Horme
What's new

What has changed

Recent releases, described by what you notice rather than by what we touched. The current version is 3.145.0.

v3.131.0 — The routine builder, rebuilt: classic programs and a form that is never lost

  • It no longer asks what it already knows. Sex, age, weight and height come from your profile, summarized and editable. The goal is picked from a single list and the emphasis (glutes, V-taper) only appears when it makes sense.
  • Classic programs inside “Routine style”. 5×5, GZCLP, 5/3/1, Texas Method, PHUL, PHAT, daily undulating and more, next to generated routines and Horme templates — only those matching your days and goal, with their evidence level stated plainly and adapted to your equipment and aches.
  • Only the methods that fit you. The methods step no longer offers plyometrics to someone after hypertrophy, or protocols for older adults to a 25-year-old.
  • The form is no longer wiped. Switching tabs, rotating the phone, reloading or pressing back returns you to the step you were on; back goes one step back instead of leaving the builder.
  • An expired session, called what it is. If your account stops recognizing the session, the app says so and offers to sign in again, instead of blaming your connection. And saving up to three routines is free.

v3.130.0 — A review of everything we say about the subscription

  • The Founder program now says the same thing everywhere. The store documents still talked about three months and a window closing on 31 October; it is two months and the window runs to 31 December 2026, for the first 200 sign-ups. The months still start counting the day billing is switched on, not on your sign-up date.
  • The prices announced outside the app match the ones inside it. A review of the amounts, the 7-day free trial on the annual plan, and what Premium includes — in the app, on the website, in the terms and in the store listings.
  • There is still nothing to buy. Subscriptions are not active: no feature is locked and nobody can be charged anything yet.

v3.129.0 — New binary package release for Android and Wear OS

  • Compiled and verified release packages. Production APK and AAB artifacts for mobile (build 160) and Wear OS (build 10160) signed with official Play Store certificates.
  • Full cross-platform parity. Continuous version synchronization across web core (PWA), Android, Wear OS, iOS and HarmonyOS.

v3.128.0 — The weight the app suggests now comes from the right session

  • An exercise in two rep ranges no longer mixes its days. When a routine does the same exercise heavy one day (6–8) and light another (10–12), the light day inherited the heavy day’s weight. Each range now progresses from its own history and its own stall counter; if there is no recent session in that range, the load is worked out from your estimated 1RM and the app tells you which set it came from.
  • An exercise you pick up again after weeks no longer asks you to go heavier. After two weeks without it, the increase you earned back then no longer applies: you return at that load minus a detraining discount, and rebuild from there.
  • Load jumps that exist in your gym. If you log in pounds, increases go up in steps of 5 (2.5 for isolation work) instead of asking for in-between numbers. And if your sets to failure show you report less reserve than you actually have, the suggested load now accounts for it.
  • Fixing a mistyped weight also fixes what was derived from it. From an exercise’s page you can edit all of its sets and its base weight; saving recalculates the record, the progression counter and next time’s load. A 400 typed instead of 40 no longer leaves a made-up record behind forever.
  • Manual food entry asks for amount and unit. Grams, millilitres or units, and whether the figures are for that amount or “per 100 g” from the label. It warns you when the calories do not add up with the macros.

v3.127.0 — Smaller packages, and both watch builds catch up

  • The installers carried the code twice. The APK, AAB and IPA all shipped the unminified source files alongside the bundle the app actually loads: 12.6 MB of dead weight in each binary. The APK drops from 135 to 130 MB, the AAB from 126 to 121 and the iOS archive from 124 to 119.
  • Garmin and HarmonyOS rebuilt for this version. The Connect IQ app is compiled again for all 205 declared watches, and the AppGallery package moves to the same version as the phone.

v3.124.0 — Six area audits: what progression, cardio and recovery were counting wrong

  • Progression counted your warm-ups. Warm-up sets and dropsets were included when computing the next load, so the app suggested going heavier on the strength of sets that were never working sets.
  • Switching between kilos and pounds reinterpreted your whole history. Already-logged weights were read in the new unit instead of being converted, so a preference change rewrote every previous record.
  • A finished mesocycle stayed in deload forever. On reaching the end of a block the app never left the deload week and kept lowering volume indefinitely.
  • Cardio: power and cadence never reached the recorded trace, the running trend added cycling kilometres to the running base, and FTP was estimated one way on one screen and another way on the next.
  • Recovery: sleep regularity reported the worst case when there was no data. Missing entries read as maximum irregularity instead of as missing; and six recovery capabilities existed in code with no screen calling them.
  • Progress no longer shows fabricated results. A family of hand-written outcomes was presented as if it were your own data, and the graph displayed coefficients that had never been computed. Without enough coverage the screen now says what is missing.
  • Home and Settings reviewed control by control. All 31 destinations render across thirteen different states with no broken text, none of the roughly 400 controls throws when pressed, the daily score is never a non-finite number, and what it reads aloud matches what is shown. And a flagged allergen now looks flagged.

