Your health data, read back to you.

1. Double-click the downloaded HealthOS-arm64.dmg to open it. A window appears with the HealthOS app in it. Drag the app onto Applications — that's in the left-hand column of any Finder window, under Favourites. That window doesn't have an Applications shortcut of its own, so you do have to drag it across. Keep it in Applications so the rest of these steps match.
2. The first time, macOS will say it can't verify HealthOS is free of malware. That's expected. Click Done, then open System Settings → Privacy & Security, scroll down to Security, and click Open Anyway next to HealthOS. Confirm with Touch ID or your password. On older Macs (macOS 14 and earlier) you instead right-click HealthOS → Open, then Open again. Either way you only do this once; after that a normal double-click works.

HealthOS asks where to store your data — pick a folder (a spot in your Documents, or an external drive, both work; if macOS asks whether HealthOS may use that folder, allow it). Your health data itself lives only in that folder — and so does any key you paste in later, in a hidden settings file inside it. (Worth knowing if you ever copy or share that folder: the key travels with it.) Two small housekeeping files sit outside it, in your Mac's own Library folder: a note of which folder you picked, and a log of what the app did. Neither holds your health records.

It only asks once, and you can move it later with Change data folder in the app window; the old folder is left alone.

It sets up — usually about a minute, sometimes a couple on a slow or external drive — then opens your browser automatically to a short setup wizard. Just follow it; you don't need to come back to the app window. If it ever gives up waiting, you get a small dialog with a Retry button rather than sitting there frozen.

Setup wizard · Welcome

The easiest first input is your Apple Health export:

1. On your iPhone: open Health, tap your photo (top right), scroll down, tap Export All Health Data. You get a file called export.zip.
2. Get it onto your Mac (AirDrop, or email it to yourself) and unzip it.
3. In the wizard's Add your data step, click Reveal folder and drop the unzipped contents into the folder it opens.
4. Continue through the AI features step to First scan — it reads through everything you added and your feed fills in.

Setup wizard · Add your data

If you have other kinds of data, you can add those too. Each one has its own Reveal folder button, and all of it is optional — add it now or years from now.

Day-to-day numbers from your watch and phone can keep flowing in automatically if you use the iPhone app Health Auto Export. Its button behaves a little differently from the others: it opens the folder in iCloud Drive that the iPhone app writes into, rather than a folder inside your data folder, because HealthOS picks the files up from there. HealthOS doesn't look in iCloud Drive until you click that button; from then on it checks for new exports while it's running.

Blood test results go in one of two ways.

• From the lab's report, once the app is open: Body → Trends → Add lab results, then pick the report PDF or photos of the printed results (PDF, PNG or JPEG, up to 30 MB; an iPhone's HEIC photos need exporting as JPEG first). Claude reads it, which needs the key from Step 4. Then you check each row against the document shown beside it and approve. Nothing is saved until you do, and a value read off a page is only as right as your check of it.
• As a plain text table, with no key needed: one line for each thing that was measured (each "analyte", meaning one substance the lab looked at, like cholesterol or vitamin D) on each date it was measured. The file is a .tsv whose name ends in _longform.tsv, dropped into the labs folder. A CSV or Excel file isn't read; if your results are in some other shape, ask whoever invited you rather than guessing.

A report PDF dropped straight into the labs folder is listed under Trends but isn't read until you click Read on it, because reading it means sending it to Claude.

DNA works, in two stages. Drop in a whole-genome file (a .vcf.gz, aligned to GRCh37/b37 — a 23andMe or array chip won't do, it miscalls) and the basic findings show up straight away. The deep read — screening your whole genome for disease-linked variants and working out which medications your genes affect — is an optional advanced step under Settings → Genome (advanced). It needs Docker and a one-time reference download of about 27 GB, so it's a sit-down job rather than a five-minute one. The panel walks you through it, and there's a separate step-by-step guide. Skip it if you just want the rest of HealthOS — everything else works without it.

Scans work differently. MRI, CT, X-ray, ultrasound and some eye scans aren't dropped into a folder — you upload them inside the app, on Body → Imaging. It takes PNG, JPEG, HEIC, PDF, or DICOM — that's the format hospitals and imaging centres give you on a disc or download, either as a single .dcm file or a .zip holding a whole series of slices. Size limits are 30 MB for an image or PDF, and 1 GB for a DICOM file or zip. You pick the body part and side each time; both are required.

If you've added an AI key (Step 4), an upload also gets an AI second read — a note from Claude pointing out things it thinks are worth a closer look, on top of the scan itself. For a DICOM series, every slice is first screened at reduced size by a fast model, and only the slices it flags get a closer read; the study shows both counts. Settings → Imaging read depth sets how many flagged slices get that closer read, and shows a cost estimate for each option before you pick one. Three things quietly switch the read off:

• Leaving the type on Auto-detect when you upload a plain image. Pick the real one (MRI, CT or X-ray) instead.
• Choosing Other for the body region. The scan gets filed, but nothing reads it. Pick a real body part.
• Uploading a HEIC photo — the kind an iPhone takes by default. HealthOS stores it but never looks inside it. If you want a read, open the photo and export it as JPEG or PNG first.

The daily feed works on its own, with no key. Claude is what powers the parts that need to read or write in words, and each of those bills to your own key. The main ones are Ask (chat with your data), the second read on a scan you upload, reading a lab report you upload, transcribing a radiology report PDF, and building a supplement timetable. There are smaller ones too: summarising research papers you pull and rating how relevant each one is, giving each new Ask conversation a title, and the Run checks button on the Ops screen. That's the shape of it rather than an exhaustive list — assume anything that reads or writes in sentences is going through your key.

Worth knowing: several of these run on their own rather than on a button you press. Uploading a scan, a radiology report PDF or a lab report starts the read straight away, and pulling a research paper summarises and rates it straight away. The short title on each feed card is written by Claude too, whenever a new finding appears. While a key is set there is no per-upload switch — every scan and report you upload is read.

To see what it's costing you: each fresh chat answer shows its own estimated cost, and System → Ops / Eval → Model spend adds it all up by day, model and feature. These are estimates at list prices, not your Anthropic bill.

Settings also has an optional Voyage AI key, for search that matches on meaning rather than exact words. With it set, the app also sends the titles and abstracts of research papers you pull — public text, not yours — to Voyage so they can be searched that way, and Ask may run one of those searches partway through answering you. Your own findings, notes and files are not sent to Voyage. Skip the key if you'd rather not involve a second company at all.

Want them? Get a key from console.anthropic.com and paste it into the wizard's AI features step. Don't want them? Skip it — everything else still works. You can add a key later in Settings.