How Iris remembers you — and how to keep your health record accurate

Most AI conversations evaporate when you close the window. You described your entire medical history, had a useful discussion, and tomorrow the AI has no idea who you are. For health investigation — where building understanding over weeks and months is the whole point — that's a dealbreaker.

Iris maintains persistent memory. Your conditions, medications, test results, investigation findings, and health context carry across every conversation. When you mention a new supplement on Tuesday, Iris knows about it on Friday. When you say "I'm still tired," it responds with awareness of everything you've discussed before.

This isn't magic. It's a specific system worth understanding, because the better you understand it, the better it works for you.

How the memory system works

Iris stores memory as markdown files organized into folders — think of it as a filing cabinet for your health information.

Three special files — called system notes — are created during your onboarding and loaded into every conversation automatically. Identity holds who you are: your conditions, medications, demographics, and health goals. Current Focus tracks what you're actively investigating and what to follow up on. Reminder Preferences stores how and when you want to be contacted.

Beyond these, your memory grows as you use Iris. Topic notes cover specific subjects — migraine triggers, sleep patterns, gut investigation findings. These get organized into folders, each with a summary file. When you start a conversation, Iris browses these summaries and selectively loads the notes most relevant to your current question.

Who manages the memory

A specialist agent called the Memory Manager handles all the behind-the-scenes organization. When you mention something worth remembering in conversation — a new medication, a doctor's recommendation, a pattern you've noticed — Iris delegates to the Memory Manager, which decides whether to update an existing note or create a new one.

The Memory Manager's core principle is consolidation: one source of truth per fact. If you mention a medication change, it updates your existing medication note rather than creating a second one. Notes are always rewritten in full to keep them clean, and every version is saved — you can revert to any previous version if something goes wrong.

What Iris loads (and what it might miss)

System notes — Identity and Current Focus — are always loaded. Iris always knows the basics about you. Topic notes are loaded selectively based on what seems relevant to the conversation. This is a tradeoff: loading everything would consume too much of the AI's working memory, but selective loading means Iris occasionally misses something relevant.

The practical fix is simple. If Iris doesn't seem to remember something, tell it where to look: "Check my notes about sleep" or "Load everything about my gut investigation." You can also ask "What notes did you load?" to see what context it's working with. Directing Iris to the right files immediately improves the conversation.

You can see and edit everything

Your memory in Iris is fully transparent. Visit the Memory Manager tab to browse every note, every folder, every summary. You can read what's stored, edit notes when information changes, add context that Iris missed, and delete anything you want removed.

This isn't a locked medical record maintained by someone else. It's your data, organized as readable files, structured to make your health investigation effective. Most users won't need to actively manage their notes — the Memory Manager handles the maintenance — but the ability to inspect and override everything is always there.

When your record goes stale

Health information changes. You stop a medication. Get new test results. Discover that something you thought was a trigger actually isn't. If Iris is still working with old information, it's giving answers based on a version of your health that no longer exists.

Some types of information go stale faster than others. Medications change with every prescription adjustment — if Iris still lists a medication you stopped three months ago, every recommendation that considers your medication profile is based on wrong data. Test results get superseded — your TSH was 4.2 six months ago, but after adjusting levothyroxine it's now 2.1. Patterns and conclusions evolve as your investigation progresses — a "trigger" you identified early on may turn out to be coincidental.

The simplest approach: update Iris whenever something changes. After a doctor's appointment, medication adjustment, or new test results, say "Update my notes — I stopped taking metformin" or "Add my new bloodwork results." Iris delegates to the Memory Manager. Takes seconds, improves every future conversation.

For a deeper cleanup, the CTA below walks you through a structured review of everything Iris currently knows about you. Worth doing every few months, or whenever your health situation has shifted significantly.

References

1. Clinical decision support and contextual information — JAMIA, 2018. Context improving health recommendation quality.
2. Medication reconciliation and patient safety — BMJ, 2021. Outdated medication data as a source of clinical error.
3. Diagnostic reasoning and patient history — NEJM, 2009. History integration improving diagnostic accuracy.
4. Patient-generated health data accuracy — Journal of General Internal Medicine, 2020. Importance of keeping longitudinal health records current.