Built for the people taking healthcare into their own hands

The shift

Healthcare has been quietly decentralising for a decade. The visible markers are wearables, direct-to-consumer labs, telehealth. These are downstream of a deeper change: tools that were once gated behind a clinician are increasingly in the hands of the person they were designed for.

Bloodwork moved first. Quest, LabCorp, and a new generation of labs like Function turned the panel from a clinical event into a recurring subscription. Continuous glucose monitors moved next. Once a device limited to diabetic patients, Levels and Stelo brought CGMs to anyone curious about their metabolism. Sleep tracking moved out of polysomnography labs and onto wrist wearables like Whoop, Oura, the Apple Watch, and the just-launched Fitbit Air. Recovery, heart rate variability, training load, body composition. Every one of them used to require a specialist appointment. Now they're a notification on a phone.

The direction is clear. Wearables are getting smaller, smarter, and more ambient. The intelligence layer is getting better at interpreting what they collect. Self-administered healthcare is moving from an edge case to a default state. The legacy health industry is not going to build the infrastructure for that future. Milligram is the iOS app that bridges the gap.

Peptides are part of the same shift, but the shift is happening differently in this category. The decentralisation isn't driven by a consumer product company building a better experience. It's driven by access. A decade ago, semaglutide was a niche diabetes drug. Tirzepatide hadn't been approved. Compounds like BPC-157, GHK-Cu, and TB-500 lived in bodybuilding forums where a few hundred people swapped notes about reconstitution protocols. Today, millions of people are running these compounds. Many through compounding pharmacies, many through the research-peptide market, many without prescriptions, and almost all of them learning as they go.

The category of user that has emerged is unfamiliar to the legacy health industry. They are not patients waiting for a prescription. They are not unhealthy people seeking medical care. They are typically curious, engaged, often technically literate, and almost always treating their own body as a system they can study and adjust. They read trial papers. They post bloodwork on Reddit. They iterate on their stacks with discipline. They are, in many cases, more informed about the specific compounds they're running than the average general practitioner they would consult.

The product category that should serve those millions doesn't exist. The peptide trackers on the App Store are dose logs with reminders. The mainstream health apps don't recognise the compounds at all. The general-purpose AI assistants don't know what your protocol is doing in your body, and most of them refuse to engage with the topic because the user isn't on a prescription pathway. The infrastructure for what these users are already doing, at scale and every day, barely exists.

That gap is what Milligram exists to close. Not as a wrapper around someone else's API. Not as a dose log with a chatbot bolted on. As the first piece of analytical infrastructure built specifically for someone running their own health protocol with the same rigour a clinician would apply to a patient.

People are going to manage their own health whether the tools exist or not. The question is whether the tools they use will be serious, or whether they'll continue to be improvisations stitched together from spreadsheets, Reddit posts, and the chat interface of a general-purpose AI that wasn't designed for the job. Milligram is a bet on the first option.

What Milligram is

Milligram is an iOS app. A tool, not a service. There is no concierge call you book. There is no clinician you talk to. There is no human in the loop reviewing your protocol. The premise is that the user is capable of running their own protocol with the right analytical infrastructure, and Milligram is that infrastructure.

The core of the app is a pharmacokinetic engine. Every dose you log gets run through trial-grade math. For sustained-release compounds like semaglutide, tirzepatide, and retatrutide, the engine uses a two-rate Bateman model with flip-flop kinetics, recognising that absorption from the subcutaneous depot is slower than elimination from plasma. The model is titration-aware, so the steady-state average concentration for a 2.5 to 5 to 7.5mg dose escalation reflects the actual sequence the user logged, not a flat-dose approximation. Confidence bands come from Monte Carlo simulation using published inter-individual coefficients of variation. When the app shows a range, that range is the p5-to-p95 spread, not an arbitrary tolerance. The numbers Milligram shows you are derived using the same methodology that Coskun 2022 and Granhall 2018 used to derive the numbers in their papers.

Around the engine sits the rest of what an operator actually needs. Dose scheduling. Bloodwork uploads with biomarker extraction. Photo scans with face landmark analysis. An AI advisor that reads your dose history and cites the trial literature when it tells you something. Apple Health integration. Notifications that arrive when they should.

Each piece exists because someone running a protocol needs it. Nothing exists because it looks good in a screenshot.

