Self-hostable · your storage, your keys

Backups you can actually restore — proven, not promised.

One control plane for servers, databases, hypervisors, cloud VMs and SaaS. Every copy is deduplicated, compressed and encrypted before it leaves your network — and every restore is verified chunk by chunk against its content address, so a corrupted or tampered backup fails loudly instead of quietly handing back bad data.

Start free See all 43 platforms

No credit card. Free tier is permanent. Self-host the whole stack if you prefer.

Data centre server racks with structured cabling

AES-256-GCM client-side S3 Object Lock BLAKE3 verified

43platforms in the catalog

9hypervisor & cloud VM platforms

12storage backends supported

100%of your data, on your own storage

The real problem

Most backup failures are discovered during the restore.

A green dashboard is not evidence. Backups silently truncate, credentials rotate, retention quietly deletes the copy you needed, and ransomware encrypts the repository along with production. RemoteBackups is built so each of those fails visibly, early, and in a way you can act on.

Ransomware reaches backups first

Attackers delete or encrypt the repository before touching production, so recovery is impossible when you need it.

Answer: write-time WORM + detection over the backups themselves

Silent corruption

A truncated archive still looks like a valid archive. Nothing complains until the day you try to read it back.

Answer: every chunk re-hashed on restore; size mismatch is an error

Vendor lock-in on your own data

Proprietary formats and per-socket licensing make leaving expensive, and make an air-gapped copy someone else's decision.

Answer: your buckets, your keys, self-hostable end to end

Capabilities

One plane for everything you have to protect.

Agentless where it can be, agent-based where it must be — with the same repository format, retention model and restore path behind both.

Deduplication that pays for itself

Content-defined chunking with BLAKE3 content addressing means an unchanged block is referenced, never re-uploaded — across every backup and every asset in the repository.

FastCDC · BLAKE3 · zstd

Encrypted before it leaves

AES-256-GCM with per-chunk authentication, keyed from your organisation's master key. The storage provider holds ciphertext and cannot see your file names or content hashes.

Client-side · you hold the key

Immutable by physics, not policy

Copies are written under S3 Object Lock, so the storage backend itself refuses deletion until retention expires. Stolen credentials cannot erase what the bucket will not delete.

Governance & compliance modes

Granular restore with undo

Browse a recovery point, restore a single file or a single database row, preview the diff before applying it — and roll the change back if it was wrong. Four-eyes approval on the sources that warrant it.

File · row · full asset

Point-in-time recovery

Replay the transaction log forward from a recovery point to a chosen moment — so the answer to "restore to just before 14:32" is a real answer, not the previous nightly.

Log replay for supported engines

DR orchestration

Recovery plans are a boot-order DAG, not a wiki page. Run them non-disruptively as a rehearsal, or for real as a failover, and get measured RTO/RPO against your targets.

Test mode · failover · failback

Coverage

43 platforms, one backup plane.

Every connector ships with a guided setup wizard and a live connection test, so you find out it works before you schedule anything — not at 3 a.m. six weeks later.

9 Databases 8 Applications & SaaS 6 Hypervisors 6 Cloud 5 Hosting panels 4 Storage 3 Servers 2 Containers

Brand marks are shown to identify the platforms our connectors protect. All trademarks belong to their respective owners.

Cyber resilience

The backups are where you can still catch an attacker.

Production is compromised, but the backup stream is not — it is the one vantage point the intruder cannot tamper with. RemoteBackups analyses it.

Ransomware detection. A backup whose data suddenly stops compressing is a backup of encrypted files. Reduction ratio, entropy, churn and mass-deletion signals are scored together by a model that learns from your analysts' verdicts.
IoC threat hunting. Sweep every recovery point for known-bad hashes, ransom notes and filename patterns — including feeds you subscribe to — and get a timeline showing the last clean point to restore from.
Answers, grounded in your data. Ask "when was my last clean backup of the billing database?" and get an answer assembled from the catalogue — with the recovery points it used shown alongside it.

Combination padlock resting on a keyboard

How it works

Five stages between your data and the bucket.

The pipeline is the product. Everything else — scheduling, retention, the portal — exists to drive it and prove it worked.

Consistent read

Quiesce the app, snapshot the volume or open a transaction — then read a coherent point in time while the source stays live.

Chunk & address

Split on content boundaries, not fixed offsets, so an inserted byte does not reshuffle everything after it. Each chunk is named by its BLAKE3 hash.

Dedup & compress

A chunk already in the repository is referenced, not resent. What is new gets zstd-compressed.

