Request preview accessHEALTH CONTEXT FOR PRODUCTS + AI
Knownset connects approved health and fitness data, turns it into clear answers with sources, and keeps people in control of every change.
01 / SEE IT IN A PRODUCT
Northstar is a made-up fitness app. Follow one answer from approved data, through a suggested workout change, to the member's final decision.
MADE-UP APP + DATA / FOR DEMONSTRATIONTwo current sources. No missing value was guessed.
Only the records needed for this question are used.
Tested code checks workload, personal ranges, and data age.
Only relevant goals and recent decisions carry into this answer.
The checked facts stay separate from the coaching explanation.
The product shows a reversible change that keeps the original goal.
The method, language, evidence, and permissions are checked.
The member decides. Their product owns the final update.
Lower sleep duration and higher reported fatigue support a smaller volume adjustment—not a different training goal.
Can we keep the lower-body session, but take a little volume out? My back feels tight.
Yes. Keep the session intent and target effort. Reduce front-squat work from four sets to three and split-squat work from three sets to two.
02 / WHY KNOWNSET
Knownset works behind the scenes. It helps health, fitness, and coaching products use approved data without taking over their customer experience.
03 / HOW IT WORKS
The person chooses which health and fitness data this product may use.
Knownset lines up units, dates, and sources so they can be compared safely.
Tested rules calculate the facts before any AI-generated explanation is allowed.
The product receives a small next step with the reason and sources attached.
Nothing changes until a person reviews it and the product confirms the update.
Product flow ready
04 / HOW ANSWERS STAY RELIABLE
This demo uses tested rules—not a language model—to calculate results. Any future AI-written explanation would still have to use those checked facts, show its evidence, and wait for approval before anything changes.
Only the goals, training, recovery, and health information approved for this person and product enter the request.
Tested code calculates the numbers, checks how current the data is, and flags gaps. AI language cannot invent the result.
The product can explain what the facts mean and offer a small next step with the supporting evidence attached.
The person sees the proposed change before it happens. Their product—not an AI model—makes the final update.
Answer flow ready
05 / WHAT PRODUCTS CAN DO
The current sample uses fixed rules and made-up data to produce a small, reversible suggestion. It does not call a language model.
The status on each tab tells you what works in the current sample, what exists in the core service, and what is not available yet.
06 / THE ANSWER RECEIPT
See what data was allowed, what the rules found, what the product may suggest, and who must approve it. The receipt keeps those steps separate and easy to inspect.
Today’s summary uses sleep, recent training, and a member check-in. Other records stay out of this request.
07 / FOR DEVELOPERS
Choose one of four working read-only tools. The demo shows the request, the safety checks, the small result, and the sources that come back with it.
MCP is a standard way for an AI product to call a tool. This example is read-only.
The sample returns a readiness summary for one day without exposing raw provider files.
{
"name": "knownset_get_health_summary",
"arguments": {
"date": "2026-08-08"
}
}Confirm the app and user are allowed to make this read.
Select only the records the person agreed to share.
Keep the evidence and known limits with the answer.
Send only what this request needs. Nothing is changed.
{
"status": "checking request"
}Knownset is checking the request before returning any data.
Step 1: Check access