots whose manifest cannot be read. They are left out of the table.
• Remote-only snapshots beyond the first 1000. They are left out of the table.
• A local index that was used with a different destination. See Changing the destination.

When the destination cannot be listed, the command warns and shows the local index alone. It still exits zero. It does the same when the destination has no marker file although the local index holds snapshots.

--json prints a JSON array instead. Each entry has these fields:

• id: the snapshot ID. It is empty for a remote-only snapshot.
• remote_key: the full 64-character remote key.
• timestamp
• compressed_size
• uncompressed_size and new_chunk_size: left out for a remote-only snapshot.
• locally_tracked: whether the snapshot is in the local index.
• remote_present: whether the snapshot is at the destination. It is null when the destination could not be listed.

Under --json, every warning above goes to stderr.

snapshot verify <snapshot-id>

Checks a snapshot at the destination. The argument is a snapshot ID, a remote key or the leading part of a remote key.

Without --deep, it checks two things:

• Every blob the manifest lists is present, with the size the manifest records.
• The snapshot's encrypted database is present.

This reads no blob contents and needs no private key. It needs permission to read at the destination, and to list it when you give a remote key.

Flags:

• --deep: download, decrypt and re-hash every blob. See below.
• --json: print the result as JSON.

The command exits non-zero when a check fails or cannot be made.

--deep

--deep needs the age private key. Set it in the VAULTIK_AGE_SECRET_KEY environment variable, or as age_secret_key in the config file. The environment variable wins when both are set. See The private key.

--deep does these checks, in order:

• It downloads and decrypts the snapshot's database. It checks that this is the database of the snapshot you named.
• It checks that the manifest and the database list the same blobs, with the same sizes.
• It checks that every blob is present with the recorded size.
• It downloads and decrypts every blob. It checks the hash of each chunk, and the hash that names the blob.

It stops at the first blob that fails.

--deep proves that the stored data is intact and decrypts with your key. It does not prove who wrote the data.

--deep writes the decrypted database to a temporary directory and removes it afterwards. See Temporary files.

snapshot purge

Removes old snapshots. You must give exactly one of --keep-latest and --older-than.

It needs the lock, and permission to list and delete at the destination. It needs no private key.

Flags:

• --keep-latest: keep only the newest snapshot of each name.
• --older-than <duration>: remove every snapshot older than this.
• --snapshot <name>: act only on snapshots with this name. Repeat the flag for more names.
• --force: do not ask for confirmation.

The command acts only on snapshots in the local index. --keep-latest keeps the newest snapshot of each name, not the newest overall.

The command removes each such snapshot from the local index. It also deletes the snapshot's directory under metadata/ at the destination. Its blobs stay. Run vaultik prune afterwards to delete the blobs that no snapshot uses any more.

The command first drops local records of snapshots that are no longer at the destination. It fails and removes nothing when the destination cannot be listed or has no marker file.

snapshot remove <snapshot-id>

Removes one snapshot. The argument must be a snapshot ID. A remote key is not accepted.

It needs the lock, and permission to list and delete at the destination. It needs no private key.

Flags:

• --force, -f: do not ask for confirmation.
• --dry-run: say what would be removed, and remove nothing.
• --local-only: remove the snapshot from the local index only. The destination is not changed.
• --json: print the result as JSON. This also skips the confirmation.

The command removes the snapshot from the local index. It also deletes the snapshot's directory under metadata/ at the destination. Its blobs stay. Run vaultik prune afterwards to delete the blobs that no snapshot uses any more.

An ID that is in neither the local index nor the destination is an error. With --local-only, the snapshot must be in the local index.

The deletion at the destination can fail, for example when the destination is unreachable or has no marker file. What happens then depends on where the snapshot was:

• A snapshot in the local index is still removed from it, with a warning. Run the same command again later to delete its directory at the destination. vaultik prune does not delete it.
• For a snapshot that is only at the destination, the failure is an error.

snapshot restore <snapshot-id> <target-dir> [paths...]

Restores files from a snapshot into <target-dir>. The first argument is a snapshot ID, a remote key or the leading part of a remote key.

