Questions, and straight answers

The download asks for a username and password.

That's the gate — use the ones whoever invited you sent. If you don't have them, ask them.

macOS won't let me open the app.

Open System Settings → Privacy & Security, scroll to Security, and click Open Anyway next to HealthOS. On macOS 14 and earlier, right-click HealthOS → Open → Open. That works for almost everyone, and it's the only step you should normally need.

If it still refuses, first check the app is actually in your Applications folder — none of this works while it's still sitting in the window the download opened. If it is, and macOS is still blocking it, there's one last resort that involves typing a line into Terminal. You do not need to understand Terminal to do this, and you shouldn't need it at all — if you'd rather not, just ask whoever invited you instead.

Press Command-Space, type Terminal, and press Return. A window with text in it opens. Copy this line in and press Return: xattr -dr com.apple.quarantine /Applications/HealthOS.app. All it does is remove the "downloaded from the internet" label macOS attached to the app, which is what's making it suspicious. Then close Terminal and open HealthOS again.

After I allowed HealthOS, macOS asked about another file, like rollup.darwin-arm64.node.

That came from an older version of HealthOS, and you don't need to allow the file. Download the latest version, drag it into Applications to replace the old one, restart your Mac, and open HealthOS again.

macOS asks whether HealthOS can access a folder.

That's the data folder you picked, usually because it's in Documents or on an external drive. Allow it; HealthOS can't read or save your data otherwise. If it asks again after an update, allow it again.

It's taking a long time to start.

The first start sets things up and usually takes about a minute, a couple on a slow or external drive. After that it opens in seconds. If it gives up waiting you get a small dialog with a Retry button.

Is this for Windows or Intel Macs?

This build is for Apple Silicon Macs (M1-M4) on macOS 12 or later. If that's not you, tell whoever invited you.

I added a file and nothing showed up.

Check the file is in the folder its Reveal folder button opens, and named the way the setup steps say: a file in the wrong folder, or with the wrong name, is quietly skipped. Then click Quit and open HealthOS again so it rescans. System → Ingestion & Jobs shows each file it picked up, and says why if one failed.

My lab results are a CSV, an Excel file or a PDF.

A PDF, or photos of a printout, can go in through Body → Trends → Add lab results: Claude reads it and you check every row against the page before anything is saved, so it needs an AI key. CSV and Excel files aren't read at all. The only no-key route is the plain-text _longform.tsv table described in the setup steps.

How do I update, and how do I know what changed?

While the app is open it checks for a new version every hour, and a notice appears across the top with the download link. Download it, drag it into Applications to replace the old one, and open it again; your data folder is kept. Settings → Version history lists what changed in each version and which one you're running.

Where is my data, and how do I back it up?

All of it is in the one folder you picked at first run. Nothing is backed up for you: to keep a copy, click Quit and copy that whole folder somewhere safe. It's an ordinary folder — if you delete it, HealthOS forgets everything. Remember the folder also holds your AI key, if you set one.

Can I add a family member?

Yes, as a separate profile: open the menu beside the HealthOS name at the top left and choose Add a profile, then give a name and pick a folder for that person's data. Each person gets their own folder and their own copy of the app's engine running, and you switch between them in a click. Don't load someone else's labs or DNA into your own profile: relatives share genetic variants, and one person's results would overwrite the other's.

A genetic finding says "not yet confirmed" or "not detected". Does that mean I don't carry it?

No. A genome file lists only where you differ from the reference, so a variant missing from the file and a spot the sequencing didn't cover look exactly the same. HealthOS says "not detected" rather than "you don't carry it" for that reason.

The Research tab finds nothing, or says a source is unavailable.

Update to the latest version first: older versions couldn't search at all on a new install, and reported that as "no new papers". Now, on a friend's Mac, searches reach ClinicalTrials.gov, Europe PMC, PubMed and OpenAlex. arXiv needs extra tools the installer doesn't include, so the tab lists it as unavailable rather than pretending it found nothing.

Can HealthOS tell me what's wrong with me?

No. HealthOS reads your own data and shows you what is in it — never a diagnosis, never a prescription, and never an all-clear. What you do with it is between you and your clinician. It can't order tests or contact anyone; a suggestion to recheck something is a note for you to take to whoever orders your tests.

Do I need an AI key?

No. Without one your files import, the feed fills in, and every screen opens. What a key adds is the parts that read or write in sentences: the chat, reading scans and documents you upload, the supplement timetable, and paper summaries.

My Anthropic key is rejected, or every AI feature fails.

