Desktop AI App Resource Audit - Where Did 150 GB Go?

net-pipe monitor · NOMINAL

38.9

GB / month est.

drag to orbit · wheel/pinch zoom

Traffic model

Workspace sync frequencyevery 10 min

Telemetry levelstandard

Context / index size2.0 GB

Audit checklist

✓

macOS: Activity Monitor → Network tab → sort by "Sent Bytes" / "Rcvd Bytes" per process.

✓

Per-app totals: Settings → Network → check cumulative data per application (or Windows: Settings → Network → Data usage).

✓

Firewall rules: use Little Snitch / LuLu (macOS) or built-in firewall to log and cap the app's outbound connections.

✓

App settings: disable auto-indexing of large folders, lower telemetry, exclude node_modules / build dirs.

Audit progress: 0 / 4

Why AI desktop apps eat bandwidth

Workspace indexing & embedding sync: the app hashes, chunks and uploads your codebase so the model has context — re-uploading deltas every sync tick.
Session streaming: every prompt round-trips large context windows; agent modes stream continuously for hours.
Telemetry: event logs, crash dumps, usage analytics — small per event, huge at "verbose".
Auto-update: multi-hundred-MB Electron builds pulled silently, sometimes per channel per week.

The math behind 150 GB

GB/mo ≈ (index_GB × resyncs/day × Δ-ratio × 30)
+ telemetry + streaming + updates

A 10 GB workspace re-syncing even 10% deltas every 10 minutes is ~1 GB × 144/day = far past 150 GB fast. The reported case is entirely plausible with aggressive sync on a big repo — which is why per-process measurement, not vibes, settles it.

How to measure it properly

macOS: Activity Monitor's Network tab shows per-process bytes since boot; nettop -P in Terminal gives live per-process rates. Little Snitch keeps historical per-app totals.

Windows: Task Manager → App history, or Settings → Network & internet → Data usage for 30-day per-app totals. Resource Monitor shows live TCP per process.

Measure a full week — sync spikes are bursty and a 10-minute sample lies.

How to limit the drain

Exclude bulky dirs (node_modules, .git, media, build output) from indexing.
Set telemetry to minimum; opt out of usage analytics where offered.
Lengthen sync intervals or switch to on-demand sync.
Cap the app with a firewall rule on metered connections.
If drain persists, file it — the 150 GB report got traction because the user brought receipts.