workout_parser 0.1.0

Turns a workout routine written as plain text (a note, a chat message, an AI reply) into days, exercises, sets, reps, holds, rest, weight, tempo and supersets, and matches each name against 884 known [...]

workout_parser

Turns a workout routine written as plain text into structured data. The text can come from a note, a WhatsApp message or an AI chat:

Monday — Push
Incline DB Press 3 × 10, rest 90 sec
Lateral Raises 3 × 15
Plank 3 × 45 sec
Push-ups 2 × failure

import 'package:workout_parser/workout_parser.dart';

final routine = RoutineParser().parse(text);

for (final day in routine.days) {
  print(day.name); // Monday — Push
  for (final e in day.exercises) {
    print('${e.displayName}: ${e.sets} × ${e.target.label}');
    // Incline Dumbbell Press: 3 × 10 reps
    // Side Lateral Raise: 3 × 15 reps
    // Plank: 3 × 45 sec
    // Pushups: 2 × to failure
  }
}

It runs offline in pure Dart, with no API calls and no Flutter dependency, so it works in apps, servers and command-line tools. I built it for neFaRec, where you paste the routine you already follow and check what was read before you train it.

What it reads

You write	You get
Bench press 4x6-8	4 sets of 6–8 reps
Squat 5x5 @ 100kg, rest 3 min	weight 100 kg, rest 180 s
Deadlift 4 x 8, 6, 4, 2	per-set reps [8, 6, 4, 2]
Plank 3 × 45 sec, Wall sit 3 x 1:30	timed holds
Push-ups 2 × failure, AMRAP	sets to failure
RDL 3x10 tempo 3-1-1-0 @ RPE 8	tempo 3-1-1-0, note RPE 8
Lunges 3x12 each leg	perSide: true
A1 Leg curl 3x12 / A2 Calf raise 3x15	superset group A
Superset: followed by lines	one group for the block
Curls + pushdowns 3 × 12	two exercises in one group
3 rounds: followed by lines	3 sets each
rest 2 min on its own line	rest for the exercise above
Day 1, Monday, PUSH, ## Legs	a new day
Tip: …, (…)	a note on the exercise above

It also understands ×, en and em dashes, bullet characters, numbered lists and Markdown bold.

Check before you trust it

Pasted text is messy, so the parser never guesses silently. Every exercise tells you where each value came from:

e.setsFrom   // Provenance.explicit, .suggested or .defaulted
e.targetFrom
e.restFrom
e.hasSuggestions     // any value the parser filled in
e.match.needsReview  // the name was ambiguous or unknown
routine.flaggedCount // how many exercises need a look
routine.issues       // lines it could not read, with a reason
routine.notes        // advice lines that are not exercises

Push ups 20 reps gives 20 reps (explicit) and 3 sets (suggested). Plank 3 × 30 gives 30 seconds (suggested), because a plank is a hold. Highlight anything that is not explicit and let the user confirm it.

Exercise matching

ExerciseIndex.builtIn() knows the 884 exercises from free-exercise-db and 1,195 aliases people actually write, such as OHP, RDL, BB row, skull crushers and lat pulldown. Shorthand like DB, BB, KB and inc is expanded before matching.

Names that aren't an exact alias are fuzzy-matched by name similarity. The matcher keeps equipment apart, so DB press never becomes a barbell lift. Generic names such as press or row come back as MatchStatus.uncertain, with the closest alternatives in match.candidates.

To match against your own exercise ids, build your own index:

final index = ExerciseIndex(
  exercises: [
    IndexedExercise(id: 'sq', name: 'Back Squat'),
    IndexedExercise(id: 'pl', name: 'Plank', timed: true),
  ],
  aliases: {'squats': 'sq', 'front plank': 'pl'},
);
final parser = RoutineParser(index: index);

index.search('incline') gives ranked results for a "pick another exercise" list.

Options

RoutineParser(
  defaultSets: 3,           // when a line has reps but no set count
  defaultRestSeconds: 90,   // null to leave rest empty when none is written
  defaultUnit: WeightUnit.kg, // for a bare "@ 60"
);

JSON

ParsedRoutine, ParsedDay, ParsedExercise, SetTarget and ExerciseMatch all have toJson(), so you can send a draft to a backend or store it.

Accuracy

The test suite checks the parser against 39 routines written the way people actually write them: bullets, Markdown, supersets, circuits, timed holds, weekday splits and AI-formatted plans. It gets all 700 checked fields and all 124 exercise names right. A typical routine parses in about 10 ms.

Licence

MIT. The exercise names and ids come from free-exercise-db, released under the Unlicense.