Every Antiaging Labs protocol is produced by nine deterministic agents, each with a typed output contract and a hard safety gate, running against a versioned clinical rulebook that is published in full. Same inputs, same rulebook version, same protocol, every time. Here is exactly how it works.
Your bloodwork PDF and intake answers enter at the top. A finished, safety-screened protocol and a 20+ page report come out the bottom. Each stage hands the next a typed JSON artifact, so every intermediate decision is inspectable, and any single agent failing halts the run rather than guessing.
Parses the lab PDF, converts units, reconciles lab-platform differences, handles missing values.
biomarkers.jsonFlags missing markers by clinical leverage, what's absent and how much it matters.
panel-completeness.jsonStructures medical history, medications, lifestyle, training, sleep, stress, family history.
intake.jsonDeterministically fires every triggered rule from the 217+ rule versioned rulebook. Authors no prose.
rule-evaluation.jsonReduces a list of independent rule-fires to one or two coherent metabolic phenotypes.
phenotype.jsonTraces the causal graph to name the 2–3 upstream mechanisms worth attacking, not just symptoms.
rootcause.jsonCross-checks every candidate intervention against medications, conditions, and contraindications.
safety.json · HARD GATEAssembles the personalized protocol, nutrition, training, recovery, supplements, from cleared interventions only.
protocol.jsonComposes the client-facing 20+ page report (HTML + PDF) with every finding traced to its rule.
report.html / .pdfNothing in the pipeline is a free-text hand-wave. Each agent emits structured JSON that the next agent consumes, and every fired rule carries its evidence trail and the rulebook hash it came from. A fired rule looks like this:
// one entry from rule-evaluation.json { "rule_id": "R-MET-04", "tier": 2, "triggered_by": { "fasting_insulin": 11.4, "hba1c": 5.6 }, "finding": "Subclinical insulin resistance", "evidence_tier": "E1", "interventions": ["resistance_training_3x", "protein_floor", "berberine?"], "rulebook_hash": "sha256:4f2a…c91d" }
Because the output is data, not prose, it can be replayed, diffed across retests, and audited by a clinician.
The berberine? above carries a question mark deliberately, it is a candidate, not yet cleared.
Whether it survives is decided by the safety screener, not the rule.
Agent 07 is a hard gate: no intervention reaches your protocol until it clears every relevant safety check against your specific medications and conditions. When a check fails, the item is blocked with a stated reason , it never silently slips through.
Every supplement is screened against your current prescriptions for known interactions, CYP-pathway effects, and timing conflicts.
Blood-thinner users get omega-3 caps and automatic blocks on curcumin, high-dose vitamin E, ginkgo, and garlic supplements.
eGFR thresholds gate creatine dosing and trigger a nephrology-referral flag below 60, never a blanket recommendation.
Pregnancy, active eating-disorder history without clearance, and type-1 diabetes halt the protocol and route to a physician.
Every run is stamped with a run-manifest: the rulebook version and hash, the model version of each agent, the input file hashes, and a UTC timestamp. Given the same bloodwork and the same rulebook version, the engine produces the same protocol, every time. This is what makes the reasoning auditable rather than vibes-based, and it's what lets a doctor verify a recommendation months later.
// run-manifest.json { "rulebook_version": "v2.4", "rulebook_hash": "sha256:4f2a…c91d", "phenoage_formula": "Levine-2018", "agents": ["normalizer@0.3", "rule-eval@0.3", "…"], "input_hashes": { "bloodwork_pdf": "sha256:…" }, "generated_at_utc": "2026-06-02T09:14:00Z" }
The rules are published. The pipeline is typed. The runs are reproducible. We would rather you, or your physician, interrogate our logic and find it sound than ask you to trust an algorithm you can't see. Open reasoning isn't a marketing posture; it's the only honest way to put AI between a person and their health decisions.
The engine doesn't read "a blood test." It reads a record. Each kind of data is an ingestion module that normalizes into one shared health record, then the same agents reason over whatever is present. Blood is live today. The others plug into the same spine.
50+ markers parsed, unit-normalized, and evaluated against the rulebook. The foundation every record starts from.
HRV, resting heart rate, sleep and glucose trends as continuous context the protocol adapts to.
Read once. Variants that change protocol personalization (B-vitamin form, caffeine, lipid response) enrich every future cycle.
DEXA and coronary-calcium findings parsed into structured fields the engine can reason over alongside blood.
Because everything lands in one record, your protocol is reasoned over your full history, not a single snapshot, and it gets more precise as more modules are switched on. This is why Antiaging Labs is a record and a reasoning engine, not a test.
Clinics, premium gyms, and corporate-health programs can collect data, sometimes with their own scanners, but they have no reasoning layer and no protocol engine. Reports get filed and forgotten.
Long term, Antiaging Labs is the data + reasoning + protocol layer beneath any longevity space: they bring the bodies and the raw data, we provide the record, the reasoning, and the branded report. The consumer program proves the engine; the clinic layer is how it scales. This is the trajectory, stated honestly, not a product on sale today.
Our clinical logic is public and versioned. You can read the exact rule behind any recommendation you receive.
Every protocol traces to a fired rule, a safety decision, and a rulebook hash. Nothing is unsourced.
Each paired Day-1/Day-90 result is structured outcome data. The engine sharpens; the dataset becomes the moat.
Start with a 30-minute consultation. We'll look at whatever bloodwork you already have and show you what the engine sees.
Book a free consultation →