Honest, specific, and within clear limits.
Cadence uses AI to read your plain-language notes, surface gentle patterns, and prepare questions grounded in what you logged. Here is exactly how, and where the guardrails are.
From your note to your summary.
Each step has a specific job, a specific limit, and a check that keeps it inside that limit.
Understanding your logs
Reading messy, human writing.
Surfacing gentle trends
Patterns over time, never alarms.
Preparing questions and summaries
Grounded in what you logged. Nothing invented.
Model-agnostic routing
The best available model, without lock-in.
Strict guardrails
The things Cadence will never do.
Understanding your logs
Reading messy, human writing.
Health notes are rarely clean. People write "feeling off since yesterday" or "BP a bit high, maybe 145?" or just "tired." Cadence uses a large language model to read those notes the way a thoughtful person would: catching the reading buried in a sentence, noticing that "off" followed a medication change, connecting today's note to last Tuesday's.
The model was chosen and evaluated specifically on this task: understanding varied, plain-language health descriptions without requiring the person to learn a format. If a note is ambiguous, Cadence treats it as incomplete rather than guessing at a meaning.
Surfacing gentle trends
Patterns over time, never alarms.
After reading each log, Cadence places it in the context of the past few weeks. It looks for things that appear more than once, or that seem to move together: morning readings that have been a little higher, fatigue that shows up after meals, days when sleep was poor and readings followed.
Cadence does not score or rank or flag urgency. It surfaces calm observations: what it saw more than once, which direction something has been moving. The goal is a picture you can bring to your doctor, not a set of alerts you have to manage.
Preparing questions and summaries
Grounded in what you logged. Nothing invented.
Before each appointment, Cadence prepares one or two questions based on the patterns it noticed. Every question can be traced back to at least two specific log entries. If the AI cannot point to what it is drawing on, it does not offer the question.
The visit summary works the same way. Each item in the summary is sourced from your own notes. Cadence does not extrapolate or fill in gaps. If a week has sparse data, the summary says so. The goal is a document your care team can trust.
Model-agnostic routing
The best available model, without lock-in.
Cadence routes requests through a thin model layer that can work with any capable language model. We evaluate models on this specific task, set the one that performs best as the default, and can switch without disrupting your experience. As models improve, so does Cadence.
This also means we are not betting the product on one provider. If a model has an outage or a regression, we can route around it. You do not notice, and your logs keep working.
Strict guardrails
The things Cadence will never do.
Cadence uses structured output validation on every AI response. Before anything reaches you, it passes through a check that confirms it does not contain a diagnosis, a dosing instruction, or a treatment recommendation. If the model output fails that check, it is discarded and you see nothing.
If your log includes something that sounds urgent, such as chest pain, difficulty breathing, or a very unusual reading, Cadence will tell you to contact your care team or seek emergency care right away. It will not interpret the symptom or suggest a cause. That is always a job for a clinician.
Your logs are used only to serve you. They are not used to train models, not shared with third parties without your consent, and not used for any purpose outside Cadence. See our privacy policy and security page for the full detail.
Care and safety
A full account of what Cadence will and will not do, and how it handles urgent symptoms.
Read the care pageSecurity and privacy
How your health data is stored, encrypted, and kept yours. No third-party sharing without your consent.
Read the security pageCommon questions about the AI.
Try it with your own logs.
Free to start. Write in plain words. See what Cadence makes of them.