Changes you can see or feel: app releases, training model changes, MCP tools. Infrastructure, refactors and internal tooling are left out. Each entry covers what shipped since the one before it, dated by its last change.
App
Your template keeps the plan, not the diary
A template says what you plan to do: 100 kg for 8, with two reps left in the tank. A workout says what happened: 102.5 for 6, nothing left. The app had been mixing the two up. If you finished a session with "Update template with today's changes" on, your day got copied into the plan, and the next analysis treated your worst set as what you meant to do. The toggle now writes only the things you meant. A swapped exercise. A set you added or took out. A target you changed before you did the set. What you actually lifted stays in the workout, and the analysis reads it there. The toggle only shows up when there is something like that to write.
AddedA note on each exercise in a template. The coach and connected clients have been able to write one since the 9th. From this build you can write one yourself in the builder, and it shows on the card while you train.
ChangedA skipped set keeps its target. The finish sheet is no longer taller than what is on it.
ChangedThe server now guards the workout-save path too, so an older version of the app cannot write a session's results into the plan either.
FixedA session left open overnight now ends at your last set, not at the tap the next morning, and it lands on the day you trained.
FixedA double tap on Add exercise no longer adds it twice.
Training model
Every number says what it counts
The strength climb compared this week with the first week in its window and nothing in between, so a lift that gained 13% and then started a new block at a lighter weight showed 0%. It now shows the peak you reached and how far below it you are, in the weekly review and to connected clients.
A half-finished week was being measured against a full one. Once alerts were switched on, that would have opened the Coach tab on a Monday with "10 muscle groups below MEV", the weekly minimum. The review now says whether the week is done, and the spike check and the alerts wait until it is. And the model had no idea you were cutting. Keep your bodyweight up to date on Profile and the weekly review picks it up, so before the coach suggests a deload it is told that a dip of up to about 3% on a lift is the diet doing its job, not a stall.
ChangedEvery figure a connected client gets now says what it is: how many weeks a trend was judged over, which muscles a volume total counts, one date format, and when it was made.
ChangedA swapped exercise is matched by the slot it replaced, not by its position in the list, so adding a lift no longer shows up as a swap that never happened.
ChangedA deload for one overworked muscle now applies itself even in an intensification or peaking week, where it used to wait for you to approve it. The check had been reading a value nothing ever set.
ChangedA volume-spike warning shows the same counts it judged, so you will not see "3 to 6 sets, 83%" again.
ChangedCoach "Active alerts" now reads real plateaus and muscles under their floor, and stays quiet until the week is over. It had never gone off before.
RemovedThe promise of a Training Score on the Intelligence tab before your first full week. It now says the weekly review unlocks then. The score is still in the MCP data, marked deprecated, until it is removed there too.
App
Connecting Claude is three steps, and the app knows when you are done
Profile → Integrations used to start with the plumbing. "Claude Desktop" and "MCP Server" sat next to each other, so if you just wanted to ask Claude about your bench press you first had to work out whether you were a developer. Now it is one row. Claude takes you through three steps, and on the third the app watches for the connection and changes when it lands. No guessing. Once connected you can see what your assistant did: reads and writes this month, a fourteen-day activity strip, and which tools it used. If the connection expires, it says so. API keys for scripts and other clients live under Developer access.
ChangedThe completion screen says what really happened. Session complete, logged short, or ended early, read from the record.
Fixed"Sign in" opened a sign-up form, so a returning email account could not get back in.
FixedThe Train card said "Day 2" on the fourth session of an upper/lower week.
FixedFinnish: buttons, chips, card headers and empty states that had been showing in English now use the translations.
ChangedThe active workout card counts working sets, keeps its toolbar within reach, and the line under a set no longer always says "up".
Training model
One definition of a hard set
Ask three parts of the app how many hard sets you did for chest this week and you could get three answers. The phone and the server disagreed on what a working set was. The weekly totals had four different writers. And how much credit a set gave each muscle depended on which screen was asking. There is one definition now. A hard set is a working set you finished with three reps in reserve or fewer, counted in full for the muscles doing the work. One engine counts every set the same way for the Intelligence tab, the coach and connected clients. One place answers what your minimum, target band and ceiling are (MEV, MAV and MRV), taking the band from your plan if you have written one and the floor and ceiling from the defaults.
ChangedYour plan's rep range now drives progression. The code was reading a field nothing ever filled in, so double progression used the built-in range no matter what your plan said.
ChangedA week that is still going is described as one. The app no longer passes judgement on a half-finished week.
ChangedA 2.5 kg step is never applied as 5 kg. The rounding is applied to the 2.5 kg step, not to the finished weight. If the equipment cannot make that step, the weight stays.
