Today, November 6, Insilico Medicine announced the open-source release of Longevity Grok, a Personal Longevity Superintelligence (PLSI) developed in collaboration with Harvard aging researcher Dr David Sinclair.
And I have already started using it.
The concept is ambitious: an AI system that can continuously integrate your health data, identify emerging risks, research relevant scientific literature, propose interventions, predict how your biomarkers might respond, and eventually evaluate whether those predictions were correct.
It is designed to work 24 hours a day, seven days a week, even when you are not actively interacting with it.
But before I explain how I use it, there is an important question:
Why not just ask ChatGPT, Claude, or Grok about your health?
This was actually my first question. We can already upload blood tests to a large language model, ask it to interpret biomarkers, request dietary recommendations, and discuss the latest longevity research.
So what makes Personal Longevity Superintelligence different?
It is the infrastructure built around it.
A conventional LLM interaction is predominantly reactive. You provide information, ask a question, and receive an answer. Some AI assistants now support memory, scheduled tasks, and external integrations, but maintaining a coherent, continuously updated model of your health still requires considerable effort. PLSI is designed around a different paradigm: persistent, autonomous, longitudinal health management.
There are four particularly interesting differences.
1. It builds a continuously updated representation of your biology.
Instead of repeatedly uploading laboratory reports, wearable summaries, and medical records, PLSI organizes heterogeneous data into a longitudinal health history. The system uses clinical standards, including LOINC for laboratory observations, RxNorm for medications, and UCUM for measurement units, to standardize information that would otherwise be difficult to compare. This matters enormously. A single fasting glucose measurement may tell us relatively little. A trajectory incorporating glucose, HbA1c, sleep, physical activity, dietary patterns, medications, and subsequent metabolic changes is considerably more informative. You can hook up your watch/ring/almost any other wearable, and let PLSI read, analyze, map trajectories and incorporate all conclusions into your n of 1 research.
2. It does research without waiting for you to ask questions.
PLSI uses a multi-agent architecture involving up to 30 specialized agents. These agents can perform different tasks, from organizing medical data and monitoring relevant literature to formulating hypotheses and evaluating proposed interventions. This introduces an important shift: instead of asking your AI what you should investigate, the system can proactively identify questions worth investigating.
3. It makes predictions that can be tested against future measurements.
This is perhaps the feature I find most scientifically compelling. PLSI introduces prediction cards that document the expected effects of an intervention, including:
The biomarker expected to change.
The predicted magnitude or interval of that change.
The timeframe over which it should occur.
The supporting scientific evidence and its strength.
When subsequent measurements become available, the system can compare the prediction against the observed outcome. For example, suppose an AI recommends a dietary modification intended to improve ApoB concentrations. Rather than simply stating that the intervention should be beneficial, it could record a quantitative expectation and evaluate the result following repeat testing.
Over time, the system could track which predictions are consistently accurate and which are not. This is conceptually closer to an individualized, longitudinal N-of-1 experiment than to conventional AI-generated health advice. Of course, observing a biomarker change does not establish that the intervention caused it. That still requires appropriate experimental design and attention to confounding variables.
4. It is open-source.
Both the personal assistant and the communal research infrastructure are being released as open-source software. That means researchers and developers can inspect, modify, and extend the architecture. Importantly, none of these features proves that PLSI currently produces more accurate clinical recommendations than a well-configured general-purpose LLM. That comparison will require independent benchmarking, but Insilico knows a thing or two about benchmarking;) (they made a Cell cover last month about it). But the architecture addresses a fundamental limitation of episodic AI consultations: health is a continuous biological process, while most interactions with AI remain isolated conversations.
How I am using it
My current setup is relatively simple, but I am already fascinated by the possibilities.
First, I connected my Apple Watch data.
This gives the system access to information about my activity, sleep, and other available physiological measurements, allowing it to examine patterns over time. I am particularly interested in how seemingly small variations in daily behavior and biomarkers might correlate with longer-term changes in health biomarkers.
Second, I have started taking pictures of virtually everything I eat or drink.
Yes, everything.
