Readiness, recovery, training, and biomarker coaching — read from the wearable and lab panel you already own no sensor, no blood draw, no clinician of ours, reasoned by a frontier AI model, on the Claude plan you already pay for. Your records stay in files you own. Every recommendation carries its study and its strength of evidence. Nothing sold.
Coaching, not medical care — no longevity or anti-aging promises; the name is the long view, not a claim on your lifespan. Ilfirin runs on your own computer, reasoned by Claude (a frontier AI model) through the Claude plan you already pay for — so "no new subscription" means no extra fitness-app fee, not "free": you still need a Claude plan and an always-on machine. Privacy by architecture — no Ilfirin cloud, account, or database. Your records stay in files you own; during a chat, the readings Claude needs pass through Anthropic. Reads today: Fitbit · Pixel · Withings · Garmin (community); Whoop and Oura addable; Apple Watch and CGM not yet.
Most of the passive signal — heart rate, sleep, steps, recovery — is read for you, with no form to fill. The rest is a short chat. Inputs in, a cited reading out.
Stated plainly, because each one is checkable.
Every recommendation ships with the study behind it and an honest tier, from Tier 1 (large cohorts, well settled) to Tier 4 (early, unproven). "Research-backed" is an adjective. Ilfirin reports which research, and how far it actually reaches.
No supplements, no prescriptions, no marketplace. That is exactly why it can flag NMN, cold plunges, and time-restricted eating as the Tier 3–4 bets the evidence says they are — rather than inventory to move.
It runs on your own computer; your data is plain files on your machine, with no Ilfirin account and no company database. Privacy by architecture, not by promise — during a chat, the readings Claude needs pass through Anthropic, and we say so.
Your watch, scale, and cuff land in one view automatically — you're not transcribing numbers by hand. The device list is stated exactly, including what isn't supported yet.
Fitbit · Pixel · Withings · Garmin (community). Most of the passive longevity signal is captured with zero ongoing input. The remainder — food, bloodwork, and how you actually feel — is a normal chat, in seconds. If you're filling out a form, we failed.
No single composite number — components, shown. These are the same five cards you read on the dashboard.
VO₂max is the strongest single reading we track — it predicts all-cause mortality more strongly than smoking or high blood pressure, with no observed ceiling (Mandsager 2018, JAMA Netw Open). Grip strength is a separate marker. The resistance-training sweet spot is a modest 30–60 min/week.
~7–8 h, and regularity may matter as much as total (Windred 2024). Heart-rate recovery and HRV are honest but noisy autonomic signals — read as trends, not verdicts. Loud snoring or witnessed apneas → test for sleep apnea; untreated it shortens lives, and it's treatable.
The Mediterranean pattern is the most RCT-supported way to eat for the long view (PREDIMED). Enough fiber and protein; less ultra-processed food. On alcohol, the health-optimal dose is zero — we report the evidence and don't moralize about it.
ApoB is the headline marker — causal for atherosclerosis, so lower reads better. Plus Lp(a), inflammation (hsCRP), glycemic control (HbA1c), and blood pressure. PhenoAge is computed as a biological-age trend — never a precise number, and labelled Tier 2. No $300 epigenetic "aging clocks."
Weak social connection carries a mortality signal comparable to smoking (Holt-Lunstad 2015). Plus air quality (no safe threshold; modern targets ~5–9 µg/m³ PM2.5), purpose, cognitive engagement, and a calendar of proven age-appropriate screenings.
The complete framework — five domains, each target with its peer-reviewed source and its honest tier Tier 1 settled · Tier 4 early, unproven.
Like a calibrated meter a trainer brings to your house — it reads your own notebooks on site and takes no copies home. The opposite of mailing your records to a company you've never met.
Records stay on your machine; during a chat the readings Claude needs pass through Anthropic, so it can coach with full detail. The default.
Raw values stay local. Claude reads "within range" or "trending up" — not your exact ApoB or weight. Some granularity traded for a narrower path.
A model runs entirely on your machine. AI quality drops; the privacy is absolute. Your call, stated as the real trade-off it is.
Walk the real instrument on sample data: the local five-domain dashboard, the single actionable call, and meal logging that cites its sources.
Five domains, one view — components shown, never a single composite score. The headline is the one signal worth acting on this week.
The one thing to do next — with the study and the tier carried alongside it, so you can see how far the recommendation reaches.
It flags the guess and cites its source instead of writing fiction — a low-confidence match is surfaced for you to confirm, not silently logged.
Tell us your devices and your setup. We provision your own instance and send a personalized installer — no terminal required. Access is limited while we keep the beta small.
An early, invite-based beta — not a finished, polished product, and not pretending to be. Coaching, not medical care. Immortality is the name, not the offer — Ilfirin coaches toward the long view, with no claim on your lifespan.