What we believe

People deserve trial-grade tools for their own bodies.

The math that runs in pharmaceutical research isn't sacred. The same two-compartment models, the same Bateman equations, the same Monte Carlo simulations that go into Phase III trial design can run on a phone. There's no good reason the person actually taking the compound shouldn't have access to that math.

The pattern of decentralising any technology is consistent. Computing. Publishing. Broadcasting. Capital allocation. Financial analysis. At first the technology is reserved for institutions because institutions are the only entities capable of using it. Then the technology gets cheap enough and accessible enough that the institutional moat dissolves, and a new category of user emerges who turns out to do unexpected things with the tools once they have them. Personal computing did this. Spreadsheets did this. The internet did this. Brokerage apps did this.

Pharmacokinetic modelling is at the front edge of the same arc. The math has been mature for decades. The data is increasingly published. The compute requirements are trivial by modern standards. The only thing missing has been a product willing to package the math for a user who isn't a pharmacologist. Milligram is that product.

The operator knows their body.

Someone running a protocol on themselves has access to signal no clinician sees. How they slept. How they felt at 3pm. When their appetite shifted. What their mood did. How their training session went. What their resting heart rate did over the night. The legacy medical model treats the patient as the subject of observation, a person whose physiology is examined at intervals by someone else with expertise.

Milligram treats the user as the observer. Every tool is designed around the assumption that the operator is actively measuring, comparing, refining, and asking questions of their own data, not waiting for instructions. The implications of this assumption run through every product decision. Notifications carry the dose and the route, not generic encouragement. Charts default to the user's data, not population averages. The AI advisor surfaces what's happening at the receptor and what the trial-band trajectory predicts, then lets the operator decide what to do with that information.

This is not a softer version of medical care. It's a different relationship between a person and the substance they put in their body. One where the person is in charge and the tool is in support.

Observation, not instruction.

Milligram will tell you that your tirzepatide buildup is at 94% and sitting two hundred times over the GIPR EC50 by published binding data. It will not tell you to take more. It will tell you that your hsCRP is trending up over the last three blood draws. It will not tell you what to do about it. The decisions about your body belong to you and your clinician. Milligram exists to surface signal, never to override it.

Observation-only is harder than it looks. Every product instinct pushes toward instruction: stronger engagement, clearer next-action UX, more legible "do this" recommendations. The pull is constant. Milligram resists it because the alternative, an app that tells users what dose to take, would compromise the entire premise of the product. The user is the operator. The tool exists to inform the operator. That's the line.

Privacy is foundational.

Health data is the most sensitive personal data there is. It outlasts you. It can shape decisions made about you for the rest of your life by parties who never asked your consent. Insurers. Employers. Governments. Future relationship partners. Future AI systems trained on it. The default posture of the technology industry, which is to collect everything and monetise later, is incompatible with handling this category of data responsibly.

The only ethical posture for a company collecting health data is the most protective one available. That's what shapes Milligram's defaults: encrypted in transit, on-device wherever possible, 60-day audit-only retention for AI conversations, no model training on user data by us or any third-party provider, no advertising identifiers, no third-party trackers, no exceptions. Some of these defaults make the product harder to build. None of them are negotiable.

What's in the app

The pharmacokinetic engine.

The core of Milligram is a deterministic numerical model that takes your dose history and returns a blood concentration curve over time. For sustained-release compounds like semaglutide, tirzepatide, and retatrutide, it uses a two-rate Bateman model with flip-flop kinetics, recognising that absorption from the subcutaneous depot is slower than elimination from plasma. This is why a once-weekly semaglutide dose doesn't drop to zero between injections. The engine is titration-aware, so the Cavg for a 2.5 to 5 to 7.5mg dose escalation reflects the actual sequence the user logged, not a flat-dose approximation. Confidence bands come from Monte Carlo simulation using published inter-individual coefficients of variation. When the app shows a range, that range is the p5-to-p95 spread, not an arbitrary tolerance. The same approach covers 100+ compounds across peptides, GLP-1 agonists, androgens, SARMs, and supportive compounds.

The AI advisor.

