Pick the meal, then tap foods to build what you are about to eat. Carbs add up as you go and the card below turns them into an indicative figure — you decide the dose. Save a plate you eat often as a routine and it comes back in one tap.
A correction shot or a 70/30 background dose is logged on its own page, not here.
Everything is editable. Scan a food, add manually, or export your data.
Fibre, sugar and GI are optional. Lookup fills them when it's confident; leave blank otherwise. Fibre feeds net-carb; GI is a band, not a number.
Foods sometimes drift into near-copies ("Bhel puri" / "Bhel Puri" / "Bhelpuri"). Find and merge them. Your past logs reference foods by name, so nothing you've logged is affected.
Add the raw ingredients you use (search works in Hindi too — "shimla mirch", "chola", "oats", "tofu"), or add any ingredient as custom. For each, set the amount either as unit multiples (×unit) or by typing the exact weight/volume in g or ml (1–1000) — the two stay in sync. Set servings, and the app totals it and saves it as your own named dish.
Foods you prepared or added yourself. Correct the numbers in place, or reopen a prepared dish and fix the recipe itself — a wrong ingredient amount is fixed where it was entered, not by back-calculating the totals.
Fill in fibre, sugar and GI band for foods that don't have them yet. New and looked-up foods get these automatically, and any food fills itself the first time you log it — this button just does the rest in one go. Needs your API key.
What the lookups cost you. Every feature that calls the API asks first and shows an estimate. Anthropic bills your key directly in US dollars — this app never charges anything and never calls out without your say-so.
Anthropic publishes its rates in USD, so they are entered in USD: defaults are claude-sonnet-4-6 as at 26 Jul 2026 — $3 / $15 per million tokens and $0.01 per web search. Costs are then shown to you in rupees at the conversion rate above (default 96.5, roughly the rate on 26 Jul 2026). Correct any of these if they drift.
Pick a food and see what it does to your blood glucose: the general indicative rise, then your own measured response — average rise, peak, how long the peak took, and whether any went low afterwards. It counts only the times you ate that food on its own, because a mixed plate cannot be separated. For what the food is made of, use What a food is made of below.
For the food picked above: how it went when you ate it early in a 70/30 window versus late in one, or outside any. Solo meals only, each past its settle window.
Pick a food and see what it is made of: carbohydrate, protein, fat, calories, and fibre and sugar where known, with the share of the calories each contributes. It also carries a one-line summary of your own effect — and unlike the card above, if you have never eaten this food alone it will still give you an estimate from your mixed meals, labelled with how confident it is.
Plain-English background, not about your data: what carbohydrate, protein, fat and fibre each do to blood glucose, and why two foods with the same carb count can behave differently. Read once, not daily.
If your numbers are running high, this looks across today's carbs, cover, background doses and activity to name what's most likely behind it — and if nothing in the data explains it, it says so.
If you were unwell, slept badly, travelled, drank, or started a new medicine, your numbers that day describe the circumstance rather than your settings. Mark it and the app stops learning from that day — while still showing it everywhere else.
The same forward projection the BG tab draws, run from your most recent reading against everything currently on board.
Your recorded 70/30 doses drawn on a 24-hour clock — where their action windows stack, and where cover thins. Built from the real times you took them, not the schedule. 70/30 only — rapid meal doses and corrections are not on this map. For all three together, use the Coverage timeline below.
From your first 70/30 dose of the day to the next one — the range your own recorded nights covered, not a line. Narrows to nights that started near your latest reading when there are enough of them.
The red line is the app's estimate of your BG through the day: meals push it up (sooner for fast carbs, later for fatty/protein-rich food), and both insulins pull it down. The dark dots are the actual BG readings you logged, and the small markers along the bottom show each insulin shot. The estimate re-anchors to your real readings — every time you log a BG, the line snaps to it and models forward from there, so a wrong morning baseline or an unlogged snack doesn't throw off the whole day. Between readings it's only a model, so the further the line is from your last dot, the more it's guessing. Where estimate and CGM disagree, trust your CGM — and the more BGs you log, the tighter the estimate tracks. The dashed amber line is your post-meal target.
Compares your logged BG when insulin was active ("covered") against times little insulin was on board ("gap"). A factual read from your readings — the interpretation is yours and your doctor's.
