Every target has an arithmetic.
What you last lifted becomes an estimated maximum — and from that comes the next target, using the method you picked for that exercise. And when nothing has moved for weeks, you get a proposal instead of a shrug.
The target comes from what you last lifted
Weight and reps become an estimated maximum. What that turns into next is decided by the method you picked for this exercise — four are available, each with a sentence on what it is good for and what it is not.
- Double progression — work inside the rep window, add weight at the top of it
- Linear — a fixed step up for as long as it holds; it is meant to run out
- Top set and back-off — one heavy set, then volume
- Wave — the window rotates from session to session
Every method names what it rests on: meta-analyses and position stands with author, year and title. No claim without a source.
Method and history, per exercise. Example data.
You don't have to create exercises, and nearly all of them show you the movement
Barbell, dumbbell, machine, cable, bodyweight, kettlebell, bands, rings, landmine, trap bar, sled, plus explosive lifts, gymnastics and conditioning — 342 exercises, each one checked individually. 323 of them play a movement clip, alongside a muscle map separating primary from secondary work.
Every image has been looked at rather than matched by filename — a wrong picture teaches the wrong movement, so an exercise sooner has none than an approximate one.
- Free text resolves: "rdl", "dips", "shrugs" find their row
- Holds count seconds and carries count metres, not reps
- What's missing you add yourself — with your own photo
The catalogue, filtered. Example data.
Every exercise explains itself
What it is, what it gives you, how it is performed, which mistakes are typical — and the muscle map with primary and secondary separated. Below it your own history on exactly this machine.
The clip plays without a signal, because it travels onto the phone before you leave for the gym. Exercises without a clip get the drawing with start and end position.
Execution, muscle map, history. Example data.
When nothing happens for three sessions, it says so
And not only that it is stuck. First it looks at why: how much volume did that muscle group get this week, measured against your own bands? How was your recovery over the last ten days?
Fatigue leads to unloading. A blunt stimulus leads to the method. Recommending variation against missing recovery would be adding load to a problem made of load — so it doesn't.
Proposals come as a ladder: the method first, then the rep window, then a different exercise, and a deload week last. Only what you confirm gets applied.
A stall shows up first on the anchor curves. Example data.
Your goal decides the targets — not a standard programme.
Four strength goals, and they reach down into the individual set: rep range, effort ceiling, number of hard sets, rest.
Four goals
Hypertrophy, maximal strength, maintenance, coming back. Return after a break and the targets are deliberately smaller — and you get moved up automatically after four weeks, without your own choice being overwritten.
Deload week
Every fourth week takes load off — switchable, if you want it differently. If a recovery signal arrives at the same time, the two don't stack: both may only reduce, never raise.
Volume bands that fit you
How many hard sets a muscle group tolerates per week depends on whether you also run and ride. Cox holds two sets of landmarks and picks between them from what you stated and what it measured.
Anchor lifts
Cox measures your progress on three to eight exercises. It derives them from your last 90 days of training — or you pick them yourself. Train only with a barbell and you get no machine curve.
And you see where you stand against your bands
Sets per muscle group against the range that is productive for you — under it, in it, or over it. Next to that the progression per exercise with the next target, so the number in the gym is never a surprise.
"Too little" is not a reproach here but a number: average sets per week against the landmark, with the gap between them.
Volume and progression. Example data.
Your machine isn't "leg press". It's the leg press in your gym.
Two gyms, two machines, two completely different numbers on the same stack. Which is why progression belongs to the machine and not to the name of the exercise.
Photograph it once, recognise it after
You photograph the machine once, and next time you recognise it in the gym — with your settings, your last weight and your own curve on it.
Which gym applies comes from your location when the session starts. If that's wrong, the line in the header is tappable: create a gym, keep the last one, or train without one — then nothing runs onto a machine nobody stood in front of.
The machine gets a face. Example data.
Bring your own plan
Got a plan from a trainer on paper, a screenshot from another app, or an export from Strong or Hevy? All three routes end in an ordinary routine you can edit afterwards.
What Cox cannot match confidently it asks about — with candidates to choose from, and the import button stays locked while a question is open. A guessed exercise would otherwise sit in your routine for three months.
With an export the history comes along. It is written as a baseline, not as an event: no two hundred retrospective personal-best notifications for days long gone.
Recognised — and asked, where unclear. Example data.
21 ready-made plans — or build your own
Across all four strength goals and three experience levels. Every plan says what happens, who it is for, who it is explicitly not for, how many sessions it costs and what it rests on.
An adopted plan is an ordinary plan afterwards: swap any exercise, change any set. There is no second, sealed-off planning system — what you edit is yours.
The library, filtered by goal and level.
Strength is the way in.
Nutrition joins when you need it — and the coach sees both. Everything is free during the closed beta.