HealthOS checks a key with Anthropic before saving it, so a mistyped key is turned away at once. If the key is right but every call still fails, it was probably created with an organization-wide scope: create a new key at console.anthropic.com with the Default scope and paste that one instead.

What does it cost?

Usage bills to your own Anthropic account at their prices. HealthOS shows estimates so it's never a surprise: each fresh chat answer shows roughly what it cost, System → Ops / Eval → Model spend adds everything up by day, model and feature, and Settings → Imaging read depth shows a cost range before you choose. For a large scan of about 1,200 slices, the app estimates around $0.43 to screen it and about $2.41 to $3.17 at the default depth. These are estimates at list prices, not your bill.

Which AI model does it use?

Claude, from Anthropic. A sensible default is set; Settings → Model lets you pick a different Claude model for the chat, with a one-line note on each, and the model that wrote an answer is named under it.

With no AI key

Your files and records stay on your Mac. HealthOS keeps everything in the folder you chose, and all the ordinary parts still work — your files still import, the feed still fills in, and you can still upload and view scans. You just don't get the AI reads. A few small things still reach the internet, and the next answer lists every one.

What reaches the internet even with no key

On its own, only one thing: the app checks this site for a version number so it can tell you an update exists. That carries nothing about you.

Three buttons reach out when you press them, all of them to public research databases. A search on the Research tab sends the words you typed to four of them: ClinicalTrials.gov, Europe PMC, PubMed and OpenAlex. Pull research on the Peptides tab sends that peptide's name to the same four, and to ChEMBL. And Find related studies on a background note sends those four a search built from the note's research focus and the names of any of your lab tests that are out of range (for example "ferritin") — never their values, and never the note itself. That last one does say something about you, so only press it if you're comfortable with that.

The optional deep genome read downloads large public reference files the first time you set it up. That's a download only: the analysis itself runs on your Mac and sends nothing.

Send to the developer, on a comment you wrote with the Comment button, sends that one comment to the person who invited you, through this site. Before it goes you see exactly what will be sent: your text, which screen you were on, and, if you pointed at part of the screen, a description of it including the words it showed. Those words can be your own values (for example a lab result), so there is a one-click switch to leave them out. It is signed with your download username, and it stays on this site until the developer removes it. Comments you only save never leave your Mac.

With your own Anthropic key

More goes out, all of it to Claude and all of it billed to you. Three of these happen on their own:

Uploading a scan sends the image to be read. For a DICOM series it sends picture versions of the slices rather than the original files: every slice at reduced size for the quick screen, then the flagged ones at full size. Uploading a radiology report PDF or a lab report sends the whole document to be read. Both happen as part of the upload rather than as a separate step you press; while a key is set there is no per-upload switch. And whenever a new finding appears, that finding's details (for example the lab value and its range, or the genetic variant) are sent to write the short title on its card — one finding at a time, never your whole record.

The rest happen when you use them. Ask sends your question plus whatever parts of your data it looks up to answer it, including your background notes if it reads them, and separately sends the opening of each new conversation to come up with a title for it. Partway through an answer it may also search the four public research databases above, with search words Claude writes from your question. Building a supplement timetable sends the list of what you take, plus timing notes worked out from your labs and genetics — not your raw lab numbers and not your genome file. Pulling a research paper sends that paper's text to be summarised and rated for relevance, automatically. The Run checks button on the Ops screen runs a batch of test questions through Claude. Saving a key sends it to Anthropic once to check that it works.

With a Voyage key as well

The titles and abstracts of research papers you pull go to Voyage AI so they can be indexed — that's public text, not yours. Ask can then run a meaning-based search over them while answering, which sends its search wording (composed from your question) to Voyage too. Your own findings, notes and files are not indexed or sent there.

Can the person who invited me see my data?

No, apart from comments you choose to send them. There's no account and no server holding your data. The username and password unlock the download and sign the comments you send with Send to the developer; each one is shown to you in full before it goes. Otherwise the only thing the app asks this site for is the version number, which carries nothing about you.

Where is my AI key kept?

In a hidden settings file inside your data folder, so it travels with the folder if you copy or share it. It's only ever sent to the company it belongs to. If you set up sending comments, your download username and password are kept in the same file, and only ever sent to this site.

Turning it off

Your genome, lab and wearable files are never uploaded in bulk. If you'd rather nothing was sent at all, leave the keys blank. If you've already set one and want it gone, use the Remove button next to it in Settings — and whether or not you do, you can revoke the key itself at console.anthropic.com, which kills it immediately and everywhere, including in a copy of HealthOS you've stopped using.