I lost track of my own data across a homelab — so I built a whole-fleet disk analyser
Description and use
The author built Fathomline, a fleet disk analyser that aggregates metadata from multiple hosts, accurately detects duplicates, and provides safe cleanup workflows—all without sending data outside the network.
Problem
The author could not locate where data lived across multiple machines, leading to wasted space and risk of accidental deletion.
Outcome
Unified view of a multi‑host estate, accurate duplicate detection, safe cleanup workflows, and no data leaves the network.
Workflow
- Deploy a read‑only agent on each host (containerised, limited capabilities).
- Agent walks filesystem, collects metadata, pushes to central PostgreSQL/SQLite via mTLS.
- Central catalogue aggregates metadata, performs progressive dedup (size → head/tail → full BLAKE3).
- UI presents estate‑wide treemaps, duplicate groups, growth trends.
- When cleanup is needed, create a dry‑run plan, sign with Ed25519, quarantine files, re‑check hashes, then delete safely.
- Optional AI concierge and sandboxed previews run locally without sending data outside.
- Repeat scans nightly or on-demand via Scan Now.
- Monitor capacity via Notification Center.
- Use setup wizard for multi‑host configuration (CA, mTLS, enrolment).
Evidence
A lightweight, read‑only agent runs on each host, walks the filesystem, and pushes file metadata over mutually‑authenticated TLS into a central catalogue — PostgreSQL in production, SQLite for staging and single‑host setups.
Content‑verified dedup, done as a ladder. Hashing every byte of tens of millions of files is wasteful, so dedup is progressive: bucket by size → sample head/tail → confirm with a full‑content BLAKE3 hash.
- Record date: 2026-09-08T04:23:28.837Z