ChangedThe recap after a session reads the note you just typed, not just the numbers.
App
The coach opens on what matters today
Coach was rebuilt over a week. It no longer opens on an empty chat box. It opens on what is true right now (how recovered you are, what you trained last, what is due tonight) and decides which of those belongs at the top of the screen. You can send it a photo of a machine or a whiteboard, interrupt it mid-answer and have it actually stop, and pick up an old conversation where you left it.
AddedChange the session from the coach mid-workout, without leaving it.
ChangedThe activity log says what the coach did in plain words. It no longer lists tool names.
App
Your notes on a lift are read, not just stored
A note you write on an exercise ("the 25s were taken so I used 20s", "left shoulder hurt on the last rep") now reaches the analyst, the coach and connected clients. A drop in weight that you explained no longer reads as going backwards. Pain you report holds back progression on that lift the same week, instead of showing up three weeks later in a trend.
AddedNotes are read before the numbers they explain, not after.
AddedThe weekly review reads the whole week's notes, so a complaint that keeps coming back stands out.
App
The active workout screen, rebuilt around the set you are on
The card you are working from is now the main instrument. It has an estimated 1RM readout that moves as you log, a toolbar for the things you reach for mid-set, and you can drag exercises into a new order without leaving the session. Warm-ups sit in their own section above the working sets.
ChangedSet counts and the finish sheet count working sets, so "3 of 3" means what it says.
Training model
Prescribed loads you can actually set on the machine
A recommendation of 63.7 kg is no use on a stack that goes up in fives. Prescribed weights now round to the steps the equipment actually has, for each exercise. Periodization also learned two more goals: powerbuilding, which builds muscle with higher reps and then pushes strength with lower ones, and muscular endurance.
AddedPlate, stack and dumbbell steps worked out per exercise. There is no single step for everything.
MCP
Write your own training plan through Claude
A connected client can write its own training plan (phases, how to progress, how often to deload) and that plan takes over from the automatic engine until you pause it. Read tools now include a data-quality block, so an outside coach can tell a thin week from a bad one, and tools that delete things are marked, so clients ask before running them.
AddedBatch workout writes, so a client can log a whole session in one call.
AddedA plan you write shows up against your real routine. It does not replace it.
App
Your data, exportable and opt-out
You can export everything the app holds about you in one file. Usage analytics can be turned off in settings, and there is no tracking prompt when you open the app. It asks for nothing before it has done something for you.
AddedAccount deletion warns first if a subscription is still active.
Training model
Periodization that moves without you asking
Starting a new phase now rewrites volume and weight across your templates. Last block's numbers no longer hang around. A daily pass keeps volume, recovery and your place in the block up to date between sessions, so opening the app on a rest day shows where you are today, not where you were the last time you trained.
ChangedPhase changes are worked out from a fixed starting point, so one phase's cut does not stack on top of the last one's.
Training model
Which muscles a lift trains, worked out from the movement
Which muscles an exercise trains, and how much, now comes from the movement itself: which joints move and what each muscle does at them. Before, it was a flat tag on each exercise. A bench press no longer gives your shoulders the same credit as your chest, and the volume landmarks mean something as a result.
ChangedEach muscle gets its share of a set. Before, every muscle involved got the whole set.
App
Something to read after your first workout
Intelligence used to need weeks of history before it said anything. It now scores from your first workout on volume alone, tells you plainly which parts it cannot see yet, and fills them in as history builds up. A chip during post-workout analysis says the numbers are still settling, so you are never shown an old figure as if it were final.
AddedVolume drilldowns from your first session.
MCP
Povver as a data source for your own AI
An MCP server at mcp.povver.ai gives Claude and other connected clients your training history: workouts, progress per exercise, muscle state, weekly reviews. It comes with docs and example prompts, so an outside coach can ask real questions of real data instead of whatever you managed to remember and paste.
AddedPremium accounts connect through OAuth from the client's own menu.
App
The Intelligence tab
One place for what the analysis found: a training score, volume per muscle against your landmarks, and the reasoning behind each recommendation. The app moved to five tabs at the same time, with Train rebuilt around starting the session in front of you.
AddedLoading placeholders and pull-to-refresh across the analysis screens.
Training model
The Training Intelligence Engine
The first version of what the app is for. Every finished set is analysed across ten muscle groups, turned into a picture of volume, fatigue and progress for each muscle, and then into specific changes to your next session. Recommendations could apply themselves to your templates, with a cap on how far one change can move, and off by default.
AddedOnboarding rebuilt around the questions the engine needs answered.
The model behind these entries
Anything tagged Training model changes what the app prescribes. The current model, and where it takes a view, is written up in full on How it works.