I send the photographs to the PLSI Chief of Staff, which estimates the macronutrient composition of each meal, including proteins, carbohydrates, and fats, alongside potential micronutrient contributions. It then considers what I have already consumed and suggests what I might want to eat next. For instance, instead of generically recommending a healthy dinner, it can consider whether my estimated fiber intake has been low, whether my dietary patterns might leave gaps in particular nutrients, or whether my next meal could complement what I have consumed earlier. Naturally, estimating nutrition from photographs is imperfect. Portion sizes, cooking methods, oils, and hidden ingredients can introduce substantial errors, especially for micronutrients. I treat these numbers as approximations, not laboratory measurements. Also, I still advocate to using your brain first, AI second. And I urge everyone to be especially cautions with AI-diet recommendations, because, even from personal experiance, it can very easily turn into obsession and borderline eating disorders. Of course, the goal is not to outsource our entire cognitive load to AI. We still need to remain attentive to how our bodies feel and what they are trying to tell us. Yet, over the past decades, many of us have become increasingly disconnected from our own physiological signals, often recognizing them only when something becomes sufficiently uncomfortable or severe to demand immediate attention. I hope PLSI can help us rebuild that connection by highlighting subtle patterns, encouraging greater awareness of our physiological responses, and helping us interpret signals we might otherwise overlook. But no matter how sophisticated the system becomes, how well it is designed, or how comprehensive its underlying data are, critical thinking, common sense, and our own biological intuition must remain central. AI should augment our ability to understand and care for ourselves, not replace our judgment or turn us into passive followers of algorithmic recommendations. Ultimately, the objective is to use PLSI to become more informed, more autonomous, and more in tune with our bodies—not less.
Nevertheless, the convenience is extraordinary.
I do not want to spend my life manually logging ingredients into spreadsheets. I want health monitoring to become as effortless and unobtrusive as possible.
Third, I am leveraging it to monitor and understand my mental health.
I am particularly curious to see corelations between my wearable data and mood patterns over time. I am so curious to see the outcomes - I will ask it for a deep dive after a month of data gathering
And finally, I have it conduct daily research into health questions relevant to my individual laboratory results.
This is my favorite application so far.
Rather than receiving generic longevity news, I want to understand developments in areas that might actually matter to my physiology. If a particular biomarker warrants attention, I want the system to examine relevant mechanisms, evaluate emerging evidence, identify potential confounders, and help formulate questions for further investigation. Ideally, the process becomes increasingly personalized as more longitudinal data accumulate. I am not outsourcing clinical decisions to AI. I am using it to expand the amount of scientific attention I can devote to my own health. That distinction is important.
From generalized longevity advice to individualized experiments
One of the biggest problems in longevity medicine is that biological responses to interventions are heterogeneous.
The same nutritional intervention, exercise protocol, or medication can produce substantially different outcomes depending on genetics, age, metabolic state, existing disease, and numerous environmental factors. Yet much of the consumer longevity industry still operates through generalized recommendations. Take a supplement. A company might cite a randomized trial showing improvement in a particular biomarker across a study population. But this does not establish that the supplement will meaningfully benefit you. Did you have the relevant deficiency or biological phenotype to begin with? Was the studied population comparable to you? Is the endpoint clinically meaningful? Would any measurable improvement occur at your baseline?
I have long advocated for a more conscious approach to supplementation and longevity interventions, grounded in objective biomarkers and scientific evidence. PLSI could help make that approach substantially more accessible. Imagine a system that not only examines whether an intervention is supported by published research but also identifies the specific biological reason it might be relevant to you, establishes measurable expectations, and examines follow-up data. We are not yet at the point where such systems can reliably establish individualized treatment effects. But the infrastructure for systematically testing predictions is an important step in that direction.
The applications I am most excited about
Beyond my own experiments, I see several particularly compelling opportunities.
1. Making physician consultations substantially more productive
Physicians rarely have the time to review years of fragmented laboratory reports, wearable data, nutritional records, and research literature for every patient. A structured longitudinal summary could help clinicians identify relevant trends, examine treatment responses, and focus appointments on the questions that matter most. PLSI explicitly reserves diagnosis and prescribing decisions for licensed physicians. That is precisely where I think its immediate clinical value could emerge: as a tool that helps organize evidence and support better-informed discussions, rather than attempting to replace medical judgment.
