Introducing
Meeds Mirror 1.0

The unit of health intelligence should not be the prompt. It should be the person.

The missing model

A laboratory result records a measurement. A watch records a signal. A conversation records what someone noticed. Each can be informative. None, by itself, is an understanding of the person who produced it.

Much of our health information lives in this unfinished state: observations separated from their circumstances, measurements separated from their history, questions separated from what came before. Gathering the fragments in one place is useful, but a larger collection is not necessarily a clearer picture.

The missing layer is the relationship between them. When was a measurement taken? What changed around it? Which details are still current? What did the person correct? What remains unknown? An intelligence system that cannot preserve those distinctions can be articulate while repeatedly starting from zero.

Personal health intelligence begins with a different design problem: building a coherent, evolving context around one person. The question is no longer only what a model can answer. It is what an intelligence system can carry forward, reconsider, and make useful when that person returns.

Introducing Meeds Mirror 1.0

Mirror 1.0 is our first practical foundation for that direction. It brings health-oriented conversation together with personal context, structured measurements, selected files, and access to public evidence. The aim is not simply to answer a question about health. It is to make the answer relevant to the person asking, within the information they choose to provide.

The core of Mirror is the system around the model: the architecture that gives information continuity, selects relevant context, enforces the user’s choices, and connects interpretation to evidence. We are building this as enduring infrastructure—so advances in AI can become advances in a person’s understanding, rather than another reason to start over.

The distinction matters for what we build next. Better models can improve an individual response. A durable personal intelligence system must also improve the way information survives between responses: its meaning, origin, timing, and conditions of use.

How Mirror works

Mirror separates observations from the context supplied to a model. Saved profile details and structured health fields can provide relevant background, including user-entered biomarker information. Dated entries can retain units and source labels. Conversation history supplies the current thread; selected file attachments supply material for that conversation.

A useful system must distinguish what matters for this question from everything it could include. The server retrieves a bounded selection of personal context and conversation history, while attachment checks establish account ownership and the stored version being used. Selection focuses the context; provenance connects the response to the information it actually received.

Specialist experiences, including sleep, nutrition, fitness, medication, and genetics, apply a focused lens to the conversation. Public-evidence tools can retrieve material from sources such as PubMed, ClinicalTrials.gov, RxNorm, and openFDA when the request and permissions allow it. These tools extend the conversation into scientific literature and reference information, while keeping personal observations and published evidence distinct.

Memory gives this architecture a further dimension. We have implemented systems for proposing persistent health context, requesting confirmation, correcting entries, and controlling future use. Automatic memory extraction and mutation remain behind release gates, as does persistent health-document processing. These are distinct from the selected-file conversation workflow available today.

Why continuity matters

A single answer can be useful and still leave the underlying problem untouched. A person returns with a new observation and must reconstruct the same background. The system encounters a number without its earlier values, or a goal without the circumstances that shaped it.

Continuity changes the work intelligence can do. Instead of repeatedly collecting a biography, it can help examine what has changed, what has persisted, and where two accounts disagree. The value lies less in remembering more indiscriminately than in keeping the right distinctions intact.

That requires an architecture beyond a long conversation window. Conversation history preserves an exchange; structured records make measurements usable; persistent memory is designed to carry selected context between interactions. The research challenge is to make these layers cooperate across months and years, so new information can refine the picture instead of simply extending a transcript.

The standard for health intelligence

Personalization raises the engineering standard. A missing measurement is different from a normal measurement. An old record may no longer describe the present. A user’s account, a device estimate, a published study, and a model’s interpretation each have a different role in understanding health.

Our standard is to make those differences legible: preserve provenance where available, acknowledge incomplete context, expose supporting sources, and make correction possible. Source quality matters as much as retrieval. The meaningful question is how a study’s methods, population, and conclusions bear on the question being asked—and what remains unresolved.

Permissions are part of the architecture, not an afterthought. Personal context depends on the user’s choices and server-side authorization. Private conversations exclude saved profile, memory, and other conversations from their context; current-message attachments can still be used. Our Privacy Policy explains processing, storage, and user rights.

Meeds brings this foundation to mobile, web, and desktop through account-backed information and shared server policies, with Apple Health support on iOS. Specific capabilities vary by platform, plan, and release.

Mirror supports understanding and better-informed conversations with qualified clinicians. Clinical decisions remain with the person and their care team.

The research frontier

Our longer-term ambition is personal health intelligence that can reason across scales and timescales: connecting an observation made today with a longer history and a deeper account of biology. This calls for advances in temporal reasoning, multimodal integration, and representations that can accommodate both new evidence and uncertainty.

A central research goal is to connect mechanistic biological knowledge with longitudinal observations: moving beyond isolated correlations toward explanations that can be examined and tested. Validated individualized prediction belongs to that research agenda, with progress measured against outcomes, appropriate baselines, and the realities of incomplete, changing data.

Experimental work on adaptive 3D representations explores how biological information can become easier to inspect. The opportunity is to move between a broad view and a specific observation, making relationships visible rather than leaving them buried in a record. We are studying these as interfaces for exploration and explanation, with source data distinguished from illustrative representations.

The language model will change as capabilities advance. The enduring work is the system around it: how observations are represented, how context is selected, how permissions remain binding, and how claims can be traced and corrected. That infrastructure is what can make better intelligence progressively more useful to one person.

The beginning

Mirror 1.0 establishes a direction: health intelligence organized around a person, not a sequence of disconnected requests. The next generation of models should meet richer context, better evidence, and a system built to carry understanding forward. That is the infrastructure we intend to build.

Our mission is to make increasingly capable intelligence useful at the scale of an individual life: helping people connect what they know, examine what changes, and understand themselves over time.

Know yourself.