Reading your data · 29 October 2026

Fitness, fatigue and wellness
in Intervals.icu, read with AI.

"Am I overreaching?" is a question about several records at once: how much you trained, the fitness and fatigue values Intervals.icu stored on your rides, and what you logged about sleep, resting heart rate, HRV and how tired you felt. An assistant connected through IntervalSync can read all of them and set them side by side by date. It cannot tell you why a number moved.

This is a way to read your own records, not medical advice and not a training prescription. IntervalSync supplies the records; your assistant writes the interpretation, and IntervalSync does not check it.

Where each number comes from

These are all read tools and change nothing; plans and daily limits are on the MCP server page.

Each tool needs its Intervals.icu permission (Activity, Wellness, or Settings for the profile); the support page explains refreshing one that was left out.

Line them up by date

A wellness record is identified by its date, and an activity carries start_date_local, in your own timezone. A ride that starts at nine in the evening can sit on a different day from the HRV you logged the next morning, so ask your assistant which record it paired with which ride. Expect gaps in both directions: a rest day can have a wellness record but no activity, so no fitness or fatigue value, and a ride on a day you logged nothing has the opposite. Nothing in the tools fills either gap.

The catalogue list_activity_fields returns has no field called form. The event-readiness review mentions form but names only icu_ctl and icu_atl for it, and says they can come from get_wellness too; the code's field lists name them only as activity fields. A form figure is your assistant's own arithmetic, and it should show which values it used.

A missing day is missing

The weekly review makes this point too. A wellness field you did not log is left out rather than returned as null, and IntervalSync treats the two as the same: not recorded. A day with no record at all is not in the list, and pagination.total_days counts the records returned, so nine records across a fourteen-day range means five days have none. Neither is evidence of a normal night or a normal resting heart rate.

The values are the ones that were logged, and the tools do not say whether you typed one, a sensor synced it, or an update you confirmed wrote it. update_wellness takes a fatigue score as a whole number from 1 to 10, and nothing in the code says which end is tired; it takes HRV in milliseconds, with no word on how it was measured. Ask which direction and source the assistant assumed.

A prompt that reads and compares

Replace the placeholders with two dates. It asks for records and gaps, not a verdict.

Ask
Read my training and wellness records for [START DATE] through [END DATE], and for the same number of days before that. From list_activities, take each activity's date, duration, training load and the fitness and fatigue values Intervals.icu stored on it (icu_ctl and icu_atl). From get_wellness, take whatever sleep, resting heart rate, HRV, fatigue score and notes I logged, and tell me the field names that came back. Put it all in one table by date, newest first. Write "no record" for a date with no wellness record and "not logged" for a field I left empty; do not fill either in or call it normal. State both date windows and how many days in each have a wellness record. Separate what the records show from possible explanations, and say what they cannot tell me. If you calculate anything the tools did not return, show the values you used. Use read tools only and leave my calendar and wellness records unchanged.

It uses two read tools and writes nothing. A wellness value changes only through update_wellness, which refuses to run without your confirmation of the exact values.

Built-in reviews that use these records

get_review_protocol lists six reviews and returns the steps of one: what to look at and which tools to call. It reads nothing from Intervals.icu.

What the records cannot tell you

Not medical or coaching advice. A low HRV or a high resting heart rate does not diagnose anything, and illness and injury are not visible in training data. If something feels wrong, ask a doctor or your coach.

To connect your assistant, follow the ChatGPT, Claude, Gemini or Grok guide. The prompts page has a shorter readiness prompt, your AI as a training analyst has more questions, and the same interval across weeks follows one session.


Every tool your assistant can use · More prompts · All articles · Connect your training