Skip to content

Configuration

This page covers the container's environment variables, the mounts the template provides, the security model, VM backup over SSH, and the off-site setup. Where backups go is set up inside the app as storage locations, not via environment variables.

Environment variables

Variable Required Description
APP_KEY Yes 32-byte hex secret (64 hex chars) used to derive the restic repo password. Generate with openssl rand -hex 32. Keep this safe: losing it makes encrypted backups unrecoverable.
LIBVIRT_HOST For VMs and ZFS datasets Unraid host reached over SSH for VM and ZFS dataset backups (template field Host SSH: Address, default host.docker.internal). The template pre-fills the placeholder 192.168.x.x, which counts as unset. Use your Unraid LAN IP, required on a custom br0.x network. When host.docker.internal does not resolve, as in a TrueNAS app, and LIBVIRT_URI names no host either, BombVault uses the container's default gateway and the boot log names it.
LIBVIRT_SSH_PORT No Host SSH port for VM and ZFS dataset backups (template field Host SSH: Port, default 22).
LIBVIRT_SSH_USER No SSH user on the host for VM and ZFS dataset backups (template field Host SSH: User, default root).
LIBVIRT_URI No Full libvirt connection URI, used verbatim instead of building one from the three LIBVIRT_* variables above (which are then ignored for the connection string). Default unset. Needed on TrueNAS Scale, whose libvirtd listens on a non-standard socket the built-string form cannot express: qemu+ssh://<user>@<truenas-host>/system?socket=/run/truenas_libvirt/libvirt-sock. See the VM backup guide. When it is a qemu+ssh:// URI, each of LIBVIRT_HOST, LIBVIRT_SSH_USER and LIBVIRT_SSH_PORT that is not set is taken from it, also for BombVault's own SSH commands (NVRAM transfer, ZFS datasets).
PORT No HTTP port (default 3000. Only used with HTTP_ONLY=true).
HTTPS_PORT No HTTPS port (default 3443. The template publishes it 1:1, so the WebUI answers on https://<ip>:3443).
HTTP_ONLY No Set true to disable the self-signed HTTPS listener and serve plain HTTP only (for use behind a TLS-terminating reverse proxy).
BIND_HOST No Address the WebUI listens on (default 0.0.0.0, every interface). Leave it unset in the container, whose published ports need every interface. 127.0.0.1 suits a run outside Docker. The healthcheck asks the same address.
TRUSTED_PROXY No Comma-separated addresses or CIDR ranges of the reverse proxy in front of BombVault (for example 192.168.20.11 or 10.0.0.0/8). Only from those hops is X-Forwarded-For believed, and the login throttle then counts failures per real client instead of putting everyone behind the proxy in one bucket. Unset (the default) trusts nobody: a header believed unconditionally would let any caller choose their own bucket.
HOST_SOURCE_ROOT No The host path mounted as Host Data (default /mnt). BombVault translates the bind-mount sources Docker reports into paths under this mount. Change only if you mounted a different host root.
HOST_MOUNT_ROOT No Where Host Data is mounted inside the container (default /host/user). On TrueNAS Host Data is mounted at its own host path, so this and HOST_SOURCE_ROOT both name that path, for example /mnt/tank (see TrueNAS SCALE).
DATA_ROOT_SEGMENTS No Comma-separated path-segment names that mark a bind-mount source as backup data (default appdata, matching Unraid's /mnt/user/appdata/<container> convention). A container's bind mount is auto-selected for backup when ANY listed segment appears as a full path segment of its host source. For example DATA_ROOT_SEGMENTS=appdata,config also picks up a .../config bind. The match ignores case, so the default also covers ZimaOS's /DATA/AppData/<container>. See Backup source detection for the other, always-on ways a container's data folder is found.
PLATFORM No Forces which platform BombVault treats itself as running on, instead of auto-detecting: unraid, generic, or truenas (default unset: auto-detects Unraid by probing for its dockerMan marker under the flash mount, otherwise generic. An unrecognized value also falls back to generic, logged). Set it explicitly on a generic Docker host or TrueNAS Scale rather than relying on the Unraid-only auto-probe. The generic compose file does this. Changes the appdata-fallback convention, the cross-instance restore-destination defaults, and whether the Unraid-only notification/companion-plugin steps are attempted at all (see internal/platform).
BOMBVAULT_SELF_CONTAINER No The name of the BombVault container itself, so it never backs up (and thus stops) itself.
BACKUP_MAX_HOURS No Maximum wall-clock hours a single backup run may hold its domain lock before it is force-cancelled (a guard so a wedged run cannot block the domain forever). Empty (the default) uses 48. Raise it for very large or slow cloud backups (a run cancelled at the cap fails with context deadline exceeded). Set 0 to disable the cap entirely. A prune is held to the same cap, and a check or prune runs as long as restic keeps reporting progress.
BACKUP_STALL_HOURS No Hours a backup may make no progress at all before it is cancelled. Empty (the default) uses 2. Set 0 to never cancel on a stall. This is the finer of the two guards and usually the one that fires: it watches whether anything is still happening rather than how long the run has taken, so a slow but healthy multi-terabyte backup is left alone while one wedged on an unresponsive share is stopped in hours instead of days. A warning is logged after 30 minutes of silence, before anything is cancelled. Scanning counts as progress: restic writes no bytes while it walks a large tree, and that phase is watched through its file and byte totals rather than through the bytes written. The two variables are independent, and BACKUP_MAX_HOURS still bounds the phases after the backup itself (retention, statistics, off-site copy), where there are no counters to watch.
DB_DUMP_MAX_HOURS No Hours one automatic database dump may run before it is stopped. Empty (the default) uses 6. Allowed values are 1 to 48, and the limit is kept an hour below BACKUP_MAX_HOURS (at half of it when that is under two hours), so a long dump is cut by its own limit and reported as such rather than taking the backup down with it. A dump that makes no progress is stopped earlier, after BACKUP_STALL_HOURS. A stopped dump fails on its own and the container backup carries on. On Unraid, add it to the BombVault container with Add another Path, Port, Variable.
TZ No Timezone for the scheduler (for example Europe/Berlin). Leave it unset and every schedule runs in UTC: one set to 02:30 then fires at 02:30 UTC rather than on your own wall clock. On Unraid you never set this yourself: the system passes its own timezone into every container. The boot log says which zone it resolved to. A zone with daylight saving skips one run in spring and runs one twice in autumn, while UTC does neither but drifts an hour against your clock twice a year.
DATA_DIR No Where BombVault keeps its database, keys and caches (default /config). Leave it alone in the container. It matters when you run the binary outside Docker.
FLASH_DIR, FLASH_TEMPLATES_DIR No Where the flash drive and its Docker templates are mounted inside the container (defaults /host/boot and /host/boot/config/plugins/dockerMan/templates-user). They follow the template's /boot mount, so leave them unset.
MODE No full (the default) runs the whole app. node runs a node: no web interface and no schedules of its own, only what its home instance tells it. Any other value stops the start with a message. Home fills it in for you in the node's compose file.
NODE_HOME On a node Which home instance the node belongs to, with home's key and relay, as one value that starts with bvhome1.. Home writes it into the node's compose file. Without it the node starts, but no home reaches it. A node that home took on refuses to start when NODE_HOME names another home, until that home releases it.
NODE_NAME No The name of the node, as home filled it in from the name you gave it.