Ps. We will host a webinar soon about how PLSI can be used to train physicians for longevity medicine, as a part of Longevity Education Hub. Stay tuned!
4. Turning personal health monitoring into collective longevity research
This is where the project becomes particularly ambitious. PLSI includes an optional communal research infrastructure called Longevity Mothership. Users can choose to contribute deidentified longitudinal health trajectories to a shared research system. In principle, this could create datasets that connect real-world behaviors, interventions, physiological measurements, and health outcomes across large populations.Such datasets would have substantial limitations, including selection bias, measurement inconsistency, confounding, and challenges surrounding privacy and consent. Nevertheless, their research potential is considerable. Today, many longitudinal aging studies are expensive, relatively small, or limited in the frequency and variety of their measurements. Imagine studying biological aging using continuously updated multimodal observations from a large, consenting population. Not simply asking which factors correlate with longevity, but exploring how individual trajectories evolve following measurable changes in behavior or treatment.
From personal health monitoring to actual drug discovery
One feature I find particularly exciting is PLSI’s potential integration with PandaOmics, Insilico Medicine’s AI-powered drug discovery platform. Instead of limiting my personal AI to interpreting blood tests, tracking nutritional intake, or searching PubMed, I could potentially ask it to investigate the underlying biology of a particular health concern. Through PandaOmics, specialized AI agents can explore disease-associated molecular pathways, prioritize potential therapeutic targets, and investigate compounds that might modulate them.
There are two ways this works. The first involves structured bioinformatics tools that rank targets, investigate disease–aging associations, and retrieve information about compounds. The second involves a more autonomous research agent that can take a broader biological question and conduct a multi-step investigation. These capabilities require separate API credentials, and some research runs are billed. Importantly, the objective is not for AI to recommend experimental drugs for personal use. The system is designed to critically evaluate its findings, cross-check them against published literature and clinical trials, assign evidence grades, and identify observations that could falsify its hypotheses.
What fascinates me is the possibility of connecting individual biological observations to the machinery of modern drug discovery. Imagine noticing an unusual pattern across your longitudinal biomarkers and having an AI investigate the molecular pathways that might explain it, identify relevant therapeutic targets, and determine whether anyone is already developing interventions against them.
This does not mean every biomarker fluctuation represents a druggable biological problem, or that an AI-generated hypothesis is necessarily correct. But it creates an intriguing bridge between personalized health monitoring and computational geroscience. And perhaps most importantly, with optional participation in the Longevity Mothership research network, hypotheses generated from individual observations could eventually contribute to broader, collaborative aging research.
We are moving toward a world in which individuals may not only benefit from longevity research but also actively participate in generating it.
And while today’s system is still an early implementation of that vision, I am excited to see where it goes. For me, the ultimate goal is not to optimize every measurable biomarker or spend every waking hour thinking about health. Quite the opposite.
I want the science of maintaining my health to become increasingly sophisticated, while the effort required from me becomes increasingly small.
What about privacy?
I am wondering how many people will have commented on “absolute breach of privacy” before they got to this section;). There is an obvious concern with giving an AI access to laboratory reports, wearable measurements, and medical histories. These are among the most sensitive categories of personal information. PLSI is designed around a zero-trust architecture in which identifiable medical records remain within the user’s own environment by default. Participation in communal research is optional, with deidentification procedures and mechanisms intended to let users withdraw contributed data. Its open-source architecture also makes independent security review possible. Anyone considering such a system should understand where their data are stored, what is transmitted to external services, and what permissions have been granted.
Try it yourself
Nothing to fear, just a few easy steps:
If you haven’t already, set up a Grok account, and get a premium tier to have access to Grok Bot.
Copy this link https://github.com/Insilico-org/plsi and tell grok bot to install it
Follow on screen insructions and start chatting with PLSI on your current situation and longevity goals!
Watch the webinar on Insilico’s social media to learn more about specific usecases.
Share! It would be great to hear from you on social media about your experiance.