To be clear about status before anything else: none of what follows has shipped. GLP Fish 1.0 is the version in App Store review, and everything below is written and tested in the working repository, headed for the next update. We would rather show our hand early than have people guess — and if something below moves or slips, this post will say so when it does. The companion post covers what 1.0 already does.
Three ways for a weight to get in
1.0 takes typed weigh-ins. 1.1 adds two more, plus an import, and each one is built around a privacy decision we made first:
Apple Health import — read-only
If you already weigh yourself with a scale that writes to Apple Health, you can import those readings into GLP Fish. The import is strictly read-only, behind the consent screen iOS gives you, and it keeps a marker so the same reading is not imported twice. The app never writes to Apple Health — not weight, not anything else. Your records in GLP Fish are yours, and the health store is not a place we publish to.
Scale photos, read on the device
You can photograph your scale's display — or pick such a photo from your library — and the app reads the number for you, using the photo's capture time to prefill the reading's time. The reading happens entirely on your iPhone; the photo is never uploaded, and only the weight you confirm is saved.
Bluetooth scales
With your permission, the app can connect to a compatible Bluetooth scale and take the reading directly, so a weigh-in is step on, tap, done. Scanning runs only while the connection screen is open, and Bluetooth is never used for anything else — no location, no tracking. The protocol specs were written from public documentation as a clean-room implementation, which is the slow way and the right way.
Meal photo estimates, and the free tier stays free
1.1 lets you photograph a meal and get a calorie and protein estimate you can edit before saving. This is the one place where a photo leaves your device: it is sent over HTTPS to our server and passed to the AI provider named in the privacy policy (currently Anthropic) to produce the estimate, and the photo is not kept after the estimate comes back or used to train anyone's models. The estimate arrives as an editable draft with its source and confidence noted on the entry — you confirm what goes into your record, same as every other entry.
Photo scanning is metered through a small in-app purchase of scan packs, because each estimate costs us money; typing a meal, logging a dose and weighing in stay free, as they are in 1.0.
One prediction instead of several exhibits
1.0 shows trial anchors, comparisons and your recorded trend, and leaves the synthesis to you. That is honest but heavy going. 1.1 adds one prediction card: an expected weight change at a year, a range around it, and the share of people who reached at least a given level — one number to hold, with the rest still underneath.
The anchor for it is real-world data, not trial data: a published study of 7,881 adults who started injectable semaglutide or tirzepatide in 2021–2023 (Gasoyan and colleagues, Obesity, 2025), in which one-year averages ran −7.7% on semaglutide and −12.4% on tirzepatide, with −11.9% for those still on treatment. Trials run on dedicated staff and high adherence; real life is where the readers of this app live, and about half the people in that dataset had stopped within a year. The card states whose distribution it is describing, and the trial curves stay visible beside it rather than being blended into it.
Two adjustments are built in. Your starting weight matters — the same drug takes a smaller percentage off a heavier body, per the subgroup analysis in the regulators' own reviews — and your own records can move the prediction toward how you are actually doing. Only slowly, and only so far: a few good weeks are weighted lightly against a year of published data, and the range can narrow to half at most, never to a point.
A correction that learns your rhythm
1.0 corrects each weigh-in against the morning fasted baseline using rough population averages, with the uncertainty shown. 1.1 makes it personal: weigh in once in the morning and once later on the same day a few times, and the app starts estimating your daily swing and blending it into the correction, with the confidence range narrowing as it learns. Bad pairs are thrown out — a delta outside what daily variation can plausibly do is a mistyped digit, not a rhythm — and the range never narrows to a point. A comparison also stops pretending a Tuesday evening reading and a Monday morning one were taken the same way: it compares against the same time of day, and says what was entered and what it was compared against.
The community grows a discussion
The Peers tab in 1.0 is posts: publish, like, comment, report, delete your own. 1.1 gives a post its discussion on the card, lets a reply be agreed with, carries posts' language and region with filtering, and opens the feed on your own region. Moderation grows with it: every comment becomes reportable like a post, blocking an author also notifies us, and the blocked list is a page you can undo from. None of that is glamorous, and all of it is the part a health-adjacent community cannot skip.
And one thing that is just nice
When a record lands, 1.1 gives it a small ceremony: your figure steps onto the scale, the number rolls to its place, and a weight or body-fat loss gets fireworks. A gain gets a neutral landing and copy that does not scold — the scale is a measurement, not a verdict. It respects Reduce Motion, and it plays on every entry path, typed or imported. Logging a weight is the act this whole app is built around; it deserves better than a spinner.
What is deliberately not on this list
No dose advice beyond the official labels, no writing to Apple Health, no prediction that hardens into a promise after a few weeks of data, and no treating a meal photo as a clinical record — it is an editable estimate. The boundaries in 1.0 are the boundaries in 1.1.
GLP Fish is listed on the App Store. Questions or feedback? Support is the fastest way to reach us.