BackupZit

Open-source NAKIVO alternative: what BackupZit covers and what it does not

If you are looking for a free, self-hosted alternative for backing up virtual machines, servers and Microsoft 365, this guide shows what BackupZit does today — from deployment and VM backup to granular restores, replicas, automation and access control — and where it stops, so you can decide whether it is worth a test in your environment.

What this guide covers

BackupZit is free and open source under the GNU AGPLv3, with no limits on machines, VMs or storage. It is self-hosted: you run the console on your own server, there is no cloud service and nothing has to be exposed to the internet. MaxZit, the company behind BackupZit, offers optional paid support. This guide does not compare prices or features of other products; it describes BackupZit only, so you can check it against what your environment needs.

NAKIVO is a trademark of its owner. BackupZit and MaxZit are not affiliated with, endorsed by or connected to NAKIVO. The name is used here only because many people search for it.

At a glance

AreaIn BackupZitMain limits
DeploymentAppliance (OVA, qcow2, VHDX), appliance ISO, install script, packages or Docker; a small Linux VMThe console runs on Linux only
VM backupProxmox VE, VMware ESXi and vCenter, Hyper-V (also failover clusters), agentless, changed blocks onlyNo XCP-ng; some disk types per hypervisor
Physical machinesWindows and Linux files, Windows disk images, Linux system backups, bare-metal restoreNo macOS agent
Microsoft 365Exchange Online mail, calendars, contacts, OneDrive, SharePoint librariesNo restore into the mailbox yet; no Teams chats
StorageSFTP, S3, SMB, Azure Blob, USB and local disks, hardened repository; deduplicated, compressed, encryptedImmutability only on S3 Object Lock or the hardened repository
RecoveryWhole VMs, in place, single files, instant recovery (Proxmox VE, VMware), Proxmox replicas, P2VNo instant recovery for Hyper-V yet
AutomationREST API with OpenAPI specification, webhooks, Prometheus metrics, email, Slack, Teams—
Access controlFour roles, LDAP/AD, SAML, OpenID Connect, two-factor, four-eyes approval, audit logOne organization per console

Deployment

The console is a Linux server with PostgreSQL, usually a small VM (2 vCPU, 2–4 GB RAM, 20 GB disk). It does not store backup data. Five ways to install it are on the download page:

  • a ready-made VM image for VMware (OVA), Proxmox VE/KVM (qcow2) or Hyper-V (VHDX) — import, start and sign in, without internet access;
  • the appliance ISO for a new VM or a physical server;
  • the install script on an existing Debian 12+, Ubuntu 22.04+, AlmaLinux or Rocky Linux 9+ server;
  • packages (deb, rpm) with your own PostgreSQL, or a Docker container.

Agents run as a service on the protected machines and on the Proxmox and Hyper-V hosts. They connect out to the console on TCP 8443 (no open ports on the clients) and write the backups directly to the storage. Nothing is installed on VMware ESXi: any agent works as the host's proxy. Console and agents update themselves with signed packages.

Virtual machines: Proxmox VE, VMware and Hyper-V

All three hypervisors are backed up without agents in the guests, into the same console, storage and retention rules. After the first backup only changed blocks are read. A job can take all VMs of a host except the ticked ones, for example the VM of the BackupZit console; Hyper-V jobs can also leave out disks by file name pattern or full path.

  • Proxmox VE 7, 8 and 9: the agent on a node backs up VMs and containers from snapshots; running VMs on LVM-thin, ZFS, Ceph RBD or qcow2 files get changed block tracking. Containers on ZFS are backed up file by file from a temporary ZFS snapshot while they run; from the second backup on, only changed files are read. The node's own configuration can be backed up with a files job. Not yet: container volumes that are directories on non-ZFS storage, and replicas of containers on ZFS; raw disks on directory, NFS or CIFS storage and thick LVM only while the guest is stopped. See the Proxmox VE guide.
  • VMware ESXi 6.7, 7 and 8 and vCenter Server 7 and 8: a proxy agent reads the VMs with changed block tracking; hosts with the free ESXi license work too (snapshots and restores over SSH). Optional hot-add (a Linux proxy agent in a VM, with vCenter or a licensed host) reads running VMs without a temporary copy on the datastore. Not yet: a vCenter with several datacenters (add the other hosts directly), raw device mappings and independent disks, VMs with snapshots of their own. See the VMware ESXi guide.
  • Hyper-V on Windows Server 2016 or newer and Windows 10/11: the Windows agent on the host uses Resilient Change Tracking. With the agent on every node, one job covers all VMs of a failover cluster, and a VM keeps its history when it moves, also during a backup. Generation 2 VMs keep their vTPM (BitLocker) when restored on the same host or a host with the original host's guardian certificate. Not yet: shared VHDX, VHD Set and pass-through disks; shielded VMs are not supported. See the Hyper-V guide.