Mounts

Mount the Docker socket, the flash (/boot) and the Host Data root (/mnt) as shown in the CA template. Backup sources and destinations both live under Host Data, and it is mounted slave so a remote share that mounts after the container starts (for example under /mnt/remotes) becomes visible without a restart.

On TrueNAS, Host Data is a dataset path such as /mnt/tank, never an ixVolume. Data an app keeps in an ixVolume lies outside it and is not backed up. See TrueNAS SCALE.

ZFS dataset backups need this mode too: the host mounts a dataset's snapshot only after the container has started. See ZFS datasets.

Local repositories default to /mnt/user/bombvault/{container,vms,flash,config,files,zfs}, created on the first backup. They are the sections' own folders and form the Local storage location. On its page under Storage locations, the Connection card changes the folder of each section.

A section's own folder can also live in a storage location you already have, such as a NAS or an S3 bucket: Use as own folder on that location's page moves it there and says first what happens to the backups already made. See A section's own folder.

The local repository folders stay at modes 0755 and 0644, so a non-root sync user over SMB is not locked out, and the repositories are encrypted, so this exposes nothing. Recovery definitions live inside each repository, so a copied folder is complete on its own. The self-backup repository stays readable by root only.

Host integration check

The Host Integration card on the overview (add it under Customize) probes every mount and CLI (Docker socket, libvirt, restic, qemu-img, rclone) and reports any missing pieces.