v3.123.0 — Strength types and methods, and a routine built to your measurements

  • Strength training types and methods. The training area distinguishes the type of work and the method used to perform it, and that is kept in the session log and in the analytics.
  • A routine to your measurements. The generator now accepts a concrete brief —days, equipment, time per session and priorities— instead of only choosing between fixed templates.
  • Watch audit. The phone bridge, the four heart-rate sensors and the cardio, rest and set-editor screens were reviewed; all five artefacts were rebuilt because of it.

v3.122.0 — Cardio to your measurements

  • Custom cardio. Cardio prescription can be commissioned with the same real-world constraints as strength —what is available, how much time and the goal— and it integrates with the training week instead of living apart.

v3.119.0 — Horme adapts to the device: desktop sidebar and a readable line length

  • A real desktop interface. On a computer Horme is no longer the phone app stretched wide: a fixed sidebar with the six areas and the active area’s sections in view, without the thumb tab bar.
  • Readable text on wide screens. Paragraphs reached 226 characters per line on a monitor; none now exceeds 80, the limit for comfortable reading.
  • Every screen on the device where it works. GPS cardio recording and progress photos are no longer offered on the computer — there is no GPS and no suitable camera — and the screen explains why and where to go.
  • Updated release packages. New verified APK and AAB builds for Android and Wear OS, plus the IPA archive for iOS and watchOS.

v3.118.0 — Data consistency, Health Connect audit and updated release binaries

  • Updated release packages. New verified APK and AAB builds for Android and Wear OS, alongside the IPA archive for iOS and watchOS, with contract parity across all platforms.
  • Comprehensive audit and native parity. Full synchronization of deferred modules and content manifest without OTA degradation.

v3.117.0 — The week now says how much data it is talking about

  • The weekly summary declares its coverage. It reuses the nutrition diary classifier: usable days —complete, estimated or inferred— are separated from partial ones, and partial days are excluded from the averages. Logging only breakfast for seven days no longer reads as the whole week diet. The nutrition score needs at least three usable days, and intake findings four.
  • Unlogged water is no longer zero water. A day with no entry returns missing data; an explicitly logged zero does count. The average only includes logged days. It reports logged water, not how hydrated you are.
  • Cardio joins the weekly story. Cardio sessions, minutes and days now appear on screen, in the exportable document and in the context Fito receives. Several activities on one date count as one day and several sessions. On the strength side, an empty or warm-up-only session no longer inflates the counter.
  • Wider health import. Height, distance and speed are read as well; active and total calories have independent readings; temperature keeps its measurement location. A local archive stores the original readings with their provenance, can be exported separately from the general backup, and repeating a read updates the record instead of duplicating it. With history permission granted, the first sync reaches back up to 365 days; without it the window stays at 30.
  • Later settings are respected. Undoing or expiring a nutrition trial now reverts only the mode and the calorie target that trial changed. Macros or other preferences edited afterwards are kept.
  • Sign-in fixed on iPad. Sign-in was sending the Android-restricted API key on both platforms, without the headers that restriction requires, so iOS returned a connection error. Each platform now uses its own key.
  • The weekly summary interface loads when opened rather than at start-up. Coverage figures are product availability rules, not clinical thresholds, and estimates remain non-diagnostic.

v3.116.0 — Goals backed by evidence, a lighter start-up and an offline core

  • Home says what it actually observed. The goals card uses the canonical profile goals and the active Progress target; it counts as active days those dates with logged strength or cardio in the last 28 days, and shows zero when there are none. The projection only appears with a recent trend and enough history, and states its limits. Empty days are described as possible rest, not as a failure to comply.
  • One date, one observation. Several weigh-ins on the same day no longer inflate the slope or the amount of evidence: the last valid entry of that date prevails. Finite weights and real dates are validated, and daily intervals are computed in UTC.
  • The wellness check-in no longer answers for you. All four signals require an explicit answer, and saving preserves sleep and notes already recorded. A metabolic timeline that showed milestones not derived from records was removed; the agenda uses configured schedules and actually planned sessions.
  • The core works offline. Pre-caching goes from six to all nine deferred bundles and now includes cover images and the glossary. A JSON or JavaScript request with no cache no longer receives HTML with status 200: the shell is reserved for navigation and a missing resource fails explicitly.
  • Recoverable saving. If a write fails, the app says so and keeps a visible notice across screens with a retry button; changes remain in memory and the browser asks for confirmation before closing. The notice only clears when a write completes successfully.
  • Lighter start-up. The Settings interface loads when opened. The main bundle drops from 2,576 to 2,411 KB and accounted start-up from 3,650 to 3,486 KB, with the budget ceilings lowered accordingly. These are code sizes, not a measured speed-up on device.
  • Labels that say what they measure. The cardio proposal bands are now called “Aerobic stimulus” and “Added fatigue”, with a note that they describe the estimated effect of that session and not your current state. The glossary no longer triggers on ordinary words or inside controls, and readability at 320 pixels was fixed.