XCP-ng is not supported.

Application consistency

  • Proxmox VE: when the QEMU guest agent runs in the VM and is enabled in its options, Proxmox freezes the file systems for the snapshot.
  • VMware: when VMware Tools run in the guest, the snapshot is quiesced, so applications and file systems are flushed first.
  • Hyper-V: Windows guests flush their applications through VSS for the recovery checkpoint. With the job option Require application-consistent checkpoints (BackupZit 0.48, agents 0.48.0 or newer) a VM whose guest cannot prepare a VSS checkpoint is not backed up and the run fails with a notification, instead of falling back to a crash-consistent checkpoint — meant for Active Directory, SQL Server and Exchange.

For databases you can also add a Databases job on the database machine (SQL Server, PostgreSQL, MySQL/MariaDB) with log backups and point-in-time restore; see the database guide.

Physical servers and PCs

Under Backup jobs an agent can back up Files and folders (Windows and Linux, from a VSS snapshot on Windows), a Disk image of one, several or all disks of a Windows machine, a Linux system (all local file systems and the disk layout) or NAS shares over SMB or NFS. Bare-metal restore boots a Windows or Linux recovery ISO and writes the backup onto an empty disk, also onto different hardware or into a VM. Supported: Windows 10, 11 and Server 2016 or newer, Windows 7 SP1, Server 2008 R2 SP1 and 2012 R2 through the legacy agent, and 64-bit Debian, Ubuntu, AlmaLinux and Rocky Linux.

Not included: an agent for macOS; Linux system jobs do not support LUKS, software RAID (mdadm) or btrfs, and a whole-system restore recreates one disk. Details in the guide for servers and PCs and the Windows bare-metal restore guide.

Microsoft 365

A Microsoft 365 job backs up Exchange Online mail (.eml), calendars (.ics), contacts (.vcf), OneDrive and SharePoint document libraries through Microsoft Graph with read-only application permissions, run by any agent. You restore by browsing and downloading, or into a folder on an agent.

Not included: restoring directly into a mailbox or OneDrive (planned), Teams chats and channel messages (Teams files are in SharePoint and are backed up with it), SharePoint lists and pages. See the Microsoft 365 guide.

Storage and immutability

Storage targets: SFTP, S3-compatible storage (AWS, Wasabi, Backblaze B2, MinIO, Google Cloud Storage), SMB shares, Azure Blob Storage, rotating USB disks, local disks and the BackupZit hardened repository. Every backup is deduplicated — also across the jobs of one agent on the same target, not between computers — compressed with zstd and, with Encrypt backups (AES-256) on (the default), encrypted on the agent. With the Recovery key backups can be restored even without the console.

  • Immutable: S3 with Object Lock in compliance mode, or the hardened repository (backupzit-repo) on your own Linux server.
  • Second copy: a Backup copy job copies another job's backups to a second target with its own retention; only data that is not there yet is sent.
  • Ransomware: backups that look like ransomware or mass deletion trigger an alert and pause the job's retention.

Limits: SMB, SFTP and local disks are not immutable by themselves; copies of disk images are restored with the command-line agent. The 3-2-1 guide shows a complete setup.

Recovery options

  • Whole VMs as a new VM or in place, on the same or another host of the same hypervisor.
  • Single files from VM backups: Browse files shows the partitions and LVM volumes of the guest's disks and opens NTFS, ext4 and XFS file systems; download files or restore them on any agent. Browsing needs the backup on network storage (SFTP, S3, SMB, Azure or a hardened repository).
  • Single files from disk images: Browse files on an image run opens its NTFS volumes; Download gives one file as it is or several as a ZIP, or restore them on any agent.
  • Instant recovery with Instantly, running from the backup: Proxmox VE VMs start from the backup and are then moved to normal storage; VMware VMs run from an NFS datastore served by a Linux proxy agent, and with vCenter Finish moves them with Storage vMotion while they run.
  • Replicas (Proxmox VE): stopped, up-to-date copies of guests on another node, started with Start replica.
  • Physical to virtual: a Windows disk image restored into a Proxmox or Hyper-V VM with Restore to different hardware, or a Linux system backup restored as a new Proxmox VM.
  • Restore tests: automatic sample restores (weekly by default), a check of all stored data (monthly by default) and boot tests of Proxmox VMs with the network disconnected.