It needs:

• The age private key, as for snapshot verify --deep. See The private key.
• Permission to read at the destination, and to list it when you give a remote key.
• Write access to <target-dir>. The command creates the directory when it does not exist.
• Free temporary space for the decrypted blobs. See Temporary files.

It does not need age_recipients, and it does not take the lock.

Each file is restored under its original absolute path inside <target-dir>. For example, /home/user/notes.txt is restored to <target-dir>/home/user/notes.txt.

With no paths, the command restores the whole snapshot. Each path names a file or a directory by its original absolute path:

vaultik snapshot restore <snapshot-id> ~/vaultik-restore /home/user/documents

A path that matches nothing is named in a warning. The command restores the other paths and then exits non-zero.

The command restores these:

• regular files, symlinks and directories, empty directories included
• permissions and modification times
• owners and groups, when it runs as root

Without root, restored files belong to the user who runs the command. The command then warns once if any owner or group differs from the snapshot.

A file that already exists in <target-dir> is replaced. Running the same restore again starts from the beginning.

Restore into a new directory that only you can write, not a shared one such as /tmp. No other user can then change the files while they are written.

Flags:

• --verify: after the restore, read every restored file and check its chunk hashes.

With --skip-errors, a file that cannot be restored is reported and skipped, and the command exits non-zero. --verify does not check a skipped file.

prune

Removes what is no longer needed, from the local index and from the destination.

It needs the lock, and permission to list, read and delete at the destination. It needs no private key.

It works in three steps:

• It drops each snapshot from the local index whose directory is gone from the destination.
• It removes unfinished snapshots from the local index. It also removes the file, chunk and blob records that no snapshot uses.
• It reads every manifest at the destination. It then deletes each blob that no manifest lists and that is older than prune_grace_period.

The third step also deletes a directory under metadata/ that has no manifest. A backup leaves one when it stops before it uploads its manifest. Such a directory is deleted only when all of these hold:

• Every object in it is older than prune_grace_period.
• Its snapshot is not in the local index.
• Its name has the form of a remote key.

prune_grace_period defaults to 24h. A newer blob is kept, because a backup that is still running may need it. See Sharing a destination for what the grace period does not protect.

The command deletes nothing at the destination in these cases:

• The destination cannot be listed.
• The destination has no marker file, although the local index holds snapshots.
• A manifest cannot be read.

It exits non-zero when a blob or a directory cannot be deleted.

The time to read the manifests grows with the number of snapshots.

Flags:

• --force: do not ask for confirmation before deleting at the destination.
• --json: print the result as JSON. This also skips the confirmation.

The JSON document has these fields:

• blobs_found: the unused blobs old enough to delete.
• blobs_deleted
• blobs_failed: blobs that could not be deleted. Left out when zero.
• bytes_freed
• blobs_too_new: unused blobs that were kept. They are newer than the grace period, or the destination gave no time for them.
• metadata_dirs_deleted
• metadata_dirs_failed: left out when zero.

info

Prints the system details, the settings in use and figures from the local index. The figures are the size of the index and its counts of snapshots, blobs and files.

remote info

Reports what the destination holds:

• the kind of destination and its location
• the size of each snapshot's database and manifest
• the number and size of all blobs
• the number and size of the blobs that manifests list
• the number and size of the blobs that no manifest lists

It needs permission to list and read at the destination. It needs no private key.

When a manifest cannot be read, the figures for blobs that no manifest lists are reported as unknown. That snapshot's own blob count and size are unknown too.

The command skips a name under metadata/ that is not a remote key, with a warning. It does not count a file under blobs/ whose name is not a blob hash.

--json prints the report as JSON. Unknown figures are null. unreadable_manifests lists the remote key of each manifest that could not be read. skipped_manifest_count counts the manifests under skipped names.

remote nuke --force

Deletes everything vaultik stored at the destination. This cannot be undone.

It needs --force, the lock, and permission to list and delete at the destination.

It deletes these, in order:

• every object under metadata/
• every object under blobs/
• the marker file

It also removes each snapshot from the local index, once that snapshot's directory at the destination is deleted.

The command fails before it deletes anything when the destination cannot be listed, or has no marker file although the local index holds snapshots.