Milligram's in-app assistant is structurally different from a general-purpose chatbot. It has access to your dose history, your bloodwork, your biomarker trends, your protocol context. It runs queries through 26 specialised tools, not just chat completion, for receptor occupancy, PK trajectory projection, trial-band comparisons, and biomarker correlation. Every quantitative claim it makes carries a citation: trial-level for outcomes (STEP-1, SURMOUNT-1, Reta-P2), pharmacokinetic-paper-level for math (Coskun 2022, Granhall 2018, Willard 2020).

The discipline that makes this work is the refusal list. The advisor refuses to recommend dose changes. It refuses to suggest sources. It treats every privacy question by routing the user to the human policy rather than improvising. Each of those refusals is more important than any single feature. They define what the advisor is by defining what it won't do.

Bloodwork, body composition, journal.

Upload a PDF and Milligram extracts the biomarker panel, flags out-of-range values, and tracks trends across multiple labs. Take a body scan and it runs face landmark analysis and visual progress comparisons. Log subjective metrics like sleep, mood, energy, and recovery, and the app correlates them against your compound timeline. Apple HealthKit integrates wherever it can without sending data off the device.

Privacy that holds up to scrutiny.

Encrypted in transit. On-device wherever possible. 60-day audit-only retention for AI conversations, then purged. No model training on user data, by us or any third-party provider. Device attestation on every server call to prevent abuse. No advertising identifiers. No third-party trackers. The full posture, including what we collect and why, lives in the privacy policy.

Who it serves

Milligram is for the person who's already taken control of their own health and wants tools that match the seriousness of that decision.

It's the 22-year-old who started retatrutide because they realised they don't need a doctor to tell them what's right, and in many cases don't trust one to. They are not going to follow generic health and wellness advice. The world they live in and the future they see is not comprehensible to the healthcare industry or to the generation above them. To this generation, health is not about doing what you're told. It is about weighing pros and cons, and taking the risks where the cons of inaction are objectively larger. They do not treat health as something to check up on at an annual appointment. They treat it as a measurable system of protocols and outcomes, of trade-offs they want to understand themselves. People call this attitude crazy. Those same people have no idea what the other end of not taking things into your own hands looks like.

It's the 35-year-old on testosterone replacement who started through a private clinic two years ago and figured out within a few months that the prescribing doctor was not the smartest person in the room about his own protocol. He has done the panels. He has read the papers. He understands his body's response to esters and dose frequency better than anyone he could pay to consult. He does not want a coach. He does not need encouragement. He wants a tool that respects what he already knows.

It's the woman in her late 20s running GHK-Cu and Epithalon as part of an aesthetics protocol. She knows the literature on these compounds is thinner than the literature on GLP-1s. She also knows that aesthetics is not a frivolous concern, that how the world responds to you is a function of how you present, and that fact is older than the wellness industry pretending it isn't. She wants to know whether her timing matters, whether her dose matters, whether anything she is paying for is actually working.

It's the recovery athlete on BPC-157 who is trying to get more out of a body they push hard, with an orthopedic medical system that mostly engages only once something has actually torn. They are not waiting for the injury to do the work that prevents it. They are running a peptide most clinicians have never heard of. They want a model of what is happening in the tissue, the data to back it up, and the discipline to know when to cycle and when to push.

It's the person doing their first stack standing at a real threshold. They have ordered the vials. They have read what's available. Now they have to do something to their body that is materially different from a supplement decision, and they feel the weight of it. They want to know they have not missed something obvious, that the dosing makes sense, that they are not about to find out the hard way what they did not know. They are asking for what any adult deserves when making a serious decision about their own body.

It's the woman on a GLP-1 agonist who knows that the scale going down is not the same thing as the right thing happening. She has done the reading on lean mass loss in the trials. She is training. She is eating enough protein. She is doing all of it without a clinician treating her like a child, because the clinical conversation about female body composition is still largely the one it was in 1995, and she is past it. She wants the tools to see what is actually happening to her body, not just her weight on a scale.

What unites these users isn't an age range, a gender, a goal, or a compound. It's a posture. They have decided that managing their own health is not a fringe activity, not a deviation from a medical norm, not something that needs justifying. It is the work of being an adult in a body in 2026. They are not waiting for permission to do it. They are already doing it. Milligram exists to give them the analytical infrastructure that work deserves, and to treat that posture as the default state of the user, not a special case the product has to accommodate.