Fortiche

Hi, I'm David. I'm a software engineer that loves to workout. Like many, I'm using AI to create my gym programs. I've used Anthropic Fable to build me the perfect workout app that takes in a Gemini/Claude/ChatGPT workout plan in full text and generate a workout program to track reps/sets/weights. I'm very pleased with the results. The app is extensively using iOS 27 APIs so will be available (for free) in September upon the new OS releases.

A strength-training app for iPhone and Apple Watch, built exclusively on the iOS 27 / watchOS 27 SDKs. Paste any workout program as plain text and on-device Apple Intelligence turns it into a structured, trackable plan.

"Fortiche" is French slang for someone who's really good at something.

  

Features

• Text → program: paste a program in any reasonable format ("Bench 3x5 @ 80kg", rep ranges, AMRAP, %1RM, RPE, rest times). Parsed on-device with the FoundationModels framework — nothing leaves the phone. A deterministic parser covers devices without Apple Intelligence.
• Live workouts on either device: full iPhone-only mode, or a watch-run session with heart rate. During watch workouts the phone shows a live, editable mirror — change weight/reps/sets from whichever device is closer.
• Quick workouts ⚡: crunches in front of the TV count too. Two steps — pick an exercise (recents first), confirm sets × reps — and you're in a real live session with rest timer and Live Activity, no program needed. Short sessions with logged sets always save. Or just tell Siri: "Start 3 sets of 20 crunches in Fortiche."
• Built for sweaty hands: one giant action per screen, Digital Crown weight adjustment, Double Tap to complete a set, no fiddly gestures.
• Rest timer with watch haptics, phone notifications, and a Live Activity countdown in the Dynamic Island.
• Live Activity with interactive Done Set / Skip Rest / Pause buttons.
• Apple Health: workouts save with one HKWorkoutActivity per exercise; they count toward your rings.
• Siri / App Intents: "Start my push day", "Log 8 reps at 80 kilos", "Skip my rest" — hands-free while lifting.
• History: personal records (estimated 1RM), weekly volume chart, last-session ghost while lifting, plate calculator, kg/lb.

Architecture

Piece	Role
FortichePack/	Shared SwiftPM package: SwiftData models, the workout engine, parsing, sync protocol, stats, App Intents
Fortiche/	iOS app
ForticheWatch/	watchOS app
ForticheWidgets/	Live Activity + widgets

Key design decisions:

• Command-sourced workout engine (ActiveWorkoutEngine): every mutation is a sequence-numbered command. The watch is authoritative during watch sessions; the phone applies edits optimistically and reconciles against echoed snapshots. State journals to disk on every command, so a crash restores mid-set.
• Transports: live sync rides HKWorkoutSession mirroring on real devices, with a WatchConnectivity fallback for simulators (mirroring's Rapport link doesn't exist between paired simulators — see docs/SPIKE-M1.5.md). Templates push phone → watch via applicationContext; finished workouts return dual-channel and upsert by UUID.
• Weights are always stored in kilograms; WeightUnit converts only at display/input boundaries.
• CloudKit-ready SwiftData models: optional/defaulted properties, no unique constraints, explicit order fields.

Building

Requires Xcode 27 (the project targets the iOS 27 / watchOS 27 SDKs) and XcodeGen.

xcodegen generate       # regenerate Fortiche.xcodeproj after editing project.yml
open Fortiche.xcodeproj

Tests live in the package:

swift test --package-path FortichePack

Set DEVELOPMENT_TEAM in project.yml for device builds. See CLAUDE.md for CLI build commands and simulator caveats.

Contributing — humans and agents welcome

This repo is deliberately structured so that AI coding agents can contribute at full quality. If you're pointing an agent (Claude Code or similar) at it:

1. Start at the repo root. CLAUDE.md carries the build commands, simulator caveats, and the hard rules; agents that respect it avoid the expensive mistakes.
2. Load the skill library. .claude/skills/ contains the project's institutional memory: a debugging playbook, the failure archaeology (every trap that cost real time, with the evidence that settled it), the architecture contract, diagnostic scripts, and a decision-gated campaign for the hardest open problem (real-device live-sync verification). Have your agent load fortiche-change-control before making any change, and the relevant sibling skill before touching a subsystem.
3. Prove it, don't eyeball it. Tests green (swift test --package-path FortichePack), both platforms building, and PRs that include the evidence — test output, screenshots, log lines. UI claims need screenshots; sync claims need logs; and nothing about watch↔phone mirroring may be claimed from simulator runs (it can't work there — see docs/SPIKE-M1.5.md).
4. Don't hand-edit generated artifacts. The Xcode project comes from project.yml (xcodegen generate), icons from Scripts/generate_icon.swift, the exercise dataset from Scripts/import_exercises.py.

Humans: same rules, fewer tokens. Full details in CONTRIBUTING.md. If the skill library ever drifts from reality, regenerate it by handing your agent docs/SKILL_LIBRARY_PROMPT.md.

Acknowledgements

Exercise data from free-exercise-db (public domain), originally derived from exercises.json by Ollie Jennings.