Backup source detection

For each container, BombVault auto-selects which bind mounts and named volumes to back up. A path is picked up when any of the following applies (you can always override the result in the Backup folders card on the container's Settings tab):

  • Data-root segment match: the bind's host source contains one of the DATA_ROOT_SEGMENTS segments as a full path component (default appdata only).
  • Docker named volumes are always included, since they have no throwaway equivalent and there is nothing to filter, but only when the volume's real host storage path is itself reachable through the Host Data mount, exactly like any other host path BombVault backs up. Docker's default local-volume driver stores a volume under the daemon's own data root (/var/lib/docker/volumes/<name>/_data unless you've customized it. Check with docker info -f '{{.DockerRootDir}}'), which is not covered by the narrow, single-directory Host Data mount the generic docker-compose.yml uses by default. An unreachable volume is silently skipped, not an error. To actually back up named volumes on a generic host, point Host Data (and HOST_SOURCE_ROOT) at a common ancestor that also covers the Docker data root. The compose file's Host Data comment explains the tradeoff (Unraid sidesteps this by mounting all of /mnt, its own universal top-level convention, for the same reason).
  • Docker Compose project directory: if the container carries the standard com.docker.compose.project.working_dir label (set automatically by docker compose up), that directory is added too, regardless of whether any bind matched a data-root segment.
  • bombvault.data label override: set the label bombvault.data=true on a container to include all of its bind mounts, for a layout neither convention above catches (for example a single /srv/plex/config bind with no Compose project). Any non-empty value other than false counts as truthy. An absent label or bombvault.data=false changes nothing.
  • bombvault.dbdump label: set bombvault.dbdump=false on a container to switch its automatic database dump off (0, no and off do the same), or name the engine (postgres, mysql, mariadb) to dump a container BombVault does not recognise by itself. The label wins over the switch in the Database card on the container's Settings tab, which is the way to do it on Unraid.

Security model

Root-equivalent control of the host

Through the Docker socket BombVault can stop, remove and recreate containers and read/write appdata, and for VM backup it logs in to the host over SSH (qemu+ssh://, root by default) to run virsh. Anyone who can reach its web UI effectively has root on the host.

Login

  • Optional password protection (Password card under Settings, Security): set a password to require login, clear it to disable. Off by default for trusted-LAN use. The password is stored with Argon2id over an APP_KEY-keyed pepper, so a copied /config is worthless without the key and slow to attack with it. A new password needs at least 12 characters. An existing shorter one keeps working until it is changed. Sessions are signed (HMAC derived from APP_KEY) and changing the password invalidates them. Logins are limited to five failures a minute per client. Changing or clearing the password asks for the Current password and, with two-factor authentication on, the code.
  • Some things need a password. Downloading the recovery kit, the bug report and the ZIP of a flash backup, exporting a flash or self-backup, depositing the key with a key keeper, allowing a deposit to be handed out, making a new recovery code, exporting settings with credentials, two-factor authentication, passkeys, adding a node, showing a node's compose file again and opening a node's entries are refused until a login password is set, and they need a sign-in at the instance itself. Deleting backups works without a password. With one set, it needs that sign-in too.
  • A location shows no password. A location can hold a sign-in, as in rest:https://user:password@host/repo, and every page and API answer shows it as rest:https://[redacted]@host/repo, with or without a login password. Change the path behind it and the stored password stays. To point the location at another server, type the whole location with its password, because a hidden password stays with its server.
  • Two-factor authentication (Settings, Security): a time code from an authenticator app on top of the password, plus eight single-use recovery codes handed out once when it is switched on. The shared secret is stored encrypted with APP_KEY, and turning the factor off again requires a current code.
  • Passkeys (WebAuthn) sit on their own card once a password is set, alongside the password and never instead of it, so removing every passkey locks nobody out. They need a real domain name and a certificate the browser trusts. The default https://<ip>:3443 is exactly what WebAuthn refuses, and the card says so instead of offering a button that fails. With two-factor authentication on, a passkey has to check your PIN, fingerprint or face, or the sign-in asks for the code as well.
  • A configuration restore (Bring back settings in the Emergency wizard) ends every session, removes every passkey and revokes every API token and MCP key, because the restored settings could bring back ones that were removed after they were saved. Sign in with the password and set your passkeys up again under Settings, Security.
  • A forgotten password is reset on the login page. Forgot password? has BombVault write a code of 12 characters, in three groups of four, to its container log (on Unraid: Docker tab, BombVault icon, Logs). That code and a new password sign you in. The reset also switches two-factor authentication off, removes every passkey, revokes every API token, MCP key and OAuth sign-in, ends every other session, signs every paired phone out and switches phone pairing off. The page afterwards lists what went. A code is valid for 10 minutes, and wrong guesses do not use it up. Widget and metrics tokens stay, since they only read.
  • Whoever can read the container log can reset the password. That is root on the host, but also a log viewer such as Dozzle, any container given the Docker socket (read-only is enough) and a log shipper that forwards container output. Give those only to people you would trust with the password. Somebody who only reaches the web interface never sees the code.

