There's a specific kind of dread that comes with opening a lab result you don't understand. You scan the numbers. Some are flagged in red. You Google one of them, land on a forum where someone's describing symptoms that sound nothing like yours, then click another link that takes you to a medical journal written entirely in a language you didn't study.
You close the tab. You open it again. You ask an AI, which gives you a thorough, confident, completely context-free answer that somehow makes you feel worse. This isn't hypothetical. Two people built Rounds because they lived it, for someone they loved.
What Is Rounds, Exactly?
Rounds is a health research tool that runs on your Mac. You feed it your medical documents — lab results, imaging reports, discharge summaries, a photo of a rash — and it reads them, pieces them together, and tells you what it thinks is going on. Not in medical jargon. In plain language, with specific things you can do next, and with every claim backed by a real source you can open yourself.
It's not a doctor. But it's also not just a chatbot you throw questions at and hope for the best. The difference is context. Most AI tools respond to whatever you type in the moment. Rounds builds a picture first — of you, your records, your family — and reasons from that picture. You don't have to ask the right question. It already knows what questions to ask.
Why Most AI Health Answers Are Frustrating
When you search for a symptom or a lab result, you're asking a question in isolation. The AI doesn't know if you're 47 or 27. It doesn't know what medications you're on, or that your potassium has been slightly elevated for two years, or that your mum has the same condition. It gives you the most statistically average answer it can, which is often accurate for nobody in particular.
Rounds approaches this differently. Before it tells you anything, it reads everything you've given it. Then it searches the medical literature for your specific situation. Then it offers hypotheses — not diagnoses — ranked by how well they fit your actual data, each one linked to a guideline or a study you can verify yourself. If it can't find a good source, it says so. It doesn't fill the silence with a confident guess.
What It Actually Looks Like to Use It
Say you drop in a blood test. Rounds reads the whole panel — not just the flagged values, but all of it together. It might notice that your ferritin is low, your TSH is suppressed, and your kidney function markers are slightly off, and recognise that these things could be related. It runs searches against clinical guidelines for each signal — not general health articles, actual guidelines from bodies like the American Thyroid Association or the British Journal of Haematology — and surfaces the ones most relevant to your numbers.
On the right side of the screen, you see a trust-ranked source list. A clinical guideline sits higher than a systematic review, which sits higher than a case report. A single anecdote from a medical forum doesn't appear at all. Then it gives you next steps — not 'talk to your doctor' — something more like: 'Ask your GP for a repeat ferritin plus iron studies, then add the result here.' Specific. Actionable. With the source that justifies why.
Worth Being Honest About the Limits
Think of it as the preparation you do before you see a doctor — so you walk in knowing the right questions to ask instead of starting from zero.
For now it requires a Mac and a Claude Code subscription, so it's aimed at technically comfortable users. It's open source though, which means anyone can inspect how it works — or help build what it becomes.
The Family Angle Nobody Else Has Thought Of
Here's the part that makes Rounds genuinely different from every other health AI tool out there. It thinks in families. You can add cards for your partner, your parents, your kids. Each person has their own record. But Rounds reasons across all of them.
If one parent has been diagnosed with something that has a hereditary component, Rounds can flag that the children might be worth screening earlier than standard guidelines suggest — and it'll show that on the child's card, linked to the evidence for why. This is how a good family doctor thinks — about a household over time, with context that carries forward.
The Privacy Question
Medical data is among the most sensitive information you have. Most health apps send it somewhere — to a server, to a company, to a database that will eventually be breached or sold or acquired. Rounds doesn't. It runs entirely on your Mac, using the Claude Code AI you already have installed. Your documents stay in a folder you control. Nothing goes to the Rounds team.
The onboarding actually shows you this as a diagram: Your Mac ↔ Claude. No middleman. It's a rare thing for a software company to put their privacy architecture front-and-centre during setup, before you've even created an account. But for a health tool, it's the right call.
Why This Exists Now
The founders didn't build Rounds because they saw a market gap. They built it because they needed it and it didn't exist. The medical system gave them information in pieces — a result here, a note there, a follow-up appointment three weeks away with no explanation in between. Nobody was putting it together into something they could understand and act on.
Rounds is the tool they wish they'd had. That origin matters. It's the difference between a product built to capture a category and a product built to solve a real problem for real people.
The Bottom Line
Nobody should have to navigate a scary health situation armed only with Google and anxiety. The information that could help them exists — in guidelines, in research, in decades of accumulated clinical knowledge — but it's locked behind jargon, paywalls, and the assumption that only doctors need to read it.
Rounds is an attempt to unlock some of that. Not to replace doctors, but to make the space between appointments less frightening. It's free, open source, and available now at github.com/Rounds-Org/rounds. If you've ever opened a lab result and felt completely alone with it, it's worth a look.
