Can a prompt replace Google One?

Build a personal file store whose distinguishing feature is knowing what is in it.

Stack: your choice, Postgres, object storage or a local disk, Docker Compose. Consider deploying an existing self-hosted file platform and building the analysis as a companion — say which route you took.

Basics: upload, download, folders, rename, move, delete to a trash with a retention period, and share links with an expiry.

The analysis layer, which is the reason to build rather than deploy:
1. **Space by folder**, as a treemap and as a sorted table, computed from stored sizes rather than by walking the disk on demand.
2. **Byte-identical duplicates**: group files by content hash, show each group with total wasted bytes, and offer replace-with-a-link keeping one real copy. Never delete automatically. Show the paths so the user can see that two of them are backups they meant to keep.
3. **Near-duplicate images**: perceptual hashing to group the same photo at different sizes or with different compression, which is where a photo library actually loses its space. Present as suggestions only, with a side-by-side view.
4. **Cold files**: not accessed in N years, sorted by size. This is the list people actually act on.
5. **Growth**: stored bytes per month over time, with a projection of when your disk or budget runs out.

Hashing must be incremental — hash on upload, never re-walk everything — and the analysis must run as a background job with progress, not block a page load.

A monthly report emailed or written to a file: what grew, what was added, the biggest new folders, and the top duplicate groups.

Write tests for duplicate grouping across identical files with different names and timestamps, for the treemap totals matching the sum of file sizes exactly, and for perceptual hashing grouping a resized image with its original.

Do not build real-time collaborative editing, a mobile app, or photo face recognition.