This shows how much insulin action is on board through the day. There are three bands layered together — the legend below names each. Rapid with meals (solid) and corrections (same colour, dashed edge, because a correction is the same insulin doing the same thing at a different moment), both sitting on your 70/30: rapid (Humalog — kicks in ~20 min, peaks around 1–1.5 h, gone in 6–8 h) sits on top of your 70/30 premix (slower — onset ~1.5 h, a broad peak, lasting 8–12 h). Read it left-to-right: where the bands are tall, you're well covered; where they thin out to nothing between doses is a coverage gap — the times your background may not be holding you, worth checking against your CGM. The height is relative insulin activity, not units.
Dose entry has its own page now — it used to sit at the bottom of this one, behind a sub-tab, which made it hard to find when you just wanted to record a shot.
Every insulin has its own tab: log a dose at the top, its full history underneath, so entering and correcting the same insulin happen in one place. Totals for the whole day are on the last tab.
Tap to confirm each scheduled dose. Skipped or different amounts change your coverage — the timeline uses what you actually took.
Pick a date and it lays that day out against your usual 70/30 times: what you took, then any scheduled slot still unaccounted for — confirm it, change the units, or mark it skipped.
Why this card and not the history below: history holds the doses that happened. This is the only place that can record one that didn't — and a skipped background dose changes your coverage, so the timeline needs to know.
These are your typical doses — the reference you titrate around by CGM (e.g. your midnight 12 dropping to 8 when you're under 100). Log what you actually took below; the timeline uses your logged amount when present, else the reference.
This card only defines your usual times and reference units. To record a dose you actually took: use Doses taken today above for a scheduled slot, Check a day against your schedule for a past day, or 70/30 history below for a dose at any other time — each of those also lets you edit or delete it. Nothing is logged from this card.
Every 70/30 shot that happened, with its BG — scheduled or not. Today's on the face; older days fold into tappable date bars. Edit date, time, units or BG, or delete, and every graph and report follows.
Doses only. To mark a scheduled dose as missed or skipped, use Check a day against your schedule above — an absence is not a record, so it cannot live in this list.
Took an extra rapid shot to bring a high down? Log it here. It feeds insulin-on-board, so your next meal dose won't stack on top of it. (Meal doses are logged with the meal on the Meal tab.)
Every Humalog 100 shot (meal & correction) with its BG. Today's on the face; older days fold into tappable date bars. Edit date, time, units, or BG, or delete — graphs and reports update automatically.
Log a dose of your second background insulin and see what is still on board from the last one. Its history is directly below, so entering and correcting happen in one place.
Every dose of your second background insulin with its BG. Today's on the face; older days fold into tappable date bars. Edit date, time, units or BG, or delete — graphs and reports update automatically.
Every insulin you have logged today, in one place — the per-insulin tabs are for entering and correcting; this is the sum.
Every insulin on one line per day, and the total alongside — which is the figure that says whether your requirement is drifting. Rapid and background separately is useful; only the total answers "am I needing more than I was?".
Add a medicine and the app checks its own reference for the standard duration, shows you what the manufacturer says, and asks you to confirm it. If it does not know the drug it says so and asks you rather than guessing. Change the figure any time your own data suggests a different one.
The app has no way to see which cartridge is loaded or how it has been kept. Logging when you open one gives it a second explanation to offer when a run of highs appears — and unlike the others, this one you can test directly.
Log exercise — it shifts insulin sensitivity and BG for hours. Tap a routine for your usual defaults, or pick from the library, then enter details. You can 📷 scan Garmin screen(s) too — select both the Summary and the Stats screen together and they're merged into one activity. A live watch face only has current HR and steps. Everything's editable before saving.
Bring this to your doctor. Over weeks it shows which meals ran you high or low at your chosen dose and timing.
Breakfast, Lunch and Dinner happen once a day — anything else is a Snack. Where a day holds two of the same, they count as separate meals everywhere in the app, which skews meal matching and the settled analysis. Listed below for you to resolve; nothing changes unless you choose.
What happened on one particular day
What happened, what you ate, what you took, what you did, how it went — and why, compared against your own days. Every comparison says how many of your days sit behind it.