If objects remain afterwards, the command names each one and exits non-zero. Running it again deletes what is left.

database delete

Deletes the local index file. The destination is not changed.

It needs a config file that loads, because the config gives the path of the index.

The next backup then reads every file again. It skips the upload of a blob that the destination already holds under the same name.

Use this command after you point the config at a different destination. See Changing the destination.

--force skips the confirmation.

completion <shell>

Prints a completion script for bash, zsh, fish or powershell.

# zsh, for the current session
source <(vaultik completion zsh)

# zsh, permanently
vaultik completion zsh > "${fpath[1]}/_vaultik"

# bash, for the current session
source <(vaultik completion bash)

# bash, permanently, on Linux
vaultik completion bash > /etc/bash_completion.d/vaultik

# fish
vaultik completion fish > ~/.config/fish/completions/vaultik.fish

version

Prints the version, the commit, the build date and the platform. It also says when the binary is a development build, not a release.

Configuration

The config file is YAML. vaultik config init writes one to start from, and config.example.yml is a longer example.

A command refuses to start when the config does not load. The config loads when all of these hold:

• Every entry in age_recipients is an age X25519 public key.
• snapshots has at least one name, and every name has at least one path.
• A destination is set, with credentials when it is S3.
• Every other setting below is within its limits.

vaultik warns when group or others can read the config file.

A value written as ${ENV:NAME} is replaced by the environment variable NAME. This keeps a credential out of the file.

A leading ~ is expanded to your home directory in snapshot paths, in index_path and in a file://~/... destination.

Where the config file is

vaultik uses the first of these that applies:

• the path given with --config
• the path in the VAULTIK_CONFIG environment variable
• config.yml in the platform config directory, which is ~/.config/vaultik/ on Linux and ~/Library/Application Support/vaultik/ on macOS
• /etc/vaultik/config.yml

A path from --config or VAULTIK_CONFIG must exist.

Settings

• age_recipients
  • The age public keys that backups are encrypted to.
  • Default: none. snapshot create needs at least one. No other command needs one.
  • Any one of the matching private keys can decrypt.
  • Adding a key later does not re-encrypt what is already stored. See Accepted risks.
• age_secret_key
  • The age private key.
  • Default: not set.
  • Only snapshot restore and snapshot verify --deep use it.
  • The VAULTIK_AGE_SECRET_KEY environment variable overrides it.
  • Do not set it on a machine you back up. See The private key.
• snapshots
  • The named sets of paths to back up. Required.
  • snapshots.<name>.paths: the paths to back up under this name.
  • snapshots.<name>.exclude: exclude patterns for this name only.
• exclude
  • Exclude patterns that apply to every snapshot. See Exclude patterns.
  • Default: none.
• storage_url
  • The destination. See Destinations.
  • Required, unless the s3 settings name the destination.
• s3.access_key_id and s3.secret_access_key
  • The S3 credentials. Required for an S3 destination.
• s3.region
  • The S3 region. Default: us-east-1.
• s3.part_size
  • The size of each part of a multipart upload. Default: 5MiB.
  • It must be between 5MiB and 5GiB.
  • A blob too large for 10,000 parts of this size is uploaded in larger parts.
• s3.endpoint, s3.bucket, s3.prefix and s3.use_ssl
  • An older way to name an S3 destination. They are used only when storage_url is not set.
  • s3.endpoint and s3.bucket are then required.
  • s3.use_ssl: false selects plain HTTP for an endpoint written without a scheme.
• chunk_size
  • The average size of a chunk. Default: 10MB. Minimum: 1MB.
  • A chunk is between a quarter of this size and four times this size.
  • A smaller size finds more duplicate content and makes the index larger.
• blob_size_limit
  • The largest size of a blob. Default: 10GB.
  • It must be at least four times chunk_size.
• compression_level
  • The zstd compression level, from 1 to 19. Default: 3.
  • Levels 1 to 2, 3 to 5, 6 to 9 and 10 to 19 form four groups. Levels in one group compress the same way.
• prune_grace_period
  • How old unused data must be before prune deletes it. Default: 24h.
  • It applies to a blob that no manifest lists. It also applies to every object in a directory under metadata/ that has no manifest.
  • 0s deletes such data at once. A negative value is refused.
  • See Sharing a destination.
• hostname
  • The hostname used in snapshot IDs. Default: the system hostname.
  • Only the part before the first . is used.
• index_path
  • The path of the local index.
  • Default: index.sqlite in the platform data directory. That is ~/.local/share/vaultik/ on Linux and ~/Library/Application Support/vaultik/ on macOS.