Phones and group members

  • Paired phones reach an allow-list. A phone that knows the group's twelve words signs in without the password, as long as Phones can pair with the group's twelve words is on in the Android app card under Settings, Security. Its session reaches only what the app shows: the overview, the runs, the backups to look at, start and restore, the anomalies to read and the Instances page to look at. Storage locations and commands reach it with their addresses and secrets removed, and a restore that replaces the original backs up the current state first. A restore into a folder takes only a new or empty folder inside the restore folder, and a phone can look into no other folder. Everything else answers that it needs a sign-in at the instance itself: keys, the recovery kit, key deposits, settings, deleting backups. A password reset switches phone pairing off until you switch it on again. A phone removed from the Instances page, or with Poll this instance switched off in its details, loses its session at once, and a removed phone that opens its app again comes back switched off.
  • Away from home a phone opens pages through the relay. Each request of the page is sealed with the group's key, which the relay cannot read or change, and signed with a key of the phone's own, which the server pins on the phone's first call and lets in once somebody approves the phone on the pairing page. The server runs a request once and only while it is recent. It answers only a phone listed on the Instances page while phone pairing is on, and only with a session it gave that phone, so signing out everywhere and a password reset end these pages as well. The server signs every answer, and the app checks the signature against the key it pinned for that server and warns when the key changes. Such a page reaches the same allow-list as a paired phone, with or without a login password, plus the sign-in status the page reads first. Signing in and out, the password reset and downloads stay at the server's own address, and one answer is capped at 16 MB.
  • Members sign what they ask for. Role requests and their answers, storage offers, the handing out of a restic password to a receiver or fetcher, receiver logins and key deposits are signed with a key of each member, which an instance pins the first time it sees that member. A call signed with another key is refused. Forget key in an instance's details drops the pinned key, for the case where the one known here is not the member's own. The member is asked for its key again at its next signed call. Members on BombVault 9.x do not sign and can be impersonated by anyone with the words, as in 9.9. A role request or a storage offer from a member on 10.4 or older is refused until that member is updated.
  • The twelve words are a password. Anyone who has them joins the group, sees the members' protection status and can ask for roles, which a person on the other instance still has to allow. Without a login password anyone who can open the web interface can read them, and the pairing card says so until a password is set.

Keys

  • APP_KEY does two things. It seals what this install stores (passwords, sign-ins, tokens) and it is the key of the backups: the restic password is derived from it, and container and VM definitions are sealed with it.
  • A taken-over key is kept apart. When the Emergency wizard takes over another server's backups, this install keeps sealing with its own APP_KEY and stores the other server's key as its backup key, sealed like every other secret. The recovery kit then prints both. If APP_KEY is changed afterwards, the backup key can no longer be opened: BombVault stops backing up and says so under Needs you, instead of writing with the wrong key. Upload the recovery kit again under Restore, Emergency, or set APP_KEY back. See Emergency wizard.
  • Encryption is on by default (Encrypt backups in the Encryption card under Settings, Security). Backups are encrypted by restic with the password derived from the backup key.
  • Key deposits. A key keeper in your group holds a sealed copy of this server's recovery kit that it cannot open itself. Only the 20-character recovery code opens it, and the keeper hands it out only after someone there presses Allow. See Key keeper.

Nodes

A node is a BombVault without a web interface that takes its orders from a home instance.