Limits: no instant recovery for Hyper-V yet or through a Windows proxy agent; a VM running from the backup is only as fast as the backup storage. Boot tests exist only for Proxmox VMs.

Automation and monitoring

  • A REST API with tokens and an OpenAPI specification at /api/v1/openapi.json: status, agents, jobs, runs, reports, storage, and starting a file restore.
  • Webhooks for finished runs carry the result, duration and size, so a dashboard needs no extra call; Slack and Microsoft Teams channels and email notifications for failures, warnings and successes.
  • Prometheus metrics at /api/v1/metrics: agents online, last status and last success per job, used and free storage.
  • Schedules, job chains (run B after A), backup windows, a speed limit, a Recovery point objective (RPO), hours per job, Reports as PDF and CSV (also emailed on a schedule) and a Calendar of runs.

Access control

  • Four roles: Administrator, Backup operator, Restore operator and Viewer.
  • Local users, LDAP / Active Directory, single sign-on with SAML 2.0 or OpenID Connect, and two-factor sign-in that can be required.
  • Single sign-on users get their role from their groups at every sign-in; with OpenID Connect the console asks for the groups scope when the provider offers it (Authelia, Keycloak, authentik), and Additional scopes covers other providers. A last administrator who lost the second factor gets back in with backupzit-server --reset-two-factor USER on the server.
  • Optional four-eyes approval: deleting jobs or storage, shortening retention and similar changes wait for a second user.
  • A tamper-evident Audit log, and Console backup for the console itself.

Each organization runs its own console; there is no multi-tenant mode for several customers in one console.

How to test it next to your current backup

  1. Leave your current backup as it is. Import the appliance as a small VM and give BackupZit its own storage target, separate from the existing backups.
  2. Pick a few VMs per hypervisor and, if you have them, a physical Windows server and a Linux machine. Schedule the jobs at other times than the existing jobs of the same VMs.
  3. Watch the overlaps. On VMware, VMs with snapshots of their own are not backed up, so check that no other tool leaves a snapshot behind. SQL Server full backups without log backups are taken COPY_ONLY; turn on log backups only where no other tool backs up the logs.
  4. Restore, not just back up: a VM as a new VM next to the original, single files from a VM, an instant recovery, a disk image into an empty VM from the recovery ISO. Turn on the automatic restore tests.
  5. Compare over a few weeks: each run page shows what was read, what was new and the compression ratio; Reports and the storage usage view show success rates and growth.
  6. Go through the limits below and the Not yet included parts of the user guide, and ask in GitHub Discussions if something is unclear.

BackupZit is a young project: the first public release was on 4 October 2026, and new versions come out often. Release candidates go to a separate testing update channel first.

Limits to know

  • No XCP-ng support and no macOS agent.
  • Not yet: instant recovery for Hyper-V, restore directly into Microsoft 365 mailboxes, Teams chats.
  • Disk types not backed up: Proxmox container volumes that are directories on non-ZFS storage; VMware raw device mappings and independent disks; Hyper-V shared VHDX, VHD Set and pass-through disks. Replicas of Proxmox containers on ZFS are not supported. A vCenter with several datacenters cannot be added as one; shielded Hyper-V VMs are not supported.
  • Databases only on the agent's own machine; no SQL Server differential backups or Always On availability groups.
  • The console runs on Linux (Debian 12+, Ubuntu 22.04+, AlmaLinux or Rocky Linux 9+, x86-64), the appliance or Docker; one organization per console.
  • The console and its guide come in 14 languages; paid support is provided in English only.

Next steps

Download BackupZit, see all features and the supported systems, or read the user guide in the console. A broader overview is the open-source Veeam alternative guide; the step-by-step guides cover Proxmox VE, VMware ESXi, Hyper-V, Microsoft 365, Windows bare-metal restore and 3-2-1 backup with immutable copies. Questions are welcome in GitHub Discussions; companies can get commercial support from MaxZit (in English).