One card a day, newest first: how the day went for your blood glucose, and how the day’s food was spread across breakfast, lunch, snacks and dinner. Tap a card and the full day opens below it — Day review, then your day hour by hour, then forensics. A day the record cannot describe says so rather than being drawn as if it could.
A factual walk-through of one day — each meal with what you ate, insulin taken, and how your BG responded, plus coverage gaps. Observations only; what to change is for you and your doctor.
Everything that happened, in the order it happened — meals, every insulin dose, anything taken for a low, activity — each with what your BG actually did afterwards, in plain words. Where two things overlap it says so rather than blaming one of them.
Takes a single day, cuts it at every dose, meal and session, and asks of each stretch: given the BG it opened at and what was on board, where should it have ended — and where did it actually end? The gap is what the model cannot account for, and that gap is the report. Candidate explanations are drawn only from what is already in your log, each with how you could tell whether it is the one.
Anything you add from this card is marked as recalled later. By default it counts everywhere — totals, charts, this page — but not where the app learns your settings, because an entry remembered in order to explain a gap is not the same evidence as one logged at the time, and the app should not end up calibrating on it.
Puts one day against another and ranks what actually differed — insulin, meals, activity — because anything the same on both days can't explain a gap. When nothing differed, it questions the sensor. Defaults to today against yesterday.
Each day's meals, tap any meal to expand its full detail — carbs, insulin taken, the indicative figure, how your BG moved, and what happened with the spike.
Tick the reports you want, then choose a version. Your phone's print dialog has a Save to PDF option — that's your file.
Every ticked report is drawn for this date range. Leave blank to use each report's own current range.
What eating does to you
Average BG rise (before → 2h, or 1h) after each meal you've logged, worst first. Shows which foods hit you hardest.
Every food you have eaten that took you past the threshold you set, ranked by how often it did it — from your own record, not a food table.
Every meal with enough readings after it is scored for its 4-hour peak. Meals that went past your threshold are then split across what was in them — pro-rata by carbohydrate for ranking, and separately as clean shots where one item carried 70%+ of the meal, which is the part actually worth believing. Each entry shows what was on board and how it was covered.
Defaults to the high alert limit on your Profile.
Meals you've eaten more than once, ranked by their average BG rise. Shows which repeat dishes hit you hardest, using your actual readings around each meal.
Once a meal's ~5h window closes, the app measures what your readings actually did — baseline, peak, how long to peak, where it settled, and any dip afterwards. Retrospective fact, not advice.
What you ate across a stretch of days added up — carbs, protein, fat and calories per day, with the insulin alongside. For seeing whether a week ran heavier than usual, not for judging a single meal.
Builds your typical daily vitamin and mineral intake from the plates you have logged, and sets it against the ICMR-NIN recommendation for someone your age and sex — with your own blood results beside it wherever you have one.
What dosing does to you
Your total daily dose, split by type, averaged over the days you logged. Units-per-kg uses your weight — a figure your doctor can read against typical ranges.
What you actually took, day by day, every insulin. The chart shows the shape; the table under it gives the figures, and each day opens to the individual doses. Whether your intake is climbing, falling or flat — with carbs and average BG beside it, because units alone do not say which.
Your profile carries a window for the background insulin, and that number is a setting rather than a measurement. This times it against your own record: after a dose given overnight, how long before BG came back up — counting only stretches with nothing eaten in them, because otherwise it times your breakfast. Nothing here changes a setting; if the record and the profile disagree, both numbers are shown and the difference is the finding.
How frequently you're taking extra rapid shots for highs — a signal worth watching with your doctor.
Whether your background dose is landing when it should — measured from what your trace does after each one, not from the label on the pen.
For each scheduled 70/30 dose, the average BG in the hours after it, from your readings. A factual look at whether your background doses line up with your numbers — what to change is for you and your doctor.
For every meal you dosed: what the app's arithmetic suggested from your current ratio/CF, what you actually gave, and how your BG moved over the next 3 hours. Descriptive only — it scores nothing and prescribes nothing. Use it to see whether your settings or your own read tracked the outcome better.
Recomputes your past meals under a candidate setting and scores it two ways — against the doses you actually chose, and against your CGM. Read-only; nothing is applied.
Lows, highs, and when they happen
How your logged BG readings sit against your targets. Needs 1h/2h BGs filled in on the Log tab.