  • Home holds the node. A node mounts the Docker socket of its server, which is root there, and home controls the node. If home is taken over, the attacker controls all its nodes as well. Give home a login password and keep its web interface off the internet.
  • A node holds little. If a node is taken over, the attacker gets no pairing group, no fortress, no other node and no other repository. The node has a link to its home alone, a login of its own on the receiving server and a folder of its own at a storage location. On the receiving server it cannot delete history, because the server is append-only. It can fill the server's disk, because rest-server sets no quota, so give the receiving server room or a share of its own. At a storage location its sign-in can usually delete what lies there. Backups kept on the node are in the attacker's hands, which is why home fetches a copy of them.
  • The link is private and signed. Home and the node talk over the relay through a link of their own, sealed with a secret derived from the node's APP_KEY. Each side knows the other's key before the first call, so there is no approval step. Home signs every call with its own key and a counter, and the node refuses a call that is not signed by home or that it took before. The node calls nothing at home but its hello.
  • Home reaches a fixed list. Requests home passes to a node run only on a fixed list of the node's routes, listed at GET /api/nodes/routes, and the node answers them with JSON only. Nothing on the list deletes, or reaches the settings, the storage, the keys, the group or the tokens. On home they need a sign-in with the password: paired phones, API tokens and MCP keys do not reach a node. Backups and restores started this way show as started by home.
  • Key deposits carry node keys. Home keeps each node's APP_KEY, and the recovery kit prints it in its chapter Nodes, so a key deposit carries it too. A key keeper that gets the deposit and the recovery code can open the repositories of your nodes, but it cannot command a node, because that takes home's signature.
  • Releasing is for good. A released node takes no home again with the same APP_KEY. A node that is away when you release it hears it from home once it is back.

Network

  • Changes need JSON. A request that changes something must send Content-Type: application/json and must not be marked cross-site by the browser, so a page on another site cannot make your browser change settings on a LAN address. A script that drives the API sends that header. Anything else is refused with 415.
  • Because the gate is opt-in, when unset the whole UI and API (including the storage locations, tamper-test routes and the restore routes) are reachable by anyone who can reach the port. Set a login password once off-site, append-only backups or encryption are in use.
  • Run BombVault only on a trusted, non-exposed network. For remote access put it behind a reverse proxy that adds authentication and TLS. Responses carry baseline security headers (CSP, nosniff, X-Frame-Options, Referrer-Policy).
  • Behind a reverse proxy every request carries the proxy's address, so without TRUSTED_PROXY the login throttle counts every client in one bucket and an attacker's failures lock you out as well. Name the proxy in TRUSTED_PROXY to get per-client counting back.
  • A reverse proxy in front of BombVault has to pass the Authorization or X-API-Key header through to /mcp and must not buffer its responses, or assistants cannot connect. See MCP server.
  • The MCP endpoint /mcp answers 404 until a key exists or sign-in through OAuth is switched on, and it asks every client for its key or token even with the login password off. No address is exempt, not even localhost. It has no restore or delete tools, and restoring a configuration backup revokes every key. See MCP server.
  • With HTTP_ONLY=true the session cookie loses its Secure flag (it has to, to work over plain HTTP), so only enable the password behind a TLS-terminating proxy if confidentiality matters.
  • The VM-backup SSH connection trusts the host key on first connect (TOFU) and pins it thereafter. Verify the host's key out-of-band if your container-to-host path is not trusted.
  • Links do not lead out of the mount. A backup or a restore in place stops with a message when a link above its path leads out of the host mount, because a container that writes into its own share could plant one. A storage location whose host or rclone remote starts with a dash is refused the same way, on save and at the next run, because ssh or rclone would read it as an option.

MCP server

The MCP server needs no environment variable. You switch it on by creating a key under Settings, Integrations, MCP server, and it answers at /mcp on the same port as the web interface (for example https://192.168.1.10:3443/mcp). Without an active key that path answers 404. Clients, certificates and limits are described on MCP server.

VM backup over SSH

BombVault backs up KVM/libvirt VMs without mounting any libvirt path. It runs virsh on the host over SSH (qemu+ssh://), so it can never affect your host VM Manager.

Mounting the host's libvirt socket into a container is fragile on Unraid: the VM Manager owns those paths, and switching "Enable VMs" can leave libvirt unable to start. The SSH key grants root on the host, the same level of trust as the Docker socket BombVault already uses.

Quick setup:

  1. Settings, Integrations, Host SSH: copy the shown public key.
  2. Append it to Unraid's /root/.ssh/authorized_keys (also persisted to the flash so it survives reboots).
  3. Click Test connection.

The template adds --add-host=host.docker.internal:host-gateway so the container can reach the host. Set LIBVIRT_HOST to your Unraid LAN IP if that name does not resolve (for example when the container runs on a custom br0.x network). If you changed Unraid's SSH port, set LIBVIRT_SSH_PORT to match. Live snapshots also need the qemu guest agent in the VM and the disk on /mnt/cache (not /mnt/user).

Full VM setup and networking guide

The complete step-by-step guide (SSH enablement, persistent key authorization, custom-network and VLAN routing, per-VM method and host-side troubleshooting) is the VM backup guide.

