Back up every Linux host, database, file share and Docker host.
Restore without the server.

Lodestore runs backups for every server in your fleet from one place. Agents encrypt on the host and write straight to your storage, so a restore needs only that storage and your key.

Protect a host in 10 minutes Try the live demo

Open source, control plane includedSelf-hosted

lodestore.internal/overview

Overview

All 5 hosts protected

Newest backup just now

Host	Status	Last success	Size
web-eu-01role=web	Protected	3 minutes ago	6.2 GiB
pg-main-01role=db	Protected	22 minutes ago	412 GiB
files-02role=files	Protected	1 hour ago	1.1 TiB
ci-runner-03role=ci	Protected	2 hours ago	88 GiB
nas-02role=files	Protected	just now	2.4 TiB

nas-02 first backup finished

The server can go down.
Your restores stay.

The server schedules, catalogs and alerts. Backups never pass through it, so losing it costs you the dashboard, not your backups.

The server is gone. Any Linux machine with lode, the bucket and your password or recovery key restores every snapshot.

any Linux machine

# the server is gone; you still have the bucket and the key
$ export LODE_REPOSITORY="s3:https://s3.example.com/backups"
$ export LODE_PASSWORD_FILE=./repository-key
$ lode snapshots
3f9a1c2e  2026-11-01 03:00:12  pg-main-01  /var/lib/postgresql
8be04d71  2026-11-02 03:00:09  pg-main-01  /var/lib/postgresql
$ lode restore latest --target ./restored --verify
$

Everything a fleet needs.
Backups never pass through the server.

Five things Lodestore does for an ops team.

Every chunk stored once, sealed on the host.

Files are cut into content-defined chunks. A chunk that already exists in the repository is not uploaded again, across snapshots and across hosts.

What leaves the host is compressed, then encrypted with XChaCha20-Poly1305. File names are not readable in the bucket either.

How the format works

web-eu-01 MonABCD

web-eu-01 TueABCE

web-eu-02 TueABF

stored in your bucket, encrypted

ABCDEF

11 chunks in three snapshots, 6 stored.

A taken-over host cannot erase its history.

On S3-compatible storage the key the agents write with has no delete permission. Deletion and cleanup run only on the server, in two phases with a wait longer than the longest backup.

It works with S3 Object Lock, so even the server cannot remove data before its time.

Threat model

agentPUT data/3f/3f9a…c2e 200 OK

agentPUT data/8b/8be0…d71 200 OK

attackerDELETE data/3f/3f9a…c2e 403 AccessDenied

attackerGET data/3f/3f9a…c2e 403 AccessDenied

servergc mark 412 unreferenced packs

servergc wait longer than the longest backup

servergc check still unreferenced: 412

servergc delete done, dry run available

Click it together, or keep it in Git.

Schedules in a time zone with jitter and catch-up of missed runs, retention, upload limits. Edit a policy in the interface, or apply the same YAML from your GitOps repository.

Policies guide

Name

etc-and-home

Paths

/etc, /home

Schedule

30 2 * * * · Europe/Kyiv

Keep

daily 7

apiVersion: lodestore.io/v1alpha1 kind: BackupPolicy metadata: name: etc-and-home spec: repository: main paths: [/etc, /home] schedule: cron: 30 2 * * * timezone: Europe/Kyiv retention: keepDaily: 7

One search across every host.

Someone deleted a config last week, on a host you don't remember. Search a path or a pattern across the snapshots of the whole fleet, pick the version and restore it to that host or another one.

Find a file

nginx.conf

web-eu-01 /etc/nginx/nginx.conf14 versionstoday 03:00

web-eu-02 /etc/nginx/nginx.conf14 versionstoday 03:02

ci-runner-03 /etc/nginx/nginx.conf3 versionsOct 28

Hear about a failed backup before you need it.

Alerts for failed, overdue and offline hosts, failed checks and agent updates. Send them to Telegram, Slack, email or a webhook, and scrape Prometheus metrics.

Alerts guide

Backup failed on pg-main-01
ops-telegram · pre-backup hook exited with 1

files-02 is overdue
#ops-alerts · no successful backup for 36 h

Resolved: pg-main-01 backed up
[email protected] · 26 minutes later

Built for the day a host is compromised.

An agent runs as root on every host. So we designed for one of them being taken over, and wrote down exactly what follows.

What a stolen host key can do

YesWrite new backups	Its job. New data is encrypted on the host before upload.
NoRead backups back	The agent key is write-only. A restore gets short-lived links to the packs of its own job.
NoDelete a backup	The key has no delete permission. Deletion runs only on the server and works with S3 Object Lock.
DependsOverwrite an object	Refused where the storage enforces If-None-Match. Where it does not, versioning or Object Lock keeps the earlier version.
NoPush an unsigned agent build	An agent installs an update only after checking its minisign signature.

Post-quantum from day one

Pack keys are sealed with X-Wing, X25519 plus ML-KEM-768. A bucket copied today stays closed if either one breaks.

XChaCha20-Poly1305X-WingBLAKE3Argon2idminisign

Tested, not promised

How we test lists the tests behind these claims, from the agent key being refused a read to a restore from the bucket alone.

Threat model How we test

From zero to the first backup in about ten minutes.

Start the serverOne container, a static binary or a Helm chart. SQLite by default, PostgreSQL for production.
Connect your storageCreate the repository and save the recovery key.
Add a hostCopy one command to it. The agent checks the checksum and registers.
Watch the first backupCreate a policy and run it. Progress shows under Jobs.

Read the quickstart

export LODESTORE_HOST=192.0.2.10
docker run -d --name lodestore \
  -p 8080:8080 -p 8443:8443 \
  -v lodestore-data:/var/lib/lodestore \
  -e LODESTORE_PUBLIC_URL="http://$LODESTORE_HOST:8080" \
  -e LODESTORE_GRPC_PUBLIC_ADDRESS="$LODESTORE_HOST:8443" \
  ghcr.io/lodestore/lodestore:0.1.0

export VERSION=0.1.0
export ARCH=x86_64
base=https://github.com/lodestore/lodestore/releases/download/v$VERSION
for f in lodestore-server-linux-$ARCH lodestore-server-linux-$ARCH.minisig \
         lode-linux-$ARCH lode-linux-$ARCH.minisig SHA256SUMS SHA256SUMS.minisig; do
  curl --proto '=https' --tlsv1.2 -fsSLO "$base/$f"
done

helm install lodestore oci://ghcr.io/lodestore/charts/lodestore --version 0.1.0 \
  --set publicUrl=https://lodestore.example.com \
  --set grpc.publicAddress=lodestore.example.com:8443

Your storage

AWS S3MinIOCephOther S3-compatibleLocal or NFS folder

Agents run on Linux (x86_64, arm64) today. Windows and macOS are on the roadmap.

How it compares

	Lodestore 0.1	restic	Plakar (community)	Bareos
Shape	One server binary and an agent on each host	A command-line program that runs when you start it	Command-line engine, ISC license	Director, a File Daemon on each client, a Storage Daemon and a PostgreSQL catalog
Protection from a compromised host	Agent key without delete rights on S3-compatible storage; deletion is server-only	Append-only mode is provided by the backend, for example rest-server	Immutable backups on object-lock storage, per the Plakar website	Tape and WORM media for ransomware-resilient setups
Platforms	Linux, x86_64 and arm64	Linux, macOS, FreeBSD, OpenBSD, Windows	Linux, macOS, Windows, FreeBSD	Linux, Windows, macOS, FreeBSD
License	Apache-2.0 (agent, format, CLI), AGPL-3.0 (server)	BSD-2-Clause	ISC license	GNU AGPLv3