← Vaulith
Verify, don’t trust

Verify it yourself: watch Vaulith on the wire

We keep saying your files never leave your device. Words are cheap — so here’s the reproducible way to test that claim on your own machine, with tools we don’t control. It takes about fifteen minutes, and you don’t need to be a network engineer.

Last updated: July 2026 · There are deliberately no screenshots on this page — a screenshot is just another claim. This is a method you run, not a capture you take our word for.

The claim you’re testing

Vaulith is not fully offline, and we’ve never claimed it is — how your privacy works lists every connection it makes. What we do claim, and what you can test here:

Method 1 — Little Snitch (macOS, the clearest view)

Little Snitch is a per-app firewall that shows every outbound connection an app attempts, live. (Its free demo mode works for this test, as does LuLu, a free alternative from Objective-See.)

  1. Install Little Snitch and open its Network Monitor (menu-bar icon → Show Network Monitor).
  2. Launch Vaulith and find it in the Network Monitor’s app list.
  3. Run a full scan — point it at your biggest folders, and let the AI analysis pass run too.
  4. Watch Vaulith’s connection list while it works.

What you should see: a version check to GitHub at launch, and — only if you’re signed in — connections to Vaulith’s backend for the license check and scan summary. During the actual scanning and AI analysis: nothing new. The AI exchange happens over loopback to Ollama on your own machine, which never crosses the network interface at all.

Method 2 — Wireshark (macOS & Windows, the deep view)

Wireshark captures the raw traffic. Vaulith’s connections are TLS-encrypted (as they should be), so you can’t read the payloads — but that’s fine, because the two things that would expose a lying cleaner are visible anyway: who it talks to and how much it sends.

  1. Install Wireshark and start a capture on your active interface (Wi-Fi or Ethernet).
  2. Apply this display filter to see every DNS lookup and every TLS connection request, including the hostname each one is for:
dns || tls.handshake.type == 1
  1. Run a full Vaulith scan, including AI analysis, and let it finish.
  2. Check the hostname list against the table below — it should be that short.
  3. Then check the volume: Statistics → Conversations, sort by bytes sent from your machine.

The arithmetic is the proof. A scan that reads hundreds of gigabytes off your disk should produce outbound network traffic measured in kilobytes. You cannot hide gigabytes of uploads inside kilobytes of traffic — encryption hides content, it can’t hide size. If Vaulith were exfiltrating your files, this one sort-by-bytes view would catch it every time.

One quirk worth knowing: the local AI traffic won’t appear in this capture at all, because loopback traffic never reaches your Wi-Fi or Ethernet interface. To watch Vaulith talk to Ollama, start a second capture on the Loopback interface with the filter tcp.port == 11434 — that’s your files being analyzed, on your own machine, going nowhere.

Method 3 — Activity Monitor (macOS, zero install)

No tools, coarse but honest:

  1. Open Activity Monitor (Applications → Utilities) and switch to the Network tab.
  2. Find Vaulith in the process list and note its Sent Bytes.
  3. Run a full scan and watch the number as gigabytes of disk are read.

Sent bytes should stay roughly flat — kilobytes, not gigabytes — no matter how much disk the scan covers. On Windows, Resource Monitor (run perfmon /res, Network tab) shows the same per-process send totals.

What you should see — the complete list

When Expected traffic
Signed out, scanning & AI analysis Loopback only: 127.0.0.1:11434 (Ollama, on your machine — never leaves it). Plus one version check with GitHub Releases at launch.
Signed in The above, plus Vaulith’s backend — currently a *.supabase.co hostname, our infrastructure provider — for the license check and the scan summary (counts, sizes, timestamps, device identifier; a few kilobytes).
Scanning iCloud/OneDrive-synced folders No new cloud-provider download traffic — cloud-only (dataless) placeholder files are detected and skipped, never downloaded just to be inspected.
Downloading AI models (one-time, only if you opt in) Ollama fetches model weights from its registry (ollama.com). That’s gigabytes coming in — model weights — with nothing of yours going out.
If the app hits an internal error An anonymous crash report — the error type and a path-scrubbed message. Never your files, file names, paths, or account details.

Hostnames reflect our infrastructure as of July 2026 and may change as we grow; the shape of the traffic — and what’s never in it — is the promise.

What you should never see

If you see something not on this list, we want to know immediately. Email support@vaulith.com with the subject SECURITY REPORT and what you captured — we treat it as a security incident, with the process described on our security page.

Run the test — it’s free

Download Vaulith, open a network monitor beside it, and scan your messiest drive. The free tier needs no account at all — which makes the signed-out test the easiest one to run. For the full inventory of what leaves your device and why, read how your privacy works.

Download Vaulith free