How often you dip under 70 (and under 54), what time of day your lows cluster, and how many followed close after a correction shot. The safety mirror to the "when you run high" report — from your own readings.
When your lows cluster across the day, and whether they follow meals or activity. For spotting patterns.
Average readings grouped by meal type, so you can see if breakfasts, dinners, or snacks tend to run higher.
Whether your BG climbs in the pre-waking hours — the dawn rise many people see. From your overnight CGM.
Your spread and stability from all logged readings in the range — coefficient of variation (CV), standard deviation (SD), mean, and a GMI estimate of A1c. CV is the single best "how steady am I" number. Reference figures for watching your own trend, not lab values.
Every reading collapsed onto a single 24-hour axis, showing your median and spread by time of day. Wide bands mean that hour is unpredictable for you; a median above the green strip means you tend to run high then. The clinical gold-standard view — descriptive only.
Average BG for each part of the day (overnight, morning, midday, afternoon, evening), from your logged readings. Shows when you tend to run high or low across the day.
Average BG and in-range share for each weekday, from your readings. Reveals whether routine differences — weekday vs weekend meals, activity, work rhythm — nudge your control on particular days.
Where it is heading, and whether the app is right
Your day-by-day average on a real calendar, with each day’s low-to-high range behind it. Unlike the AGP below, this keeps the dates apart — so you can see a week going wrong, or a change bedding in.
Your time-in-range week by week, so you can see control improving or drifting rather than a single snapshot.
Estimated average glucose (eAG) and the rough A1c it maps to, from all your logged readings in the range. Longer ranges (up to 90 days) mirror what an HbA1c reflects. Spot readings tend to read higher than a lab A1c — use it to watch your trend, not as a lab value.
Your GMI, estimated A1c, variability and time-in-range, each shown against the reference bands the sources use. Your data vs published ranges — never a diagnosis.
Scores the app's own forecast against what actually happened, over every reading that has an anchor an hour before it. Split by what BG was doing at the time, how dense that day was, the hour, and whether insulin was on board — so a weakness shows up as a number rather than an impression.
Scoring every prediction takes a second or two, so it waits to be asked rather than holding up the page.
For each day, how far the estimated-BG line sat from your actual readings (average gap). Lower and steadier over time means the model is tracking you well. Use it to see whether your control and the estimate are converging.
The score above and the learning card are computed over the same stretches: from each thing you logged to the next one. A correction, a dose, some movement or a rescue ends one stretch and starts the next, so nothing is left out for having something else in it. Against each one sits what simply assuming your BG stayed put would have got — that is the bar worth clearing.
Tests candidate corrections to the model against the audit above. Each one is learned on the older half of the window and scored on the newer half it has never seen, so an improvement has to be real rather than the model fitting itself. Your correction shots are one of the training signals — a correction is you judging the app to be reading wrong, which is exactly the label a model needs.
Everything else in the model works out a departure from your last reading — insulin coming down, food going up. Nothing in it knew that a reading of 245 comes down on its own and a reading of 70 comes up, so the forecast could walk downhill indefinitely. This measures the level your own readings have actually drifted back to when nothing was acting on them.
It is a description of where your readings have been, not a level to aim for. It is measured from your record and it is not a target, not a goal and not a recommendation — your targets are the ones you set in Profile and this does not touch them. It changes the forward projection only: no dose, no ratio, no target and no suggestion moves because of it.
Every projection you read is insulin pulling down and food pushing up. On your record those two almost never happen apart, so the app cannot say which of them it is getting wrong — and six runs of measurement have now stopped at that same wall. This counts the occasions that would separate them, and how far off it is. It asks nothing of you.
When you dose a meal, the app also works out what it would have suggested — and what your BG would have done at that number instead. It never shows you that number and it never puts it anywhere near your readings; it files it, with the assumptions it was working under at the time, and waits for your meter to settle the question. This is how it learns whether it is any good without you having to change a thing. Every value it computes is marked as computed. Nothing here is a measurement except your own readings.
Most plates never repeat, so waiting for a plate to come round again is waiting for nothing. The ingredients do repeat. This is what the app can work out about a meal from its parts alone — which items it already knows, which it has watched enough times to have measured on your record, and which still need you to tell it. The exact plate only overrides this when you have actually eaten it before.
