Workout Tracker

Your gym notes, turned into an app you keep on your home screen.

Workout sessions that open with last time's weights already filled in. No account, no server, no signal required.

Pick a day. The overdue one is badged.	Last time is already in the boxes.	Every kilo you moved, per day.

Most tracker apps make you look up what you lifted last time. This one puts it in the box before you get there.

You tap Push. Every exercise is already showing 85x9 80x8 75x8 in grey — exactly what you did last Push day. Hit ✓ on the ones you repeated, type over the ones that moved. A rest timer starts itself. You never search for anything.

Your sessions live in your phone's own storage. It opens in the basement gym with no bars of signal, because after the first load it doesn't need the network at all.

_{Eight weeks at a glance, then every session underneath it.}

Progress has two halves, because "am I getting stronger?" and "am I doing more work?" are different questions. Lifts answers the first, one exercise at a time. Workload answers the second, one training day at a time.

Each session becomes one number, labelled est. 1RM — an estimate of the heaviest single rep you could manage, worked out from a set you actually did. You never have to attempt a true max to get it.

It uses the Epley formula, on your best set of that exercise that session:

est. 1RM  =  weight × (1 + reps / 30)

So 80x9 scores 80 × (1 + 9/30) = 104 kg. The point is comparability: it turns 3×10 at 60 kg and 1×5 at 85 kg into one number, so the curve tells you whether you're getting stronger without you having to eyeball weight and reps separately.

Three things to know before you read too much into it:

• It's your best set, not your workload. Six hard sets and one hard set score the same if the top set matches. Total volume is shown per session in History.
• Drop sets don't raise it. 70x8+55x3+45x3 scores exactly like 70x8 — the drops are fatigue work, not a new strength ceiling. They do count towards volume.
• It drifts optimistic above ~10 reps. A 20-rep set estimates at weight × 1.67, which almost nobody could actually lift. The bias is consistent, so the shape of your curve stays honest even where the absolute number isn't.

Exercises measured in reps or seconds skip the formula entirely and chart best set or best hold instead. The manual has the per-exercise details.

The chart going flat is the thing you most need telling about and the thing you're least likely to notice, because nothing about loading the same weight again feels like a decision.

So every exercise is counted against its own best. Four sessions without beating it and it's flagged — as a chip on the card during the workout, while you can still do something about it, and at the top of Lifts as a list, worst first.

Each one comes with somewhere to restart from: about a tenth lighter, rounded to a jump your equipment can actually make. The point isn't the lighter weight, it's clearing the old number with room to spare. Rep-only exercises get no suggestion — "do fewer pull-ups" isn't a plan.

Counting from your best rather than a rolling average is deliberate: progressive overload says the number goes up again sooner or later, so that's what it measures you against. Four is low enough to catch a plateau early, high enough that one tired week doesn't trip it.

Workload adds up every kilo you moved — weight × reps across every set and drop set — and charts it per session, with the last four compared against the four before them.

It's always one day at a time, never a single combined line, and that's the whole design. A Legs session moves several times the tonnage of a Push one, so a combined chart would mostly track which day you happened to train rather than whether you're doing more of anything.

Two things worth knowing:

• Only kilos count. Plate-numbered machines and timed holds stay out of the total — their numbers aren't kilos, and adding them in would make it mean nothing.
• Dumbbell and one-arm work counts twice, once you've said which exercises those are under Settings → Which exercises count twice. A set written 30x8 is 480 kg off your chest, not 240. It can't be guessed from the name, so it's a list you tick — and because workload is worked out fresh each time, one tick fixes your whole history, not just what comes next.

The live demo above runs on one person's routine. The app has no sign-up because it has no server, so "using it" means running your own copy with your own exercises in it. That takes about five minutes and costs nothing.

1. Grab the code

git clone https://github.com/sartaj1995/workout-tracker.git
cd workout-tracker
npm install
npm run dev

2. Paste in your own workout

Open src/data/notes.ts and replace what's there with your routine, written the way you'd scribble it in a notes app:

Push
Chest press (sitting) - 85x9 80x8 75x8 70x8+55x3+45x3
OR
Dumbbell press - 30x10 25x11 25x10 25x9
Lateral raise - 25x7 22.5x9 22.5x8

Cable crunch - 40x15 40x12

Those starting numbers are just seeds; the app takes over from your first logged session. Machines numbered by plate, and lifts counted in reps or seconds instead of kilos, go in OVERRIDES at the bottom of the same file — the manual has the full reference.

You only have to get this roughly right. The same notes are editable inside the app afterwards, so nothing here is a decision you're stuck with.

3. Put it online and on your phone

Vercel detects Vite by itself — preset Vite, build npm run build, output dist, no environment variables. Open the URL on your phone and use Add to Home Screen. It then runs full screen with no browser chrome, and works with airplane mode on.

Edit your notes later and redeploy; the app fingerprints them, notices the change and rebuilds its exercise list on its own. Everything you've logged is left untouched. Delete an exercise and it's retired, not erased — it stops being offered, but its past sessions and chart still render.

Or skip the laptop entirely: Settings → Edit exercises opens the same notes in the app, so you can add a machine while standing in front of it. Your edits take over from the file at that point, and a redeploy won't overwrite them. Renaming an exercise asks whether it's the same one under a new name, because ids come from names and a silent rename would leave its history stranded.

Every line of this app was written by Claude Code. Not scaffolded by it, not autocompleted by it — written by it, a pull request at a time over a handful of days, from one person describing what they wanted in the gym and what annoyed them about the last version.

The human contribution was the taste. This app exists because every workout tracker on the store wanted an account and a subscription to do something a paper notebook already does better. So the prompts were things like "the numbers should already be there" and "a fresh phone must never overwrite the backup" — decisions about behaviour, not instructions about code.

It's an honest look at where that gets you: six thousand lines of TypeScript with two runtime dependencies (React and React DOM), no component library, no state library, no test suite, and a hand-rolled service worker. It's also unapologetically single-user software — built for one routine, one phone, one person. That's exactly why it's pleasant to use, and exactly why you should fork it rather than sign up for it.

If you want your own version of this, that's the whole method: open Claude Code in an empty folder, describe the app the way you'd describe it to a friend, then keep using the thing and complaining about it until it's right.

These live in the manual rather than here, because you don't need any of them to start:

• Editing exercises in the app — the notes syntax in full, what happens to history when you rename something, and why removing an exercise retires it instead of deleting it.
• Fixing a saved workout — what a correction does to next session's prefill, your best, and the chart.
• Backing up to Google Drive — optional and off by default. One backup file in your own Drive, re-uploaded after every workout you save, using a scope that can't see any of your other files. About fifteen minutes to set up. Until then, Settings → Export backup writes a JSON file whenever you want one.
• How it's built — the project layout and the handful of decisions that explain the odd behaviours: substitute days that don't advance the rotation, one exercise shared across two days, and why editing your notes resets an in-app choice.
• More than one gym — a design note for a feature that doesn't exist yet. What would have to change to train at two places whose machines share names, and why the answer isn't a second copy of the app.