Sizes take a number of bytes or a unit. MB and GB are powers of 1000. MiB and GiB are powers of 1024.

Environment variables

• VAULTIK_AGE_SECRET_KEY: the age private key. See The private key.
• VAULTIK_CONFIG: the path of the config file. --config overrides it.
• VAULTIK_INDEX_PATH: the path of the local index. It overrides index_path.
• RCLONE_CONFIG: the path of the rclone config file, for an rclone:// destination.
• EDITOR: the editor that config edit starts.
• TMPDIR: the directory for temporary files.
• NO_COLOR: when set, status lines have no color.
• VAULTIK_CPUPROFILE: write a CPU profile to this path. For debugging.
• VAULTIK_MEMPROFILE: write a heap profile to this path on exit. For debugging.

The private key

snapshot restore and snapshot verify --deep decrypt data, so they need the age private key. No other command needs it.

Give the key in one of two ways:

• the VAULTIK_AGE_SECRET_KEY environment variable
• age_secret_key in the config file

The environment variable wins when both are set.

The value can be the whole file that age-keygen writes. Comments are ignored, and every key in the file is used.

Set the variable from the file, so that the key is not typed into your shell history:

export VAULTIK_AGE_SECRET_KEY="$(cat vaultik_backup_private_key.txt)"

Do not put the private key on a machine you back up. Anyone who breaks into that machine could then decrypt its backups. Supply the key only on the machine you restore on.

Exclude patterns

A pattern excludes a file or a whole directory from a backup. Patterns are matched against the path below the configured path.

• A pattern that starts with / matches from the configured path. In a snapshot of ~, /Library/Caches matches only ~/Library/Caches.
• Any other pattern matches at every depth. .cache matches every file or directory named .cache.
• * and ? match within one name. ** matches across directories.
• A pattern that matches a directory excludes everything in it.

exclude at the top of the config applies to every snapshot. exclude under a snapshot name adds patterns for that snapshot.

Destinations

storage_url selects one of three kinds of destination.

S3

s3://bucket/prefix?endpoint=s3.example.com&region=us-east-1

Any S3-compatible object store works.

• endpoint is the host of the store. The default is s3.amazonaws.com.
• region overrides s3.region.
• ssl=false selects plain HTTP for an endpoint written without a scheme.
• No other parameter is accepted.

The credentials come from s3.access_key_id and s3.secret_access_key in the config. A URL that carries credentials is refused.

Local directory

file:///path/to/backup

• The path must be absolute.
• A percent-escape in the path is decoded, so my%20stick names my stick.
• The directory must exist before the first backup. vaultik never creates it.

A directory on a removable volume is gone while the volume is not mounted. A backup then fails. It does not write to the disk under the mount point.

This does not hold when the destination is the mount point itself. The empty mount point stays behind when the volume is unplugged. There only the marker file stops the backup.

rclone

rclone://remote/path

This uses any storage provider that rclone supports. Define the remote first with rclone config. vaultik reads rclone's own config file.

The URL takes no parameters and no credentials.

Interrupted uploads

An upload that is cut off must not leave a partial object under the final name. A later backup would take that object for complete.

• S3 shows an object only once its upload has finished.
• The local directory destination writes a temporary file and then renames it.
• An rclone remote that can move objects is written the same way. The temporary name ends in .partial.
• An rclone remote that cannot move objects is written in place. On such a remote, a killed upload can leave a partial object under the final name.

A leftover .partial file is ignored. You can delete it.

Destination permissions

Each command needs different permissions at the destination.

