zolder

Appearance

zolderStatus you can trust for a local-first backup

Observes your Syncthing mesh and wraps your restic backups, then emits one structured status report and pages you when something breaks. Never a sync engine.

Getting started
Configuration reference
View on GitHub

Observe, never reconcile

zolder reads Syncthing over its local REST API and reads restic snapshot freshness. It never diffs trees, resolves conflicts, or moves your data — Syncthing is the sync-of-record, restic is the backup-of-record.

Oneshot by design

zolder check runs from launchd / systemd timers / Task Scheduler, writes zolder.json, alerts on change, and exits nonzero when unhealthy. There is no long-running daemon to babysit.

Edge-triggered alerting

One page on a new failure, one all-clear on recovery — a persistent outage will not re-page every tick. A heartbeat dead-man's switch catches the case where zolder itself stopped running.

Cross-platform, config-gated

A single Rust binary for macOS, Linux, and Windows. Every check is opt-in per machine, so a laptop drops [zfs] and a box outside the mesh drops [syncthing].

What zolder is ​

zolder is the observability and orchestration layer over a local-first personal storage setup:

  • Syncthing carries your files between devices.
  • restic backs the synced tree up to object storage.
  • zolder watches both, plus canary files and (on the home node) ZFS pool health, then emits a structured zolder.json status report and alerts when something is wrong.

The design line is deliberate and load-bearing (ADR-0002): zolder observes Syncthing and wraps restic — it is never a sync engine.

What zolder is not ​

  • It has no reconciliation logic — no tree diffing, no conflict resolution, no rename detection, no tombstones. The moment a change would need any of those, the answer is "configure the underlying tool" or "file an issue upstream", never "implement it here".
  • It is not a backup tool. Taking snapshots is restic's job; zolder run backup only wraps restic to capture its exit code and duration.
  • It is not a monitoring service. It is a oneshot binary that a timer runs; a heartbeat ping to an external dead-man's switch covers the one failure mode a local tool cannot report on itself.

"A zolder status you trust beats a sync engine you don't."

Not yet built ​

Fleet aggregation (rolling every device's zolder.json up to the node), a system-tray indicator, and a serve dashboard are on the roadmap but not shipped yet — this guide documents only what the current release does.