Inspiration
Most people who train seriously juggle four apps: one for workouts, one for calories, one for sleep and recovery, and one for journaling or breathing. None of them talk to each other, so none of them can tell you the thing you actually want to know in the morning: how hard should I train today, and what should I eat to recover? I wanted one app that sees all four, and a coach inside it who has seen your whole week, not just the screen you're on.
What it does
Awj is built around four pillars, each with its own colour so you always know where you are:
- Forge (training): programs for your goal, fast set logging with rest timers, progression that tells you when to add weight, and a Live Activity while you train.
- Fuel (nutrition): calorie and macro targets from your own body and activity, barcode scanning, a large food database, and one-tap re-logging of your usual meals.
- Fortress (recovery and mind): last night's sleep stages, HRV and resting heart rate, guided breathing saved as mindful minutes, and a private journal you can lock with Face ID.
- Foundation (your day): a readiness score every morning from Apple Health, one dashboard for today, habits, and a weekly score across training, recovery and food.
Awj is also the name of the AI coach inside the app. You can ask it anything ("I slept badly, what should I train today?"), log a meal by photographing it or typing "3 eggs and a coffee", talk to it hands-free during a workout with Awj Live from your iPhone or Apple Watch, and have it design a training program that you review before it replaces yours.
It runs on iPhone, iPad and Apple Watch, with widgets and Siri shortcuts, and it's available in 31 languages with full right-to-left Arabic and Hebrew.
How we built it
Awj is native Swift 6 and SwiftUI with SwiftData, on iOS 26 and watchOS, with strict concurrency throughout.
- AI: Google Gemini. Chat, food photos and plain-language meal logging use gemini-3.5-flash-lite through Firebase AI Logic with structured output, so a described meal comes back as data the app logs directly. Awj Live uses the Gemini Live API for real-time voice; the Apple Watch relays its microphone through the iPhone so the Watch holds no credentials. Awj acts through more than a dozen function-calling tools: it logs meals, workouts and mood, queries records and readiness, remembers facts about you, and builds programs by searching the exercise library and submitting a draft you approve.
- Security: the Gemini key never ships in the app. A Cloud Function issues single-use Live tokens, and every request carries an App Attest check through Firebase App Check.
- Monetization: RevenueCat. Everything is free except the AI coach, which is free for everyone's first 7 days; Awj Pro (weekly, monthly or yearly) keeps it going with a larger daily allowance. The paywall reads the RevenueCat "default" offering, the "pro" entitlement unlocks the allowance, and restore works from Profile.
- Data: Apple Health for sleep, HRV, heart rate and activity. Workouts, meals and journal sync through the user's own private iCloud with CloudKit; health data stays on the device, in a store that never syncs.
- Quality: over 1,000 unit tests, design-system lint tests that enforce the colour and material rules, and screenshot checks of every screen in English and Arabic, light and dark.
Challenges we ran into
- Making an AI coach safe to trust with health data: health data only reaches Gemini with explicit consent, which can be switched off anytime, and it never goes to iCloud (App Store guideline 5.1.3).
- Real-time voice on the Watch: the Watch can't hold API credentials, so its microphone is relayed through the iPhone into the Live API session.
- Languages: the coach answers in 31 languages and 17 regional varieties, from Egyptian and Gulf Arabic to Brazilian Portuguese and Cantonese, and the whole app mirrors for right-to-left.
- Keeping numbers honest: one definition per number (calories, macros, readiness) so the dashboard, the widgets and the Watch never disagree.
Accomplishments that we're proud of
- One coach that can see training, food, sleep and mood together, and act on them with tools rather than just chat.
- A design system where colour carries meaning: each pillar owns one hue, glass carries the surface, and a lint test fails on any stray colour.
- A complete Apple Watch app, not a mirror of the phone: workouts with heart-rate zones, guided breathing, readiness, and Awj Live from the wrist.
What we learned
An AI feature is only as good as the context and the actions behind it. Giving Awj structured tools (log this, query that, draft a program) made it far more useful, and far more predictable, than a better prompt ever did.
What's next for Awj
Log in or sign up for Devpost to join the conversation.