v3.115.0 — Meal planning, training and a clearer interface

  • More reliable content updates. Fixed compatibility between the publisher and integrity checks. The app protects its bundled content from older cached or server revisions.
  • Clearer in both languages. Expanded interface and accessibility translations. The Home summary describes available records without claiming unverified trends.
  • Expanded menus and recipes. Planning with preferences, restrictions, variety, pantry items and substitutions; explanations of trade-offs when no solution meets every target.
  • Activity-aware cardio. Indoor and outdoor workouts with dedicated controls, preparation before the start, and improvements to maps and route history.
  • Strength and safety. Improvements to routine generation and volume accounting; activity-readiness screening, eating-behaviour signals and response analysis with explicit confidence limits.
  • Readability and responsiveness. Fixes for alignment, cards, buttons and long content on small screens; deferred loading of secondary areas.
  • App estimates and signals are not diagnoses or clinical validation. Physical-device testing and distribution requirements are verified separately.

v3.112.1 — Horme becomes a personal fitness operating system

  • Training now behaves like an operating system: the pre-session check-in turns time, equipment, discomfort and readiness into a session budget; recommendations show confidence and offer a Plan B. Personal learning requires a minimum sample, remains inspectable and can be reset independently.
  • Modality-specific models and private collaboration: strength, cardio, intervals and mobility no longer share one universal formula. Coaches and other roles use permissions, versioned proposals, explicit acceptance, comments and revocation; no marketplace opens without verification, payments, moderation and legal coverage.
  • Nutrition v2: per-nutrient provenance, missing values retained as null, expenditure backtesting, day completeness, confidence-scaled photo uncertainty, hard allergen filters, budget-aware planning and a safety gate that blocks automatic deficits in risk contexts without supervision.
  • Longitudinal progress: Progress Score v2 declares coverage; it compares equal-length blocks, supports milestones and annotations, creates private monthly and annual stories, and answers local queries without mixing units. The gallery adds optional biometric locking plus independent export and deletion.
  • Cardio finally reaches every surface: Sensor Hub arbitrates each source; weather and altitude adjust the target band; the week shows impact load; swimming computes SWOLF, critical speed and five zones; fuelling, anaerobic reserve and imported zones surface with their limits.
  • Settings, privacy and control: context profiles, presets, intent search, system diagnostics, purpose-specific permissions, inspectable Fito memory, export, retention and visual-privacy controls turn configuration into an operational product surface.
  • The library grows to 525 exercises, including carries, trap-bar and kettlebell variants and more equipment alternatives. The website opens a bilingual page for every exercise and counts the catalogue directly from app data.
  • Cardio no longer ends when the session is saved. The lab brings race prediction, load, fitness, tests and plans together; race planning explains the target, pace, uncertainty and missing evidence before committing to a number.
  • Wearable samples now pass quality checks before moving a baseline. Horme normalises units, rejects physically impossible values, avoids double-counting sources and chooses one coherent source for each night of sleep.
  • Quality travels with the data: heart-rate coverage, sample count, rejected artefacts and whether sleep comes from stages, classification or a simple session. Wrist SpO₂ and skin temperature are presented as consumer trends, not clinical measurements.

v3.111.0 — Cardio keeps going with the phone locked and reaches the wrist in full

  • GPS recording continues while the screen is locked through the system location service, started only when you tap Start and stopped at the end. Background maps can be enabled or disabled; the locally recorded trace remains the route source.
  • Watches receive the route and heart-rate zones, not just a summary. Wear OS, Apple Watch, Garmin and HarmonyOS share the same target and effort language decided by the phone.
  • With separate explicit permission, Horme can copy five record types created in the app to Apple Health/Health Connect: weight, body fat and, from its cardio, calories, distance and average heart rate. It never modifies records created by another app.