Encrypt

AES-256-GCM, each chunk bound to its content address so a blob cannot be swapped for another. Keys never leave your control plane.

Place & lock

Write to the chain you defined — primary, copy, archive — applying Object Lock retention at write time on the tiers you marked immutable.

Restore is the part we take seriously

On the way back, every chunk is decrypted, decompressed and re-hashed. If the bytes do not hash to the address the manifest claims, the restore stops. If fewer bytes arrive than the manifest declares, the restore fails. The manifest itself is authenticated with a keyed MAC over every field, so an edited chunk list is detected before a single byte is read.

Per-chunk BLAKE3 verification on every restore
Keyed manifest authentication — a forged chunk list will not verify
Short reads are errors, never silent success

Open hard disk drive showing the platter and read head

Deployment

Run it yourself, or let us run it.

Same code either way. Choosing managed should be about who carries the pager, not about what you are allowed to keep.

Self-hosted

One VPS, one Compose file, your own MySQL and your own buckets. No component phones home, so it works air-gapped — the portal ships every icon and image it renders rather than fetching them.

Docker Compose · Kubernetes Helm chart · systemd

Managed

We run the control plane; your data still lands in storage you own and is encrypted with keys we cannot read. Multi-tenant with recursive sub-organisations and quotas, so MSPs can resell it cleanly.

MSP tenancy · per-tenant quotas · white-label roadmap

Pricing

Priced per workload, not per socket.

Storage is billed by your provider at their rates — we never mark it up, because it is your bucket.

Starter

Free

For a homelab, a side project, or proving the restore path before you commit.

Up to 3 protected sources
Full dedup, compression & encryption
Bring your own storage
Community support

Get started

Business

$49 /month

For teams protecting production. Everything in Starter, plus the parts an auditor asks about.

Unlimited sources
Immutable WORM repositories
Ransomware detection & threat hunting
Point-in-time recovery & DR plans
Tamper-evident audit log

Start free trial

Enterprise / MSP

Let's talk

For service providers and regulated estates that need tenancy, sovereignty and a contract.

Recursive multi-tenancy & quotas
Self-hosted or dedicated deployment
Air-gapped and sovereign-cloud options
Priority support & onboarding

Contact sales

Questions

The things people actually ask.

Where does my data physically live?

In storage you own and pay for directly — S3, Wasabi, Backblaze B2, Cloudflare R2, MinIO, Ceph and others. You add the bucket credentials; we never proxy or re-host your bytes. If you stop using us, the data is already where you put it.

Can you read my backups?

No. Chunks are encrypted with AES-256-GCM before upload using a key derived from your organisation's master key. On a self-hosted install that key never leaves your server. Storage providers see opaque blobs under HMAC-masked names — not your file names, not your content hashes.

What actually makes a backup "immutable"?

S3 Object Lock. Retention is applied at write time, and the storage backend itself refuses to delete the object until it expires — governance mode allows a privileged bypass, compliance mode allows none. This is enforced by the bucket, not by our software, which is the point: compromising the control plane does not grant deletion. Object Lock must be enabled when the bucket is created; it cannot be added later.

How do I know a restore will work before I need it?

Three ways. Every connector has a live connection test you run at setup. DR plans have a non-disruptive test mode that rehearses the recovery without touching production and reports measured RTO/RPO. And every restore verifies each chunk against its content address as it writes, so a corrupted or tampered backup fails rather than returning bad data.

Do I need to install an agent?

Usually not. Databases, hypervisors, cloud VMs, object storage and SaaS are all protected agentlessly over their own APIs. The agent is for machines that cannot be reached that way — physical servers, hosts behind NAT, and anywhere you want LVM/ZFS block snapshots with application quiescing.

Is it open enough that I am not locked in?

The repository format is documented and the whole stack is self-hostable: control plane, portal, data mover and agent. There is no proprietary appliance and no per-socket licence. You can run it entirely on your own hardware without any component reaching the internet.

Which platforms are production-ready today?

Maturity is shown honestly on every connector in the catalogue rather than buried. A connector marked Stable has a verified end-to-end round trip against the real platform; Beta means the integration is implemented and contract-tested but has not yet completed live certification. Of 43 connectors, 18 are Stable and 22 are Beta. We would rather tell you which is which than claim everything is finished.

Find out whether your backups restore.

Connect one source, run one backup, and restore it somewhere harmless. That is the whole evaluation — and it takes about ten minutes.

Start free Talk to us