• snapshot create needs only permission to write objects. A machine you back up can therefore hold credentials that can neither read nor delete.
• snapshot list, snapshot verify, snapshot restore and remote info list and read.
• snapshot purge, snapshot remove and remote nuke list and delete.
• prune and snapshot create --prune list, read and delete.

A machine that runs snapshot create --prune can delete its own backups. To keep it to write-only credentials, run prune from a separate, trusted machine.

A backup tries to look up the marker file and each blob before it writes. It goes on when the destination refuses those look-ups.

The marker file

Every backup writes vaultik-destination.txt at the top of the destination. The file holds one fixed line of text.

The marker tells an emptied or missing destination apart from a new one. An unplugged volume leaves an empty directory at its mount point. Without the marker, that directory would look like a destination whose snapshots are all gone.

The check applies while the local index holds snapshots. When the destination then answers that the marker does not exist, commands treat the destination as not present:

• snapshot create refuses to write to it.
• prune, snapshot purge and remote nuke fail.
• snapshot list shows the local index alone.
• snapshot remove deletes nothing at the destination.

To fix this, mount the volume. If you emptied the destination on purpose, run vaultik database delete and back up again.

The check is skipped when the look-up fails for any other reason. S3 refuses the look-up to credentials that can only write. A machine with such credentials therefore does not notice an emptied destination. Its next backup would leave out the data its local index records as stored. After you empty a destination, run vaultik database delete on such a machine before it backs up again.

Changing the destination

The local index belongs to one destination. It records which blobs are stored there.

The first command that changes data records the configured destination in the local index. Later, these commands refuse to run when the config names a different destination:

• snapshot create
• snapshot purge
• snapshot remove
• prune
• remote nuke

The destination that is compared is storage_url. When storage_url is not set, it is s3.endpoint, s3.bucket and s3.prefix.

There are two ways out:

• Change the config back to the earlier destination.
• Run vaultik database delete. The next backup then starts a new index for the new destination.

Sharing a destination

Give each machine and each user its own destination, such as its own prefix or directory. The rest of this section applies only when two of them share one.

A prune from one machine can meet a backup from another that is still running. A backup uploads its blobs first, then its database, and its manifest last. Until the manifest exists, no manifest lists the new blobs.

prune_grace_period protects those blobs. A prune keeps a blob that no manifest lists until the blob is older than the grace period. It keeps a directory under metadata/ that has no manifest for the same time. Set the grace period longer than the longest backup of any machine or user that shares the destination.

The grace period does not protect one case. A backup does not upload a blob that is already at the destination. Such a blob keeps the time of its first upload. It can be left over from an earlier backup that stopped before its manifest. A prune from another machine can then delete that blob while the new backup runs. The new snapshot cannot be restored afterwards.

remote nuke ignores the grace period.

Locking

Commands that change data take a lock, so that only one of them runs at a time. These commands take it:

• snapshot create
• snapshot purge
• snapshot remove
• prune
• remote nuke

A second one does not wait. It exits at once with an "already running" error.

The lock is the file vaultik.pid in the platform data directory. That is ~/.local/share/vaultik/ on Linux and ~/Library/Application Support/vaultik/ on macOS.

The lock covers one user on one machine. It does not stop another user or another machine from writing to the same destination. See Sharing a destination.

No other command takes the lock. snapshot list, snapshot verify, snapshot restore, info and remote info run while a backup is in progress.

Output

stdout and stderr

stdout carries the output you asked for:

• status and progress lines
• tables and reports
• JSON documents
• config get values
• completion scripts

stderr carries the rest:

• the startup banner
• log records, including those from --verbose and --debug
• the error that a failed command ends with

So vaultik snapshot list > out.txt captures the list and leaves the diagnostics on your terminal.

Log records are colored lines when stderr is a terminal. They are JSON, one object per line, when stderr is redirected.

Status lines are colored only on a terminal, and never when NO_COLOR is set.

--json

Under --json, stdout holds one JSON document and nothing else. The banner and the status lines are left out, and warnings go to stderr. A command can therefore be piped as it is:

vaultik snapshot list --json | jq .

--quiet and --cron

--quiet leaves out the banner, the status lines, the tables and the reports. --cron on snapshot create does the same.

