Glasshouse

What it actually does

Most privacy apps ask you to trust them. Glasshouse does the opposite: it turns your own phone into the surveillance machine for a few minutes so you can see it — then proves every byte stayed on your device. Four chapters:

• Understand — a short, plain-English walkthrough of how the modern web quietly follows you, plus the real human cost when your attention is the product.
• You — builds the dossier a broker could assemble from your phone (an exposure score, the ad segments you'd be filed into), then plays “A Day in Your Life”: your real morning, commute, purchases and late-night scroll, scene by scene, each one cited and showing what you're worth — roughly $300–600 a year, anchored to the ad industry's own filings.
• Defend — a built-in Safari tracker blocker, plus genuinely private tools to cut the tracking down.
• Proof — “How do I know I'm not being tracked?” Put the phone in Airplane Mode and it all still works; a live ledger shows 0 bytes uploaded; and you can read the exact data shown to you, with nowhere for it to go.

The promise that makes it different: 100% on-device. No servers, no accounts, no network — the AI runs on your phone's own chip. The author has zero access to your data because there's literally no channel for it to reach anyone.

Two parts: learn it, then see it

Glasshouse walks you through it in order, so the results actually land.

Part 1 — “How the web tracks you” (interactive)

A hands-on walkthrough that runs the whole ad-tracking machine twice — once assuming your ad tracking is off, once assuming it's on — so you can see exactly what that switch does (and what it doesn't). It covers how they identify you, the split-second auction where you're sold, how it quietly steers your prices and feed, and how AI runs all of it. Every heavy idea ends with an “Explain it to me like I'm 5” note.

Part 2 — “See what they could get on you”

Then it reads the scraps on your phone — where your photos were taken, what's on your calendar, who's in your contacts — and asks an on-device AI, “what could a data broker infer about this person?” You get a polished, interactive report:

• Exposure Score — a 1–10 dial summarizing how exposed you are, and why.
• What they can infer — age band, where you likely live/work, job or income bracket, your social circle, habits and routines.
• Ad-targeting buckets — the exact audience segments an ad network would file you under.
• How you'd be targeted — tap to expand how a broker could approach you (clearly marked as possibilities, not ads you're getting).
• Who already has this — the apps detected on your phone, each mapped to the real, named company that monetizes that kind of data (Match Group, Meta, PayPal/Venmo, Chase's ad network, and the brokers who merge it all).

Save it as a PDF if you like — on your device, your choice where it goes.

The honest part most apps skip

Glasshouse goes out of its way not to give you false comfort:

• Turning ad tracking off isn't “safe.” It removes the easiest key (your IDFA) — but you're still tracked by the apps you log into, by your email and phone hashed into identity graphs, and by a hidden device signature (your screen, fonts and settings — your device, not your finger). One lock on a house with many open doors.
• Not seeing targeted ads isn't proof you're not targeted. The real work is invisible — the price you're quoted, the timing of a pitch, lookalike audiences, which posts surface. You can't trace it back, so it feels like nothing's happening. That's it working as designed.
• The profile is what could be inferred, not your ad feed. The ads you see are driven mostly by browsing and search this app can't read — so the two won't match, and the report says so.

It doesn't just show you — it makes you feel it

Two of the things the app does to you on purpose, then explains:

• It re-themes itself to fit you. After it profiles you, the whole app quietly switches color scheme to match your “vibe” (active, money, traveler, home, nightlife, tech). Then it tells you it did — because that's personalization: your own data reshaping your environment, exactly how feeds and stores tailor everything to keep you comfortable.
• It pulls the same re-engagement trick advertisers use. Once you've been through the “how it works” walkthrough, if you go quiet the app does what apps do when they sense you drifting — a buzz, a beep, a “wait, don't go — you seemed into X” notification using what you engaged with most — then reveals that it just profiled your attention and dangled your own interest back at you. (You can switch this off anytime.)

Then it helps you do something about it

Block the trackers

Glasshouse includes a built-in Safari content blocker that stops the very companies you just learned about — Google, Meta, Criteo and dozens more — across all your browsing. It reads nothing and sends nothing; Safari does the blocking from a bundled list. It also points you to other free, no-paywall privacy tools that meet the same bar (AdGuard, Brave, Firefox Focus, DuckDuckGo).

See the human cost

A short, carefully-sourced section on real people harmed by engagement- and ad-driven content while the platforms profited — so it never feels abstract. It follows safe-messaging guidelines and links crisis resources (988, 1-800-GAMBLER).

