GroundStateAI ID · the PrePrompt Method
Your AI already knows what you told it. It's time it knew who you are.
GroundState collects your DNA, bloodwork, the drugs and supplements you take, personality, wearable data and biography into one encrypted profile on your own machine. Before any AI sees a thing, you approve the smallest slice it needs.
Software first · No hardware required · Every finding labelled for evidence · Don't go unprompted
The gap
Same question. A different answer.
A lab result comes back with a high creatinine. Here is what an AI says without your data, and with one small slice of it that you approved.
“My creatinine came back high. Should I be worried about my kidneys?”
without your data
A high creatinine can be a sign of reduced kidney function. Dehydration, certain medications and kidney disease are common causes. Please see your doctor for follow-up testing.
Reasonable, and the same answer everyone gets.
with one approved slice
scope: substances · creatine, 5 g/day, current
Your reading may be overstated. You take creatine, which raises serum creatinine without changing how well your kidneys work. Ask for cystatin C alongside creatinine, and tell your doctor about the supplement.
●●●● strongMeta-analysis behind this
Both answers are illustrative and written for this page. They are not a measurement; our time-and-accuracy benefits remain assumptions until a pilot measures them.
The PrePrompt Method
Collect first. Prompt second.
Five layers of personal data, stored encrypted on your machine and cross-referenced by a rule set drawn from published research. Each rule is labelled for how strong its evidence is, so a finding no single layer could give you arrives with its limits attached.
01
Genetic
What your body can handle. Consumer genotype or whole genome.
02
Bloodwork + substances
What it's doing now, plus the drugs and supplements that change how a lab value should be read.
03
Personality
How you process and what motivates you.
04
Wearable
What your body is doing day to day. Sleep, training load, recovery.
05
Biographical
Values, relationships, the shape of your life.
1 · Collect
Import what you already have. Add the supplements and drugs you take, with dose and status.
2 · Store
Encrypted in a vault on your own machine. You hold the key.
3 · Check
The rules run locally and produce flags, each with an evidence tier and a source.
4 · Approve & prompt
You pick the smallest slice, read it, approve it, then prompt any AI.
Private by design
Your profile is not a file an AI can just read.
Most health-and-AI tools ask you to hand over the data first. GroundState keeps it in an encrypted local vault and releases only what you approve, one scope at a time.
Encrypted at rest
A vault on your machine
AES-256 encryption with the key held in your system keychain. Nothing is stored in a place an AI session reads by default.
Scoped release
The smallest slice that does the job
Ask for communication, training, sleep, nutrition, substances, labs or derived flags. Each release needs a typed approval from you.
You approve, not the AI
An assistant can't approve itself
Release needs a human at a terminal. An agent working on your behalf can ask, but it can't answer for you.
Tamper-evident log
A receipt for every disclosure
Every release is recorded in a hash-chained audit log, so an edited or deleted entry shows up when you verify the chain.
the honest part
A slice you approve is a disclosure. Once an AI provider receives it, we can't take it back, which is why the slice is small and you read it first. Hardware custody (Hito) is an optional upgrade we're planning; today the key lives in your system keychain.
Interactive demo
See it work, with a persona who isn't you.
A working prototype built from the real check code, run in your browser on a synthetic persona, “Alex”. Every number is made up, and the demo page makes no network requests.
Six scenes
1The gap. Same question, two answers.
2Your vault. Encrypted, with a tamper-evident log you can try to break.
3The check. Every finding says how strong it is.
4Your supplements. What they do to your lab results.
5For fun: astrology. With the receipts.
6What this is. Real, approximate, not yet.
In scene 4
Type your own supplements. See which lab results they can shift.
The supplement checker runs entirely in the page. Add creatine, biotin, vitamin D, a statin or other drugs and supplements, and it lists the lab values to read with care, each with its tier.
Open the interactive demo
Synthetic persona · no sign-up
Evidence
We label our confidence. Most don't.
Every finding carries an evidence tier, set by the weakest input it depends on. If a rule combines a strong lab finding with a contested personality claim, the finding is graded Contested. Some inputs aren't graded yet; we say so rather than invent a tier.
And a caution we repeat: no study has tested your exact combination of genes, labs, supplements and habits. Each rule is real research applied to a person it was never run on. You decide how much weight to give it.
61 rules today: 47 across the five layers and 14 about substances. The count grows as the research does.
strong
Meta-analysis or clinical guideline behind it
moderate
Real research, modest or mixed effect size
contested
Real studies, genuine disagreement
unestablished
No research yet, or not an empirical claim by design. You can hide these with one switch.
non-scientific
Astrology mode is for fun. It is always labelled, it never feeds a rule, and a fair test of it is part of the demo.
Who benefits
Not just a better answer for you
You
The aim is time and accuracy: fewer round-trips to a usable answer, built from your own data instead of a generic guess.
The LLM
Constrained, high-signal input instead of a noisy conversation history. No retraining required.
The environment
Fewer wasted exploratory turns would mean less compute. A hypothesis a pilot has to confirm.
Privacy-minded people
You choose the slice, read exactly what goes out, and keep a local receipt. Useful without handing over everything.
Where we are
A working prototype, labelled honestly.
We would rather tell you what isn't done than have you find out.
real today
- An encrypted local vault with scoped, approved release and a tamper-evident audit log.
- A local check that runs the rules on your machine and prints only flags.
- Every finding carries an evidence tier and a source you can open.
- A confounder gate: drugs and supplements that change how a lab value should be read.
approximate
- The rule thresholds are assumptions we've set, not clinical cut-offs.
- Many inputs are still ungraded; we don't invent a tier for them.
- MBTI, Enneagram and self-report rules stay Unestablished by design.
- Intake doesn't collect smoking, shift work, caregiving or social connection yet.
not yet
- No pilot, so no measured outcomes.
- Hardware custody (Hito) is a plan, not a shipped feature.
- EU health-data and medical-device review hasn't been done.
- None of this is medical advice or a diagnosis.
Planned pricing · not yet on sale
One-time purchases, planned at every tier
Figures below are our current plan and may change before launch. No subscription trap; hardware custody is something you can add, never a gate.
Starter
$29 one-time
Personality layer (MBTI, Enneagram, Big Five) plus your structured profile.
Standard
$79 one-time
Starter, plus DNA analysis and a genotype-informed blood panel requisition.
PLANNED FLAGSHIP
Complete
$149 one-time
Bloodwork and substances analysis, wearable integration, and the full cross-layer check.
Ongoing
$9 /month
Annual re-analysis, monthly wearable trend reports, updated findings as research and your data change.
+ optional hardware custodyplanned add-on, any tier, any time$79
Don't go unprompted.
Join the waitlist and be first in line when GroundState opens up.
groundstate.me · No confirmation email. One email when we launch.