Neither flag hides these:

• warnings and errors
• config get values
• JSON documents
• confirmation prompts

Exit status

• 0: the command succeeded.
• 1: the command failed.
• 130: the command was interrupted by Ctrl-C or SIGTERM.

An interrupted command removes its temporary files before it exits. It gives up after 30 seconds.

Temporary files

Temporary files go to $TMPDIR, or to /tmp when TMPDIR is not set.

A backup needs free temporary space for these:

• One blob at a time. Each blob is written whole to a temporary file before it is uploaded, so allow about blob_size_limit.
• About twice that for an rclone remote that cannot stream uploads.
• Copies of the local index, made when the snapshot's database is prepared.

With TMPDIR not set, SQLite writes one of those index copies to /var/tmp.

A restore keeps the downloaded blobs on disk, decrypted, until no remaining file needs them. snapshot restore and snapshot verify --deep also write the decrypted snapshot database there.

These files are readable only by their owner, and are removed when the command ends. A kill -9 or a power loss leaves them behind. Delete any leftover vaultik-* entry in the temporary directory yourself.

What the destination reveals

File contents, names, paths and metadata are encrypted. Someone who can read the destination still learns these:

• when each backup finished
• how many blobs each snapshot uses, and their sizes
• which blobs two snapshots share
• whether a hostname and snapshot name they guess exist there

The last point holds because a remote key is a plain hash of the snapshot ID, made without a secret.

docs/REPOSTRUCTURE.md describes the layout of the destination, what it protects and the risks it accepts.

Limitations

• No extended attributes. ACLs, macOS Finder metadata, quarantine flags and SELinux labels are not backed up.
• No hard links. Two hard links to one file are backed up as two files. The content is stored once, but a restore creates two separate files.
• No sparse files. A sparse file is backed up at its full size, zeros included, and restored as a full file.
• No bandwidth limit. Uploads and downloads use all the bandwidth there is.
• One download at a time. A restore downloads blobs one after another.
• No device nodes, named pipes or sockets. They are skipped without a message.
• No consistent view of changing files. A file that changes while it is read can be stored in a mixed state.
• Owners need root. A restore sets owners and groups only when it runs as root.
• No resume. An interrupted restore starts again from the beginning.
• The local index can become unusable after an update. This applies before version 1.0. Run vaultik database delete and then a full backup. The destination is not affected. See Schema migrations.

Development

These documents are for developers:

• ARCHITECTURE.md describes how the code is organised.
• docs/DATAMODEL.md describes the local index.
• docs/REPOSTRUCTURE.md describes the layout of the destination.
• AGENTS.md holds the rules for AI agents.
• REPO_POLICIES.md holds the repository policies.

Requirements

• Go 1.26.1 or later.
• Docker, with a daemon you can reach. The tests and the linter run only inside Docker.

Workflow

• Make every change on a feature branch.
• Write the tests with the change.
• Update the documentation that the change affects.
• Run make fmt, then make check, before each commit.

make check runs the tests, the linter and the format check. All three must pass.

Entrypoints

The scripts in script/ are the entrypoints. The Makefile targets only call them. The layout follows Scripts to Rule Them All.

• script/bootstrap: installs what development needs. It installs Go from script/install-go when the machine has none, and it fails without Docker.
• script/setup: prepares a fresh clone. It runs script/bootstrap and script/install-precommit.
• script/test: runs the whole test suite, by building the test stage of the Dockerfile.
• script/lint: runs the linter, by building the lint stage of the Dockerfile. The linter never runs on your machine.
• script/lint-fix: applies the linter's automatic fixes to your files. It needs a local Docker daemon. Run script/lint afterwards.
• script/fmt: formats the Go code.
• script/fmt-check: checks the formatting and changes nothing.
• script/check: runs script/test, script/lint and script/fmt-check.
• script/precommit: runs go mod tidy, go fmt and script/check. The git pre-commit hook calls it.
• script/install-precommit: installs that hook.
• script/docker: builds the Docker image. The build runs the linter and the tests too.
• script/cibuild: what CI runs on every push. It runs script/bootstrap, script/check and the image build.
• script/version: prints the version to build into the binary.
• script/projectname: prints the project name, which tags the Docker image.
• script/install-go: installs the Go version that go.mod names into .tool/go. It checks the download against a checksum in the script.
• script/install-goreleaser: installs the pinned goreleaser into .tool/bin, checked the same way.
• script/release: builds and publishes the release files.
• script/release-snapshot: builds the release files into ./dist and publishes nothing.