The app measures its own lean against your readings and re-measures as more arrive, so the correction it applies keeps up with your data instead of freezing at the first answer. It proposes; you decide. Every figure here comes from this device alone — a second phone learns its own, from its own body.
Every insulin's action window is currently whatever the manufacturer's label says, because that is a sensible place to start. It is not the last word: the same insulin runs shorter in one person and longer in another, and yours is the only record that knows which. This re-runs your own readings against each candidate window and says which one fits how your BG really moved.
The app looks across your settled meals and clean correction windows for a consistent one-way drift against your current settings. If it finds one, it proposes a change with the evidence — and applies nothing unless you tap Apply.
What activity does to you
Minutes per day across the range, with rest days marked. The effect of activity on your numbers is in the Activity → next-day BG report further down.
Compares your in-range % and average BG on the day after an active day vs after a rest day. A rough signal of whether exercise is helping your control — not a controlled study, just your own pattern.
Goes past "cardio / strength / yoga" and asks what actually drives the effect: how hard, how long, what time of day, and whether it carries overnight. Coverage comes first. Every split below states how many sessions it could use and whether that is enough to say anything — a number computed from four sessions is not a finding, and this card says so instead of drawing it.
Every build of this app, newest first
What each version changed, taken from the build log inside this file — so it cannot drift from what actually shipped. The build you are running is marked.
What your watch records while you are not looking
Whatever your Health export carries, kept as its own series. Paste a HEALTH HR block into the box on the Move screen — it can be heart rate on its own, with no workout in it at all.
Hours when your heart rate sat outside the middle 80% of what you have recorded for that hour, with what your blood glucose was doing stated beside it. It does not say why.
Each night as your watch recorded it — core, deep and REM, with time in bed beside time asleep.
Health’s own resting figure, and the floor your own samples sat on. They are measured differently and are shown apart.
Whatever your export carries, kept as its own series and compared against nothing until there is enough of it.
How long your heart rate takes to return to your usual for that hour after a session that pushed it above it.
What your heart rate did when your glucose was low, measured against your own usual for that same hour. It appears only when there is enough of it to mean anything.
Who you are and the numbers every other card calculates from — your carb ratio, correction factor, targets, insulin names and windows. Change something here and every projection, report and indicative figure in the app moves with it.
These go at the top of every report you print or share, and into the backup file, so a doctor receiving it knows whose numbers they are. Fill in as much or as little as you want — blank fields are simply left out.
Once these are filled in, the backup file identifies you by name. That is the point — but it also means the file is no longer anonymous, so send it deliberately.
Update your weight any time it changes — the insulin-per-kg report uses it.
Published onset, peak and duration for the insulins you actually take, so the app’s windows can always be read against the label rather than remembered. These are population figures from healthy-volunteer studies — your own response is the one that counts, which is the whole point of this app.
A credential, not a food setting — it lives here with the rest of your account details (v4.90.0). Photo scanning and food lookups need internet and this key, which is separate from your Claude subscription: get one at console.anthropic.com. Stored only on this phone and never sent anywhere except Anthropic.
Since v5.35.0 this key is not written into your backup. It stays on this phone; a backup file you send to somebody no longer carries it. After restoring onto a new phone, paste it in here again.
Each meaningful rise and fall read off your minute-by-minute trace — when it started, how fast, how far, how long it lasted — with whatever the record holds around it: the meal, the dose, the movement, the low you treated. Where the record does not explain one, the app asks you instead of calling it unaccounted for. Your answer is stored and that rise is never counted as unexplained again.
A correction given and BG kept climbing; a meal dose followed by a rise far bigger than the carbohydrate accounts for; corrections stacked in one stretch with no fall after them; a background dose that stopped holding early. For each: what was given, what this app’s model expected, what actually happened, and how big the gap was against the spread the model normally runs out by. A description of what the app noticed — never a judgement on the dose, and never a suggestion of a different one.
Pick a kind, then pick either two records of that kind or one record against everything of its kind that came before it. Each side shows what it was — the inputs of the thing itself — and what followed, over a window worked out from your own record and printed on the card. The last block names the differences the app can see. It names them; it does not rank them, and it never says one of them caused anything.
Everything you logged with a time, newest first. Tap one and it opens on its own page: the event, what was taken, what the app showed beside it, and what your readings did 1, 2 and 4 hours later, with the curve between. Each page describes that one event and stops there. It does not compare events, rank them, or say what to do.
