MemeScanr privacy: we literally don't have a server

What MemeScanr does not collect

• Your photos. Never uploaded. Never transmitted. Never mirrored to a server. Every photo stays on your device.
• Photo metadata. EXIF, location, timestamps — all read locally, never transmitted.
• Hashes of your photos. The 64-bit perceptual hashes MemeScanr computes for duplicate detection are stored locally in SQLite, never uploaded.
• Scan results. What MemeScanr found, how many duplicates you had, which categories dominated — all local only.
• Your contacts. Duplicate contact detection runs on-device via the iOS Contacts framework.
• Your account. There is no account. No email, no phone number, no password, no sign-in.
• Advertising identifiers. MemeScanr does not read IDFA. It does not show ads.
• Third-party analytics. No Mixpanel, no Segment, no Amplitude, no Facebook SDK. Only Firebase Crashlytics for crash reports.

What MemeScanr does collect (and only if you opt into diagnostics)

Apple's App Store privacy nutrition label for MemeScanr declares two data categories, both under "Data Not Linked to You" and both marked "Not Used for Tracking":

• Crash Data — via Firebase Crashlytics. Triggered only by an actual crash. Includes: stack trace, iOS version, device model, MemeScanr version. Contains no photo data, no gallery stats, no location, no personally identifying information. Used solely to fix bugs.
• Product Interaction — anonymous event counts (e.g. "scan completed", "Memory Lane opened"). Used for product analytics. Not linked to a user identity. You can disable this entirely in Settings.

On-device scanning: how it actually works

MemeScanr's scan engine runs a six-phase pipeline — Index, Hash, Classify, Group, Refine, Results — and every phase runs on your iPhone's CPU and Neural Engine.

• Hashing — 64-bit perceptual hash (pHash) computed in Swift, stored in local SQLite.
• Classification — Apple ML Kit models (memes, screenshots, blurry) run on-device. Apple ML Kit does not send inputs to Google or any server.
• Grouping — Hamming distance comparison of local hashes. No network call.
• Refinement — local re-check pass.
• Results — written to local SQLite. The scan results never leave your device.

The entire pipeline works with Airplane Mode on. You can turn off WiFi and cellular and run a full scan — nothing breaks. That is the simplest proof you will get that no upload is happening.

Why this matters: the privacy problem with "AI" photo cleaners

Most cleaner apps that advertise "AI-powered cleanup" or "smart AI" do their classification in the cloud. The user flow looks like this: the app quietly uploads every photo in your library to a third-party inference server, runs classification there, and sends a small JSON response back with "this is a meme / this is blurry / this is a duplicate." The convenience is that the classifier can be bigger and smarter than anything that fits on a phone. The cost is that your entire photo library — including screenshots of your banking app, your ID, your DMs, your medical portal, and anything else you took a picture of — is now on someone else's machine.

MemeScanr does not do this. Classification is Apple ML Kit, on-device, offline. The model is smaller than a cloud model, but it runs with zero exposure of your photos to anyone. For a photo cleaner, that trade is correct. A photo cleaner does not need a PhD-level classifier — it needs to reliably group duplicates and flag screenshots. Small on-device models handle that fine.

Deleting MemeScanr removes everything

If you uninstall MemeScanr from your iPhone, every bit of local data goes with it: the SQLite database, the hash index, the Backroom vault, the scan history, the Phone Therapist cache. There is no residual server-side copy because there is no server. Reinstalling starts from zero.