Every docker build in these scripts uses --no-cache. A cached layer would skip the tests or the linter and still report success.

The test suite has no separate integration target. Every test runs every time. Each test binary is stopped after 90 seconds.

To change the Go version, change these together: go.mod, the version and checksums in script/install-go, and the two golang image digests in the Dockerfile.

Messages to the user

Status lines, warnings and errors go through internal/ui. The data a command exists to print is written plain, with no marker and no color. That covers tables, reports, JSON documents and config get values.

internal/ui has these kinds of message:

• Banner: the startup line. No marker.
• Begin: an operation is starting. Marker 》.
• Complete: an operation has finished. Marker 》, in green.
• Info: a neutral status line. Marker 》.
• Notice: a note the user should see. Marker 》, in cyan.
• Warning: a problem the command survives. Prefix ⚠️ Warning:.
• Error: the command stops. Prefix 🛑 ERROR:.
• Progress and Detail: an indented line under another. Marker 》.

Messages follow these conventions:

• A message is a full sentence that ends with a period.
• Name things in full. Write "backup destination store", not "storage".
• An operation that prints a Complete also prints a Begin.
• Use the tense to show the state: "Uploading" for a Begin, "Uploaded" for a Complete.
• Say what a time belongs to, as in "blob upload elapsed: 30s".
• "ETA" is a clock time, not a remaining time. Print it with ui.Time, and print a remaining time with ui.Duration.

Build a message from the value helpers in internal/ui, and never write color codes by hand:

• Hex: a blob or chunk hash, cut to 12 characters.
• Snapshot: a snapshot ID.
• Path: a filesystem path.
• Size: a byte count.
• Speed: a transfer rate, in bits per second.
• Duration: an elapsed or remaining time.
• Time: a clock time, in local time.
• Count: a whole number, with thousands separators.
• Percent: a percentage.

Releasing

Version numbers

The version a binary reports comes from git. It is the output of git describe --tags --always --dirty:

• On a tagged commit, it is the tag, such as v1.0.0.
• On a commit after a tag, it is <tag>-<N>-g<short commit>.
• With no tag, it is the short commit.
• With uncommitted changes to tracked files, -dirty is added.

script/version gives this value to the Makefile. script/docker and script/cibuild compute it themselves.

A binary built without git reports dev. An image built by script/docker or script/cibuild outside a git checkout reports unknown.

goreleaser builds the release binaries. It uses the tag without its leading v, so the tag v1.0.0 gives vaultik 1.0.0. script/release-snapshot stamps dev- followed by 12 characters of the commit.

vaultik version calls a build a release only when its version is a plain tag, such as v1.0.0 or 1.0.0. Every other version is a development build.

make stops with an error when script/version gives no version.

The release workflow

.gitea/workflows/release.yml runs when a tag that starts with v is pushed. It does this:

• It installs Go with script/install-go and goreleaser with script/install-goreleaser.
• It runs script/release.
• script/release builds archives for Linux and macOS on amd64 and arm64, and a checksums.txt.
• It publishes them to this repository's releases, as a draft.

The workflow needs one repository secret, RELEASE_TOKEN. It is a Gitea access token with the write:repository scope, from an account that can publish releases here.

script/release refuses a goreleaser on PATH that is not the pinned version.

make release-snapshot builds the same files into ./dist and publishes nothing.

Each archive holds the binary, LICENSE and README.md. The release files are not signed, have no SBOM and are not built reproducibly.

license

MIT

author

Made with love and lots of expensive SOTA AI by sneak in Berlin in the summer of 2025.

Released as a free software gift to the world, no strings attached.

Contact: sneak@sneak.berlin

https://keys.openpgp.org/vks/v1/by-fingerprint/5539AD00DE4C42F3AFE11575052443F4DF2A55C2