By Engineer Alex · 2025-04-15
A no-fluff setup guide from the engineer who built half of the app. The exact 14-day onboarding I wish every new user followed — what to turn on, what to ignore, and the small settings that make the AI dramatically smarter about you.
Hey, Alex here. I'm one of the engineers who built Habits, and I spend a lot of my week looking at how people actually use the app. The single biggest thing I notice is this:
The people who get great results in month 1 all do the same five things in week 1. And the people who churn skip them.
So this is the playbook. Two weeks. Boring but high-leverage. Do these and the AI gets smart about you fast.
Open the app and do these in order:
1. Pick your medication and dose. This is the single most important field — almost every recommendation downstream uses it. 2. Set your dose day (e.g. Sunday morning). Reminders, side-effect tracking and the weekly review all hinge on this. 3. Enter your starting weight, height and goal weight. Don't worry about precision; we use trend lines, not single readings. 4. Set a daily protein target. If you're not sure, tap "use the suggested target" — we calculate it from your lean mass, age and goal. 5. Allow notifications. I know, I know. But the smart-reminder system is one of the most under-rated features in the app, and it's switched off if you decline.
That's 10 minutes. You're set up.
This is where most people leave gold on the table. The AI is dramatically better when it has signal.
- Apple Health / Google Fit — turn this on first. Steps, weight, HRV and sleep flow in automatically. No daily logging needed. - Apple Watch / Wear OS — workouts auto-import. You'll never log a session manually again. - WhatsApp reminders (optional, available in select countries) — opt in with your number for hydration, movement and weekly-summary nudges outside the app. Reply STOP at any time to turn them off.
If you don't connect anything, the AI is still useful, but it's flying with one eye closed. Connect something.
This is the bit most people get wrong. They log the "real" meals and quietly skip the handful of nuts, the half a biscuit with coffee, the kids' leftovers, the spoonful of peanut butter from the jar. On a normal day that's fine. On a GLP-1, those tiny bites are often the difference between hitting your protein target and missing it, and over a week they add up to far more calories than people realise.
So for the first 14 days: log every food and drink that goes in your mouth. Not to police you — to give the AI an honest picture of how you actually eat. That's how it learns:
- How often you snack, and at what times of day - The patterns behind your snacking (stress, boredom, low blood sugar, certain meetings) - Your true protein intake vs. your target - Your real calorie pattern, not the cleaned-up version - The foods you actually reach for (so meal plans don't suggest things you'd never cook)
The fastest way to log: snap a photo. Our food vision model handles the macro estimate in a couple of seconds. Adjust the portion if it's wildly off — those edits train your personal model. For tiny snacks, the quick-add tile is usually faster than the camera.
Pro tip: The first 20 photos are the most important. After that, the AI starts auto-recognising your usual meals and snacks (your Greek yogurt bowl, your standard lunch, the 4pm dark-chocolate squares) and one tap is all it takes.
After two weeks of full logging you'll have one of the most useful pictures of your eating you've ever had — and a far smarter AI to coach you on it.
End of week 1, open Insights → Weekly Review. Read all of it, even if it's short.
Two things happen the moment you do:
1. The AI confirms which patterns it's actually picked up about you (so you can correct it if it's wrong). 2. It surfaces your first "test-and-learn" suggestion — a tiny, specific change to try in week 2. Do that one thing. Don't try to fix everything.
Now that the foundation is in:
- Add side-effect logging for nausea, fatigue or anything else GLP-1-related. One tap. The doctor-ready report depends on this. - Try one AI meal plan. Generate a 7-day plan, swap any meals you don't like, then commit. We've seen people stick to plans they helped tweak about 4× longer than ones generated cold. - Talk to the voice coach once. Even just for 90 seconds — it'll feel weird the first time, but for "what should I eat tonight?" while you're cooking, it's faster than typing.
I see these toggles get ignored constantly. Turn them on:
1. Smart reminders → Context-aware. Reminders shift based on your day, not a fixed clock. 2. Insights → Show data sources. When the AI surfaces a pattern, you'll see exactly which days/meals/workouts it was based on. Builds trust fast. 3. Privacy → On-device food recognition (where supported). Faster + photos never leave your phone. 4. Doctor report → Auto-generate weekly. So the PDF is always there when you need it. 5. Health connections → Background sync. Otherwise data only refreshes when you open the app.
- The streaks. They're motivating later, but in week 1 they create pressure. Just use the app. - The leaderboard / comparisons. None of that matters when you're learning your own baseline.
We genuinely read every piece of feedback. If something feels broken or stupid, the in-app "Tell us" button goes straight to a small Slack channel that the engineering team watches daily. It's the fastest way to influence the roadmap.
Two weeks of doing the above and the app stops feeling like a tracker and starts feeling like a coach that actually knows you. That's the whole point. — Alex