What Iris actually does — and how it's different from asking ChatGPT about your symptoms

You can ask any AI chatbot "why am I always tired?" and get a list of possible causes. Sleep apnea, thyroid, iron deficiency, depression, stress — the same list you'd find on WebMD. It's accurate in the way a phone book is accurate: everything's in there somewhere, but it doesn't help you find what you're looking for.

Iris does something different. It investigates.

The difference between answering and investigating

A chatbot answers questions. You ask, it responds, the conversation ends. Next time you open it, it doesn't know who you are, what you've tried, or what you ruled out last week. Every conversation starts from zero.

Iris remembers you. It knows your conditions, your medications, your test results, what you've been tracking, and what you're currently investigating. When you say "I'm tired," it doesn't give you a generic list — it gives you a response informed by the fact that your thyroid was normal last month, your sleep has been fragmented for three weeks, and your energy crashes correlate with high-stress days.

That's not a smarter chatbot. It's a different tool for a different job.

What Iris is built to do

Track your health from natural conversation. Say "had a headache after lunch, maybe a 6/10" and Iris creates a structured data entry — headache, severity 6, timing afternoon, possible trigger food. You don't fill out forms. You talk, and the system handles the structure. Over weeks, this builds a dataset that's actually analyzable.

Find patterns you can't see. Your brain notices obvious connections — "I always get headaches on Mondays." It's terrible at multi-variable correlations across weeks. You won't notice that your pain averages 2 points higher on days following less than 6 hours of sleep, and that this doubles when combined with stress and no exercise. Iris's Data Analyst runs that analysis across your entire tracking history.

Remember everything across conversations. Your medications, conditions, investigation history, and patterns all persist in organized notes. You can see, edit, and delete anything. When you start a new conversation, Iris loads the relevant context — you don't re-explain your medical history every time.

Check its own work. Every message Iris sends gets reviewed by a separate Supervisor agent that checks for safety, overconfidence, and factual accuracy. If something's off, a review annotation appears alongside the response. One AI generating answers and a different AI checking them — the same principle as peer review.

Help you prepare for your doctor. Two weeks of tracked data, analyzed for patterns and summarized into a clear timeline, is more useful in a 15-minute appointment than whatever you can recall from memory. Iris turns your investigation into something your provider can act on.

The cross-specialty problem

Healthcare is organized by specialty. Your gastroenterologist manages your gut. Your orthopedic surgeon manages your shoulder. Your psychiatrist manages your anxiety. Each specialist sees their slice — and that's appropriate. They're experts in their domain.

The problem is that your body doesn't respect these boundaries. Your gut condition affects your anxiety. Your pain medication has GI side effects. Your sleep disruption makes everything worse. No single specialist is thinking about all of these interactions, because that's not their job. Your GP is supposed to hold the big picture, but they have 15 minutes and dozens of patients.

Iris holds your full context across every condition, medication, and symptom you're tracking. When you're seeing a new specialist, Iris knows what the other specialists are managing. When a medication is prescribed, Iris can flag interactions with conditions being handled by a different provider. This isn't a theoretical benefit — it's the kind of thing that matters when your surgeon doesn't know about your GI condition, or when your new prescription has side effects that interact with something you're already taking.

No single appointment covers your whole picture. Iris does.

How it works under the hood

You talk to Iris. Behind the scenes, Iris coordinates a team of specialist agents — each handling a specific job.

The Tracking Agent turns your natural language into structured health data. The Memory Manager organizes and updates your notes — conditions, medications, findings, life context. The Data Analyst runs queries and statistical analysis on your tracked entries. The Supervisor reviews every response for safety and quality, recaps your conversations, and periodically reviews your overall investigation trajectory.

You don't need to think about any of this. You talk to Iris. The agents do their work. But if you want to see what's happening — what's stored in memory, what entries were created, what the Supervisor flagged — every agent has a transparent tab you can inspect.

What to expect in your first week

Day one is onboarding. Iris learns about you — your conditions, medications, what you're dealing with, what you've tried. This creates the foundation that every future conversation builds on.

The first few days, focus on talking to Iris about what's actually going on — your conditions, what you've tried, what you're worried about, what you want to figure out. Most of the relationship lives here: thinking out loud, asking questions, building up the context Iris carries for you across months and years. There's no daily ritual to perform.

If a real question surfaces — "is dairy actually a trigger?", "what's driving these afternoon crashes?" — Iris may suggest a tracking sprint: a focused stretch of consistent logging for a week or two so the Data Analyst has something to work with. Sprints are short, specific, and conversational. Iris asks the few things the investigation needs at the times you've agreed on, and you answer in chat. When the sprint ends, the daily logging ends with it. Tracking is one mode Iris can shift into, not the price of admission.

References

1. Patient-generated health data and clinical outcomes — Health Affairs, 2016. Longitudinal patient data improving health investigation.
2. The value of time-series symptom tracking — PubMed Central, 2020. Precise timing data revealing correlations missed by recall.
3. AI ensemble methods for reliability — PNAS, 2021. Independent models catching each other's errors.