v3.110.0 — Exercise order stops lying to your progression

  • The app now knows the fourth exercise is not the first one. Your whole progression compares today's session against the last time you did that exercise, and it assumed something that almost never holds in a real gym: that both times happened under the same conditions. You move the bench press to the end because the rack was busy, you drop one rep, and the app logged a failure. Two of those and it deloaded an exercise that was not stalling at all.
  • The app now estimates the performance discount from where an exercise sits in the session — larger the more prior work hit the same muscle — and uses it for two honest things: comparing your e1RM against its "fresh" equivalent, and explaining why today's suggested load is not last time's. It rests on the systematic review of exercise order (Simão, 2012), which is on the References page.
  • Your training notes finally do something. "The bench was taken", "shoulder twinge on the last set", "slept 4 hours" were the only qualitative data in the app and ended up on three screens with no consequence. They are now classified by topic, tied to the exercise you wrote them on, weighted by how recent and repeated they are, and turned into concrete proposals: swap the exercise, drop the load, move it in the session, open the pain screening or lengthen the rest. The text never leaves your phone: what reads it is a local dictionary, not a model.
  • The HarmonyOS watch is approved on AppGallery, with the corner truncation the review flagged already fixed: on a round screen, what fits centred depends on the distance to the edge through the chord formula, not on a percentage set by eye. The set editor and the history screen were recomputed with it.

v3.109.0 — Every area explains what it was already computing

  • An AI copilot in all five areas: Strength, Cardio, Nutrition, Recovery and Progress. The on-device engines still compute and decide; Fito receives an aggregated summary and its job is to prioritise ONE signal and turn it into a small action. It does not recompute tonnage, zones, VO₂ max, TDEE or load, and it cannot unlock a plan.
  • Recovery was the last area without a copilot, and most of what it computed was invisible to the model: sleep debt, regularity, social jet lag, adherence, wearable baselines and heart-rate recovery from cardio. All of it now feeds in.
  • The muscle map stops being half a mirrored silhouette and becomes an anatomical plate with two complete outlines — front and back, with head, hands and feet — and the 16 groups drawn over them.

v3.107.0 — Cardio grows up: its own tab, plans and maps

  • Cardio is now the second tab, and the lifting one is renamed “Strength”. It is not a relabelling: an area with full prescription, four routes of its own and GPS recording did not fit inside Training, and from there it did not even highlight correctly when opened.
  • The area went from measuring to prescribing: 19 structured sessions across five families with voice guidance, 7 periodised plans — from “start running” to the marathon — with a deload every fourth week, and 7 field tests each with its protocol, its source and its margin of error.
  • Cardiorespiratory capacity with FRIEND percentiles, age projection, critical speed, a mean-max curve and a heart-rate recovery trend. Test history stopped overwriting itself: repeating a Cooper adds a point instead of erasing the previous one — which is exactly what the screen recommends doing.
  • Real map background, optional and off by default. The route is always drawn from your own points and works offline; if you enable it in Settings → Maps, the provider’s tiles appear underneath. The track still never leaves the phone, and what you share carries Horme’s own drawing, never the provider’s map.
  • GPX import and export, with the route trimmed by the privacy distance you configure. Re-importing the same file updates the run instead of duplicating it.
  • Cardio from your watch lands in the same history as what you record and what you type, using the same activity vocabulary, and it counts towards the training-load model instead of living in a separate list.

v3.108.0 — Horme connects its data and Fito gains memory

  • Intelligent Home: the daily recommendation prioritises real states — safety, active session, fasting, Plan B, capacity, nutrition and recovery — and shows which signals it used and with what confidence. The first reading runs on-device; deep AI analysis only happens when requested, is not stored and expires when your data changes.
  • Horme Intelligence brings Data → Science → Models → Action into one flow. It answers “what changed?”, “why might it be happening?” and “what should I do now?” without inventing another score or claiming causes the data cannot prove. “Explain my body” runs locally and states coverage, noise and limits.
  • Fito now keeps up to 12 context messages, accepts 2,000-character questions, restores drafts and offers Retry or Edit when a response fails. Stop is available from the beginning of the operation.
  • Verifiable temporary privacy: temporary history and drafts are excluded from persistence, sync, exports and backups. The interface also explains whether a question travels through included Fito or directly to the provider with your own key.
  • Fito can propose logging wellbeing or pain with sanitised values and safety limits. Like every action, it shows a preview and waits for confirmation: understanding an intent does not grant authority to change your data.

