Hevy

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me a weight-training log that replaces Hevy: routines, sets ticked off as I do them, and proof that the weight is going up.

STACK
- A PWA that works entirely offline, or a native app if you want the watch and health integrations properly
- Local storage first: IndexedDB or SQLite on device. A gym has no signal, and an app that needs one is useless
- Optional sync server: Node 20+ with Fastify, SQLite through better-sqlite3
- Caddy in front

THE DATA MODEL
- exercises: id, name, muscle_group, secondary_muscles_json, equipment, kind, is_custom, notes — kind is weight and reps, reps only, time, distance, or distance and time
- routines: id, name, notes, folder_id, position
- routine_exercises: id, routine_id, exercise_id, position, superset_group, target_sets, target_reps, target_weight, rest_seconds, notes
- workouts: id, routine_id, name, started_at, ended_at, notes, bodyweight_kg
- sets: id, workout_id, exercise_id, position, kind, weight_kg, reps, rpe, rir, distance_m, duration_s, is_warmup, is_completed, completed_at — kind is normal, warmup, drop, failure
- personal_records: derived, never stored — best weight per rep count, best estimated one-rep max, best volume, per exercise
- measurements: id, kind, value, unit, at — bodyweight, circumferences, whatever is tracked
- Everything in one canonical unit internally, with the display unit a preference. Store kilograms and metres, show pounds and miles if that is what the user wants — mixing units in storage is unrecoverable

LOGGING A SET, WHICH IS THE WHOLE INTERACTION
- The screen during a workout shows one exercise, its previous performance, and the sets as rows to fill
- The previous session's numbers are pre-filled as the starting point. This is the single most important detail: nobody wants to type 100 and 8 again, they want to see what they did last time and change one of them
- Big targets. This is used with sweaty hands, one-handed, in a hurry, between sets
- A tick per set, and the rest timer starts by itself from the routine's rest for that exercise
- Plate maths shown for a barbell: what to load each side to reach the target
- Warmup sets marked and excluded from records and from volume
- Editing any set afterwards, including a whole past workout

THE REST TIMER
- Starts automatically on completing a set, counts down, and notifies at zero even with the screen locked
- Adjustable in place with two taps, and the adjustment can be saved to the routine
- Keeps running if the app is backgrounded, computed from a timestamp rather than a ticking counter

PROGRESSION, WHICH IS THE POINT
- Per exercise: a chart of top-set weight, estimated one-rep max, and total volume over time
- Estimated one-rep max from a named formula, with the formula stated. Epley and Brzycki disagree, and a number without its formula means nothing
- Personal records surfaced the moment they happen, during the workout, because that is when it is motivating
- Volume per muscle group per week, which is what actually drives programme decisions
- Frequency: sessions per week, and which muscle groups are being neglected
- Every number computed at query time from sets

ROUTINES
- Build a routine from exercises with target sets, reps and rest
- Supersets as a group that alternates
- Progression rules per exercise if wanted: add weight when the top set hits the rep target for two sessions
- Duplicate a routine, and a workout can deviate from it freely — the routine is a starting point, never a cage
- Folders for training blocks

THE EXERCISE LIBRARY
- A few hundred exercises with muscle groups and equipment, which is the tedious part. Assemble it from a permissively licensed open dataset rather than typing it, and credit the source
- Custom exercises with the same fields
- No animations needed. A clear name and the muscles worked is enough, and the licensing of exercise illustrations is a swamp

SYNC AND ESCAPE
- Optional sync of the whole database, last-write-wins per row with a version, conflicts kept rather than discarded
- Import from the common exporters' CSV formats, so a training history moves with you
- Export everything as CSV and JSON in two taps. A training log is years of your own effort and must never be trapped

WHAT NOT TO BUILD
- The social feed. It is what keeps people logging in the original, and it needs other people — a feed of one is worse than nothing. If you want the accountability, share a workout as an image or a link and stop there

OPERATIONS
- If there is a server: .env with DATABASE_PATH, BASE_URL, SESSION_SECRET; migrations on boot; nightly backup
- On device: works offline forever, and the export is two taps

WHAT MATTERS MOST
The logging screen and pre-filled previous numbers. Build that one screen, use it for a fortnight of real training, and fix everything that annoyed you. This app is used forty times a month in a noisy room and every extra tap is felt.

Give me the app, the sync server if you want one, the exercise dataset with its licence noted, and a README with build steps and the export format documented.