Where the figures on this record sit against published standards, each with the body that publishes it named; which way they have moved over the dates on the record; and which two measures have moved together, with how strongly. Every figure carries how many observations sit behind it and over what dates.
Every setting you save in Profile is stamped with the date. Once there is enough data on both sides, this compares the window before against the window after — average BG, in-range, above target, below 70, and how far the 1h prediction missed. It will often say it cannot separate the two, which is the honest answer when your day-to-day spread is wider than the change.
From the exercise you log (and Garmin scans): total sessions, active minutes, calories burned, and a breakdown by activity type. Movement shifts insulin sensitivity, so this sits alongside your BG picture.
Everything on this phone, in one file. Nothing here is sent anywhere — the file is written by this phone and stays wherever you put it.
Moved back here in v3.5.1. The v2.7.0 tab split cut the Food database out of Profile and took this block with it — it had been sitting at the end of that card — so it ended up on the Food database page and was never even drawn, because the render calls still pointed at Profile.
Your data lives on this phone only. Export a backup file now and then. To restore it — new phone, cleared data, or Android later — import that file and everything comes back.
Opens the share sheet with the file attached — Files, iCloud Drive, Drive, anywhere. A copy that is still inside this app is not a backup: removing the app takes it too. Only a file that was actually saved is counted.
That one replaces what is on this phone — right for a new phone or after clearing data, but anything logged since the file was exported is lost. To add old lab reports to what you already have, use the additive option below.
Automatic on-device copies, kept for the last few days. For undoing a bad import or a mistaken edit — taking one costs nothing and asks nothing.
Takes a full export or just a list of markers. Adds only, skips anything already on file, never overwrites a record, and leaves meals, doses, foods and settings alone — so it is safe to run against live data.
The app keeps a second copy of your data inside itself, refreshed hourly. It can undo a bad import or corruption — but it is deleted along with the app, so it is no substitute for the export file above.
Four routes in. They were on three different screens until v5.22.0, and one of them was inside an activity form — which is why the heart-rate import could not be found without reading the code.
Upload your weekly CGM PDF and the app pulls in every hour's high and low reading — far richer than typing spot values. Import daily; overlapping days simply refresh. Where the CGM covers a time, its value is used; your manual readings at other times are kept.
Open the PDF, select all, copy, and paste here — then tap Parse. Use this if you're offline or the PDF won't read.
Your CGM already writes to Health, so the readings are on the phone — but no web app can read Health directly. There is no browser API for it and there will not be one; Apple opens Health only to native apps. What can be done is have Shortcuts fetch them for you, which gets this down to one tap.
If an earlier import landed at the wrong times, paste the same text above and use this. It works the shift out itself by matching values, removes only readings matching on both time and value, and re-imports at the corrected times.
Imports made before this app knew the difference left the report's hourly high and low sitting alongside a full day of real readings. Averaged together they count every hour's extremes twice. This marks those hourly points as superseded on days that already hold a dense trace of their own — and only those days. A day whose only data is the hourly points is left completely alone. Nothing is deleted and the whole run can be put back.
Paste a HEALTH HR block here — heart rate samples with their own times, with no workout in them. Your Garmin shortcut only runs on days you train; this is for every other day.
This box only ever stores heart rate. It cannot create an activity, it does not touch your glucose readings, and nothing you tapped above changes what it does. One day at a time is fine — pasting the same day again adds nothing and takes nothing away.
For more than a day at a time, or for anything other than heart rate, use 📥 Import a health file below — a clipboard paste tops out around a thousand samples.
The .zip Health Auto Export writes, or a single .csv or .txt. Saved to On My iPhone in Files.
This route takes everything the file carries, not only heart rate: blood glucose to your readings, heart rate to the heart-rate series, and steps, distance and flights climbed kept alongside them. Units are read from the file's own headers every time and named on the result — the same day exported twice has come out in mmol/L and mg/dL.
Glucose already on file for a minute is left exactly as it is — this only fills gaps. It creates no activity.
Read on your phone. Nothing leaves the device.