v3.105.0 — Audio mode, and deciding before diagnosing

  • Hands-free audio mode: it calls out the set, narrates the rest, counts you down and understands spoken commands. With your phone in your pocket and the timer running, you touch nothing.
  • Home now opens with "Now": what to do comes before the day's diagnosis. Only a safety signal or a change of plan gets in front of it.
  • The Progress summary gives one verdict and one actionable priority, and states how many families of data it is speaking from instead of headlining on just one.
  • A muscle heat map over a real silhouette, with all 16 groups and their volume for the week. One look tells you which muscle is short and which is overcooked.
  • A routine analyser: it audits your whole programme across eight dimensions and returns findings with the concrete fix, not a bare score.
  • All four watches now show the full day at a glance — training, water, nutrition and streak — not just the workout.
  • On iPhone, four widgets for the home and lock screens, plus three complications on the Apple Watch face.
  • A seven-step quick tour plus ten per-area mini-tutorials, always available from the "?" on each screen. It used to be a thirty-one-step tour nobody finished.

v3.104.0 — Dictation in two languages, and the Huawei watch on its own

  • Voice dictation works in Spanish and English, and follows the app's language rather than the phone's. Recognition happens on the device when the language pack is present: your voice never leaves it, and it works offline.
  • The HarmonyOS watch trains on its own: it starts a session from its own catalogue and saves it to its own history with no phone involved.

v3.103.0 — Five screens overhauled and a precision audit

  • Home, Nutrition, Recovery, Progress and Settings reworked from top to bottom. In Recovery, three analyses that were written but never showed up surfaced: nothing was saving the data they needed.
  • A precision audit of the engine: nine calculations returned a number with the same confident face as a correct one. They affected prescribed load on long sets, overload warnings, predictions and the calories in alcohol.
  • Settings' diagnostics used to give a green tick while the system had notifications blocked. Now it says so.
  • A full entry per food, a database explorer, and protein distribution across the day.

v3.102.0 — One report, and a weekly review

  • The report inside the app and the one in the web dashboard were two, and said different things about the same data. Now it is one, wherever it comes out.
  • A weekly review: when the week closes, what you did, what fell short and what to change next week.
  • The web dashboard gains its own summary and charts. Sign in from any browser, nothing to install.

v3.101.0 — Household measures and effective volume

  • Log food in household measures — "2 eggs", "1 cup of rice", "1 tablespoon of oil" — covering all 2,160 foods, not just the popular ones.
  • Weekly volume is now weighted by RIR: sets far from failure stop counting as full sets.
  • Sets logged without an RIR are estimated from the prescribed RIR and your own median on that exercise.

v3.100.0 — Unified watch protocol

  • The four watches stop duplicating the phone's decisions: training readiness, heart-rate target and VO₂ max are computed in one place only.
  • Fewer discrepancies between what the watch says and what the app says.

v3.99.0 — More predictable AI

  • Fito's structured responses no longer depend on a single provider: they work the same across all four.
  • Fewer answers cut off halfway through on long requests.

v3.96.0 — Second wave of icons

  • The remaining generic emoji are retired and replaced with in-house icons consistent with the rest of the app.

v3.95.0 — Animation and sensor audit

  • Twenty-nine exercises were showing an animation that was not theirs. Now none are.
  • Several wearable signals were being misread, producing false accumulated-fatigue warnings. Fixed.

v3.87.0 — Calculated level and level-aware techniques

  • Your training level stops being whatever you ticked on a form: it's computed from relative strength, body composition, rate of progress and consistency.
  • Intensity techniques are offered according to that level, not to everyone alike.

v3.85.0 — Sharper photo and voice logging

  • What the AI detects in a photo or dictation is anchored to the local database instead of being eyeballed.
  • Calories are checked against macros: when they disagree, arithmetic wins.

v3.84.0 — Adapt today

  • Half an hour instead of one? At home instead of the gym? Today's session is rebuilt around what you have — and only today's.

v3.83.0 — Reminders that do not nag

  • Local reminders with quiet hours and no server involved: no connection or account needed.
  • Aggregate scores stop scoring when there are fewer than two reliable signals.

v3.81.0 — Wearables at a glance

  • The watches show status, next workout, heart-rate target and the day's summary without opening your phone.
  • Liquids are logged in millilitres rather than grams, which is how they are labelled.
  • The volume week is anchored to your first training day rather than the calendar Monday.

How updates reach you

Content fixes — a guide, a reference, a food's data — reach you without reinstalling anything. Functional changes do go through the store, like any other app.

Catch up

If it's been a while since you opened it, half the app will be new to you.

Download

Free. No account required. Works offline from the first launch.