Ask your coach.

A few real patterns we see:

• “Why was today’s interval session so hard?”
• “Can I do a long ride tomorrow if I swam 4k today?”
• “My race is in 3 weeks — how should I taper?”
• “Is my Saturday long run progression safe?”
• “Should I skip Thursday’s threshold or push through?”
• “Why is my HR higher than usual at the same pace?”

Every answer grounds in your data. If the coach doesn’t have the information to answer honestly, it says so and tells you what’s missing. No hallucinated numbers.

The chat context is rebuilt for every turn:

• Your athlete profile — FTP, threshold HR, threshold pace, weight, goal race, experience level, injury history, training preferences.
• Current fitness state — CTL, ATL, TSB, activities in the last 7 and 14 days, weekly zone mix.
• Detected issues from the rule engine (the same one that drives the weekly review).
• Your last workout-by-workout feedback — “too hard”, “too easy”, “just right”, skipped — with your own notes.
• Snippets from your three most recent chat threads, so follow-up questions a week later still have context.
• On-demand tools: it can fetch the last 14 days of activities, pull the detail of a specific day, compute a weekly breakdown, or re-read the most recent weekly review — whichever the question needs.

The bigger, static context — coaching rules, zones reference, your profile — is cached on Anthropic for an hour, so repeat questions in a session stay fast and cheap.

Every AI coach has to be honest about its limits. Ours declines entire categories of question by design:

• No medical advice. Injury diagnosis, medication, dosing, persistent pain → see a sports medicine professional.
• No nutrition prescriptions.We’ll mention fueling during long sessions if relevant, but we won’t write a meal plan.
• No rigid 16-week plans.We prescribe 1–4 week blocks because that’s the horizon on which your data gives us signal.
• No magic predictions.If the coach says you’re ready for a 40-minute 10k, it’ll show the recent workouts that support the claim.