Use AI for good

The flip side: the same kind of AI, pointed at helping people instead of profiling them — real projects to admire (Be My Eyes, Flood Hub, iNaturalist, Tarjimly…) and hands-on links to start building something good yourself (fast.ai, Kaggle, Omdena).

How it was built — and how you could build one too

Honest version: I'm not an iOS developer — I don't know Swift. I built Glasshouse by telling an AI assistant (Claude) what I wanted, letting it write the actual code, then testing it and going back and forth until it worked. One person, no server, no data collection, no big budget. That is the point: the barrier to building something good has dropped to having the idea and caring enough to finish it.

What it's made of under the hood — the AI wrote it, I directed and tested it:

• The app: SwiftUI (Apple's free UI framework). I couldn't have written it from scratch; the AI did, and I refined it.
• The AI inside the app: Apple's built-in on-device Foundation Models (iOS 26). No model to download, no API bill, no server — it runs on the phone's own chip, which is exactly why nothing has to leave the device.
• The data: standard Apple frameworks any app can use — PhotoKit, EventKit (calendar), Contacts, CoreLocation, HealthKit, CoreMotion — all read locally.
• The blocker: a Safari content blocker extension (a simple bundled list of tracker domains).
• The privacy guarantee: the design choice to keep the AI on-device and ship with no network — so there's literally nowhere for your data to go.

Want to build your own? You don't need to know how to code.

I didn't. Here's the actual path — all free or low-cost:

• Build it with AI: describe your idea to Claude (or a coding agent like Claude Code) and have it write the app — then run it, tell it what's wrong, and iterate. That's the whole method I used.
• If you want to understand the code too: 100 Days of SwiftUI (free) and Apple's SwiftUI tutorials — optional, not required.
• Use on-device AI: Apple's Foundation Models framework — the same on-device model this app uses.
• Point AI at good problems: fast.ai / Kaggle, then build for impact with Omdena or DrivenData.

The real point: the same technology used to profile and nudge you is free, on your phone, and — with AI doing the heavy lifting — buildable by someone who isn't a programmer. Pick a problem you care about and start small.

What it can't see — and who can

Glasshouse is honest about its own limits. iOS sandboxes apps, so even with every permission granted, a third-party app like this physically cannot read:

• Messages your texts / iMessages
• Web browsing your Safari / Chrome history
• Purchases your payments, banking, what you buy
• App usage which apps you use and for how long
• Calls your call logs and precise carrier/cell location
• Email your inbox
• Other apps anything inside other apps' sandboxes

Here's the uncomfortable part: all of that is mined — just not by an app like this. It's collected by your phone maker, your carrier, your ISP, and the apps you've already trusted (Google, Meta, your bank, and the ad SDKs buried inside other apps you use every day).

So whatever dossier Glasshouse shows you is the floor, not the ceiling. It's the small slice one honest, on-device app can see — your real exposure, in the hands of companies that do upload and sell it, is far bigger. That's the point.

How to install

Not on the public App Store — so you install it directly. Two ways:

① TestFlight — easiest, recommended

No computer, no refreshing — just install Apple's free TestFlight app, then tap the invite:

1. Install TestFlight from the App Store (Apple's official beta app).
2. Tap the Glasshouse invite: testflight.apple.com/join/xVpeHGvA
3. Tap Install, open Glasshouse, allow the permissions you're comfortable with, and tap “See what they can get on you.”

✅ Live now — Apple approved the beta. Install TestFlight, tap the invite, and you're in.

② Sideload (no TestFlight) — for the impatient/technical

Install it yourself with your own Apple ID, using SideStore or AltStore (free sideloading tools):

1. Set up SideStore (or AltStore) on your iPhone once — you'll need a computer for the initial setup (their site walks you through it).
2. Download the app: Glasshouse-sideload.ipa
3. Open the file in SideStore, sign in with your own Apple ID — it signs the app to your account and installs it.
4. Open Glasshouse, allow the permissions you're comfortable with, and tap “See what they can get on you.”

Heads-up: with a free Apple ID the app needs a quick refresh every 7 days in SideStore (this is Apple's limit on sideloading, not the app). SideStore can do this on-device automatically once set up.

Requirements (both ways): iPhone 15 Pro / 15 Pro Max or newer (16/17 Pro), iOS 26, and Apple Intelligence enabled in Settings — the AI runs on the phone's Neural Engine.