Garmin Connect → Activities → Export CSV. It carries every session with its distance, heart rate, cadence, pace and step count — far more than a screenshot can, and it is Garmin's own summary rather than a reading of a photograph. Sessions you already have are filled in and corrected rather than duplicated, and importing the same file twice changes nothing. A .zip is read directly.
The phone has no CSV export. Open the activity in Garmin Connect, choose Export File, and save the .zip to Files; choose that zip here, or a bare .fit. It carries what the activity list does not — time in each heart-rate zone, run, walk and standing time and distance to the centimetre — and it has no moving time, which the list does. A session you already have, from either route, is filled in and never duplicated, and importing it again changes nothing.
Before v5.23.1 a single CSV could have its glucose column read as heart rate. If that happened, choose the same file here: every heart-rate row whose value matches this file’s glucose at the same minute is removed, and nothing else is touched.
It matches on the exact minute AND the exact value, so genuine samples on the same day survive. Then import the file normally above.
This one still lives inside the activity form, because what it produces is an activity and it needs the type you are logging. Tap a routine on the Move screen, then look for From a Garmin screenshot.
Carb ratio, correction factor, how long each insulin lasts, and your targets. These were called Titration settings and sat inside Profile; the name did not say what they were and Profile is where identity lives, not arithmetic.
These are your values, the ones you already use. Editable any time — you flagged the ratio drifts.
Everything on this card saves itself the moment you leave a field — no button to find.
The pre-meal target is what a meal-time correction is measured against — if your BG is above it when you eat, the correction part is positive; below it, negative. The post-meal figure is the ceiling used in the retrospective reports, not in the meal maths.
A banner appears at the top of every page when a reading, or the model's forward track, crosses one of these. It is a warning to go and look at your CGM — nothing more. Set them where you want to be told.
A sabzi and a paratha are not cooked in the same thing. Set each one — the plate picks whichever fits the dish, and you can still change it per dish.
Short window drives insulin-on-board. Both editable since you said they drift.
The estimated-BG line uses a carb-sensitivity value (mg/dL rise per gram of carb). It starts from your correction ÷ ratio and is checked automatically every week against your actual readings — if it drifts, you'll get a banner on the Coverage tab asking permission before anything changes. You can also check now.
Both are measured from your own data in Reports → Self-check & proposals, where the evidence and the uncertainty are shown before you apply anything. They shape the forward projection only — never a dose. A negative activity number means that activity pulls BG down.
This only affects the on-screen estimate. It never changes your ratio, correction factor, or any dosing figure — those stay exactly as you set them.
Name your two insulins. If you switch to a different insulin in future, change the name here — past shots keep the name they were logged with, only new shots use the new name, so your history stays accurate.
Pick from the list and the app fills in the name, the type and the modelling window from the manufacturer’s figures — or type your own underneath if yours is not listed.
The app's projection was measured against your own readings and found to sit low — it under-predicts. A single constant, learned from your older data and tested on newer data it had never seen, closes most of that gap. This is where the figure lives; the Calibration lab (Trends & accuracy) is where it is measured and where you adopt one.
Changing this changes the number you read before dosing. It is arithmetic from your own record, not advice — set it from the lab rather than by hand unless you have a reason.
If you take a separate flat basal alongside the two above — a Lantus, Toujeo, Levemir or Tresiba — switch it on here and it gets its own dose stream, its own action window and its own place in insulin-on-board. Leave it off and nothing about the app changes.
Switching this on is dose-adjacent: those doses start pulling the projected track down, so the figure you read before dosing will be lower. That is arithmetic from your own settings, not advice — and worth raising with your doctor before you rely on it.
The numbers you or your doctor/dietitian work with — the daily summary compares your actual intake against these. Leave blank and it just summarizes without comparing.
Tap what you took and put in how much — the carbohydrate is worked out for you. BG and details can be filled in afterwards. Recording it is what stops the rebound being read as a meal.
For anything else, open Something else below and type its name — the app works the carbohydrate out from your food list, and looks it up if it has never heard of it. You should not have to know the grams while you are low.
This card records what you took. It does not and will not say how much to take — that is a dose, and it belongs with your doctor. If lows are happening often, the Low-glucose timing card and the rescue log are what to show them.
When each sensor went on, and when it came off — finished if it ran its course, removed if it came off early. Those are different things and the app keeps them apart. Earlier sensors are worth adding too: type over the date and time and a sensor from July goes in exactly like today's. Nothing here is calculated from a typed sensor lifetime; how long one lasts is taken from the ones you record.
