Two of the most-requested features since Mother Nature AI launched were the same thing in different words: "I want it to remember what I told it last time," and "I want to see what's actually changed since I started this protocol." Today both are live.

Memory persists your health context across conversations. Journaling captures the day-to-day inputs — symptoms, supplements, sleep, mood — that the AI then uses to surface patterns and check progress against goals. The two work together, and along with the wearable and EHR integrations we already had, they bring Mother Nature AI noticeably closer to functioning like a personal-physician layer rather than a stateless Q&A tool.

This post is the practical version of what we shipped, why each design choice was made the way it was, and what is coming next.

The problem we were solving

Most AI assistants are stateless. You tell ChatGPT about your hypothyroidism on Monday, and on Wednesday it has no idea. For general consumer questions, that is mildly annoying. For health, it is actively dangerous. Without persistent context, the AI cannot remember that you are on levothyroxine when you ask about ashwagandha (which can suppress thyroid hormone). It cannot notice that you have mentioned afternoon headaches three times in two weeks. It cannot track whether the magnesium you started last month is actually doing anything.

The other half of the problem is that asking the right question at the right moment is hard. Most people do not show up to a chat with "my HRV has been declining for three weeks and my sleep efficiency is down 8 percent — what should I look at?" They show up with "I feel off." Without journal data, the AI is starting from scratch. With it, the AI can answer the question the user is actually asking.

What Memory does

Memory is opt-in. You enable it in settings, and from that point forward, the AI keeps a structured profile of your health context across conversations. The profile is a small set of facts: conditions and diagnoses, current medications and supplements, allergies, sensitivities, goals, and stated preferences. Everything in it is editable. You can see exactly what is being remembered, change anything that is wrong, or wipe the entire profile in one click.

Concretely, here is what changes once Memory is on:

Every supplement or herb suggestion is automatically interaction-checked against the medications and supplements already in your profile.
Allergies and sensitivities are flagged before they appear in any recommendation, not after.
The AI can refer back to past conversations: "When you described this in March, you also mentioned afternoon energy crashes — is that pattern still happening?"
Recommendations are scoped to your stated preferences (vegan, no caffeine, prefers tinctures, etc.) without you having to repeat them.

We deliberately kept the Memory profile small and structured. Every "remembered fact" is something you can see in plain language and edit. We do not store conversation transcripts in Memory itself — those stay in the chat history where you can review or delete them. Memory is a separate, much smaller layer that holds the durable health context.

What Journaling does

Journaling sits adjacent to Memory and adds the longitudinal layer. You log daily entries — as detailed or as brief as you want — covering whatever you care about: energy, mood, sleep quality, symptoms, supplements taken, meals, training, anything else. The AI then analyzes the journal alongside your wearable data and surfaces what it notices.

A few examples from beta users (paraphrased and anonymized):

"Your energy ratings have gone up 30 percent since you started B-complex three weeks ago."
"On the days you logged poor sleep, your next-day anxiety scores have been about twice as high. Sleep looks like the highest-leverage thing to address right now."
"You have been logging evening ashwagandha for two weeks and your morning HRV from Oura has improved each week. Worth keeping for now."
"You added a new prescription last Friday — I checked it against your supplement stack and there are no flagged interactions."

The journal connects directly to the integrations Mother Nature AI already supports — Apple Health, Oura, Whoop, Garmin, Fitbit, and MyChart — so wearable data and lab results show up alongside what you typed. You do not have to manually log heart rate or sleep when your watch is already doing it. You do have to log the things only you know: "had a stressful day," "missed dinner," "started feeling congested."

Every Sunday, Mother Nature AI generates a weekly summary based on the journal and your wearable data. The summaries are designed to be shareable with a clinician — a clean snapshot of what changed, what looks like it is working, and what to keep an eye on. Several integrative-medicine practitioners in the beta cohort started asking patients to bring these summaries to appointments, which surprised us in a good way.

The design choices behind it

A few decisions worth flagging because they were genuinely contested internally:

Memory is opt-in, not opt-out. The default-off design makes onboarding slightly worse — most users do not turn on a feature they have not used. We chose it anyway because health data is the highest-stakes category to get wrong on consent, and an opt-out default would have felt deceptive even if technically defensible.

Memory is editable, not append-only. You can delete anything you wrote at any time. There is no internal "history of edits" the AI uses behind the scenes. If you remove the fact that you have PCOS from your profile, the AI does not silently keep referencing it.

Memory and chat history are separate. The conversational history in the chat panel is its own thing, with its own retention controls. Memory is a much smaller layer of structured facts, sourced from chats but stored separately. This matters because deleting the chat panel does not silently delete what the AI knows about you, and clearing Memory does not erase your past conversations. Two distinct controls, two distinct mental models.

No model training on your data. We do not use journal entries, Memory profiles, or chat transcripts to train Mother Nature AI's underlying models. Period. The relevant section of the privacy policy spells this out in detail.

The journal is structured but flexible. We considered making journaling a strict daily form (mood 1–10, sleep hours, symptoms checklist, etc.) and we considered making it pure free-text. Both extremes had problems — strict forms feel like work and people stop logging; free-text is hard for the AI to reason over. The shipped version sits in the middle: a small set of optional structured fields plus a free-text area. The AI extracts what it can from the free-text and you can correct anything that gets parsed wrong.

What it does not do

Worth being explicit about the limits.

Memory is not a medical record. It is a context layer for AI conversations. The AI uses it to give better recommendations and catch interactions, but it is not a substitute for a clinician's chart, an EHR entry, or a prescription record. If you need that level of formality, the VitalIQ continuous monitoring platform and the EHR integrations through MyChart are the right tools.

Journaling is not a diagnosis engine. The AI is doing pattern recognition over your inputs and your wearable data; it is not making a clinical diagnosis. Anything it surfaces — "your symptoms suggest you might want to talk to your doctor about thyroid function" — is a flag to discuss with a clinician, not a verdict.

Memory does not persist between accounts. If you sign in on a new device, your Memory comes with you (it is tied to your account, encrypted at rest, transmitted over TLS). If you log out without an account, nothing persists.

What is coming next

A few things are in active development:

Smart reminders. Time-based prompts based on your logged routine: "you usually take your evening supplements around 9 — they are not in today's journal yet."
Lab integration with annotations. Upload a lab result and the AI walks through it with the context of your medications and prior labs, then suggests what to ask your doctor.
Cohort insights, fully opt-in. Anonymized and aggregated patterns across users with similar profiles ("most users in your age group with PCOS who tried inositol reported X"). This is opt-in, k-anonymized, and uses no individually identifiable data. We are being careful with this one because it is the area where it would be easiest to get the privacy story wrong.
Practitioner workspace. A dedicated interface for integrative-health providers to review patient journals (with patient consent) and pull AI-generated summaries into their charts.

Memory and Journaling are the foundation underneath all of those. A persistent, personalized, longitudinal AI is fundamentally a different product than a stateless chatbot, and the next several quarters of our roadmap depend on having those two layers in place.

If you want to try them, both are live now in the chat. If you have feedback — particularly on what the AI is getting wrong about you — please send it. Every miss is a useful signal.

— Christopher Dobbie

Try Memory and Journaling now. Open Mother Nature AI — connect your wearables (Apple Health, Oura, Whoop, Garmin, Fitbit) and your MyChart, and the AI will start building your personal context from your first conversation.