Off-site setup

Off-site copies go to storage locations. Open Storage locations from its settings tile (or pin it to the sidebar) and choose Add storage location. See Off-site & recovery for the full workflow (delete protection, tamper testing and DR drills). In short:

  • Backends: SMB/CIFS and WebDAV through rclone without a host mount, NFS (mount the export on the host and use it as a local folder), native restic backends without rclone (s3:..., rest:http://host:8000/repo, sftp:user@host:/repo), or any rclone remote (rclone:<remote>:<bucket>/path). Backblaze B2 has no native backend here: reach it through its S3 endpoint (s3:https://s3.<region>.backblazeb2.com/<bucket>/<path>) with the key ID and application key as the S3 credentials.
  • Sign-ins are stored encrypted, as one list of saved sign-ins in the Sign-ins card under Settings, Storage locations. A star marks the default, which is used wherever no sign-in is picked. A location's own page picks the one it uses in its Sign-in card. See Sign-ins.
  • SSH targets need nothing installed on the far side. sftp: only needs an SSH server. Add the public key from Settings, Integrations, Host SSH (also at /config/ssh/id_ed25519.pub) to the target user's ~/.ssh/authorized_keys, or let Add key in Add storage location put it there with the account's password.
  • rest-server with a self-signed certificate: put its certificate, or the one of the authority that signed it, as a PEM file into the certs folder of BombVault's data folder (/config/certs in the container, appdata/bombvault/certs on Unraid). Backups, copies and the delete protection test trust it without a restart.
  • Off-site copy: BombVault copies new snapshots with restic copy on a best-effort basis, on top of a (usually local) primary. A location's Copy card sets whether it copies After every backup or on a plan of its own, and Settings, Schedule, Off-site copies per section gives one section its own copy plan. Replicate now sits in the section's row of the Off-site copies card under Settings, Storage locations.
  • Several storage locations per section: the Storage locations per section card on the Storage locations page switches each line from a section to a location on or off with one click, Local included. A section with only off-site locations backs up straight to them, and one with every line off is not backed up. A location's Rules per section give one section its own repository, retention or compression there.
  • Placement per entry: each container, VM and folder set lights Local and the locations that get its backups, in the Schedule and storage location card on its Settings tab. Entries without a choice of their own follow their section. See Placement per entry.
  • Retention: each storage location keeps backups by its How long to keep card, and a section can keep its own values there under Rules per section. That includes the copy in a section's off-site field. All zero never trims a location's snapshots.
  • Bandwidth limits: cap the restic upload/download rate in the Off-site bandwidth card under Settings, Storage locations, or per location in its Copy card. You enter KiB/s, and the field shows the same rate in MB/s and Mbit/s.
  • Streaming first: in the Streaming card on the Storage locations page, pick the media servers (Plex, Jellyfin and Emby are preselected by image name), the send rate from which one counts as streaming, the upload limit while it streams and how long after a stream the normal limit comes back.
  • Cold and archival storage class (S3): for an S3 storage location, pick a restore-readable tier (Standard, Standard-IA, One Zone-IA, Intelligent-Tiering, Glacier Instant Retrieval) on its page. rclone remotes set their class in the rclone config.
  • Own folder in a storage location instead of local: a section's own folder can live in one of the storage locations above, with no local copy and no copy step. See A section's own folder.

Anomalies

Anomaly detection needs no environment variable. It is set up in the Anomalies card under Settings, Verify, described on Anomalies.

Portable settings (export and import)

The Export / import settings card under Settings, System writes your whole BombVault configuration (section settings, storage locations, schedules, retention, notifications) to a portable JSON file you can import on another instance, so moving to a new box or cloning a setup does not mean re-entering everything by hand. Import shows a preview and asks for confirmation, and it never touches your backup data or history. Links in the file to storage locations or repositories that no longer exist are left out, and the preview names them. While a ZFS replica runs, an import is refused before anything is written.

The export can contain credentials

You choose whether to include the off-site, notification and MQTT broker credentials in the file. With credentials included, the export is as sensitive as your recovery kit, so store it somewhere safe. Without them, the file holds only non-secret settings.

The APP_KEY does not travel with the file

The repositories are locked with a password derived from the APP_KEY, and the file does not hold it. For a planned move, give the new box the old APP_KEY in its template, or take the key over with the recovery kit under Emergency. Under another key the repositories answer with a wrong password. Nothing is lost, and setting the right key opens them again.