Log a finger-stick or spot CGM reading at any time — not tied to a meal or shot. Use this to fill gaps the CGM import doesn't cover. These feed the estimated-BG graph and your reports.
Ask about insulins, diabetes, heart, kidney, or any health topic. Answers are fetched with web search and every claim is cited with its source and date — this is reference material to verify, never medical advice. The app is a librarian, not a doctor.
Type your reading, or scan the monitor. Systolic is the top (higher) number, diastolic the bottom. Pulse is optional. Everything stays on this phone.
Recent readings with a systolic/diastolic trend line. Category labels follow standard adult reference bands — descriptive context to watch your own pattern, not a diagnosis and never medication advice. Discuss any change with your doctor.
Upload a photo or PDF of a lab report. It reads the values and explains each one in plain English — what it measures, and how it sits against the standard reference range printed for that test. Key markers (like HbA1c) are saved so you can watch the trend over time.
Everything is processed for this one explanation and shown back to you — the app keeps only the marker values you choose to save. This is reference reading to help you understand your own report and talk to your doctor. It is not a diagnosis and never a treatment or dose change.
Values you've saved from scanned reports, newest first. Editable reference figures for watching your own trend — confirm anything against the actual lab report and your doctor.
Four questions a long record can answer and a short one cannot. Every figure is measured from your own data, and each says how sure it is.
For each meal that stood alone, the ratio it would have taken to land back where it started. Insulin sensitivity is not the same at 7 am as at 8 pm for most people; this asks whether it is for you.
Your correction factor is a number you set. This is what corrections actually did — only shots with nothing else going on, banded by where they started and by what was already on board.
Every evening dose against the lowest reading between 03:00 and 08:00 the next day. If your evening dose barely varies there is nothing to learn here, and the card will say so rather than fitting a line through nothing.
Days ranked by time in range, best third against worst third, comparing every daily figure you log — carbs, late carbs, insulin, corrections, movement, last meal, water.
Tap a glass or bottle to log what you drink. Dehydration concentrates blood glucose and makes readings look higher than your control actually is, so a thin day here is worth knowing about when a report looks off.
Plain water only. Tea, milk, chaas and sharbats belong on the plate, because they carry carbohydrate and this does not.
A few plain questions. Nothing is chosen for you and nothing is worked out from what is already on here — an empty record means nothing has been collected yet, which is what this is asking about.
How large and how dark the text on every card is right now, where those figures came from, and what the app measured on this device to arrive at them. Nothing here was chosen for one person's eyes: the size follows the text-size setting on the phone, and the colours are held to the published contrast minimums. Change the text size in the phone's own settings and this card, and the whole app, follows it.
Every report card and everything the app can collect, each with its own switch. Off means it stops being collected — not merely hidden. Switching anything off never deletes what is already stored; it stays in your record and in your backup.
What a watch writes into Health Connect differs by watch, by app version and by which permissions you granted — sleep stages especially. So this reads your own export and reports what arrived, rather than assuming a shape.
Every call the app has made, what it was for and what it actually cost — by day, by feature, down to the individual call. Recorded from each response, not estimated.
When a number, a card or an import looks wrong, put it here the moment you notice. It is kept with your data and travels in the backup, so the next time you hand a backup over the list comes with it and nothing depends on you remembering a week later. Yours alone — nothing is sent anywhere.
Alongside the fault log, and deliberately not inside it: a fault is the app doing the wrong thing, a request is the app not doing a thing at all, and mixing the two makes both lists harder to read. Write what you want and why you want it — the why is the half that decides how it gets built, and it is the half that is gone a week later. Stamped with today's date and the build you are on, kept with your data, and it travels in your backup exactly as the fault log does. It sends nothing anywhere.
Compares the copy running on this phone against the copy published on the web. Two things can leave you on an old build now: the ordinary browser cache, and the offline cache the app keeps so it opens without a signal. Neither announces itself. This is where it shows — it always reads the published file directly, never either cache, so it cannot agree with itself.
The check re-downloads this file (~0.9 MB) to read what is actually published. It runs by itself the first time only; after that, tap.
Theme, font and text size. Pick what's easiest for you to read — especially at 3am.