A language-model assistant (for example, Claude via API) that operates strictly within the user's role and permissions.
For the trainee
- Answers questions about their training and nutrition from their own plan and data.
- Explains how to perform an exercise and suggests a substitute for an injury or missing equipment.
- Calculates the calories of a meal and suggests swaps that fit their macros.
- Summarises their week: “You completed 4 of 5 workouts; protein was behind on three days.”
- States progress toward the goal and what would improve it most.
- Reminds and motivates.
For the coach and admin
- A summary at the top of each trainee record: compliance, progress, risks.
- Flags struggling trainees and suggests an intervention.
- Suggests plan changes (a draft requiring coach approval).
- Generates a draft program or meal plan from the questionnaire.
- Drafts replies to trainee messages for the coach to review.
- Summarises a new questionnaire before approval.
- Answers operational questions: “Who expired this week?”, “Who is not complying?”
AI governance (important)
- It is not a doctor: a permanent disclaimer is shown; it does not diagnose or prescribe medication.
- On alarming symptoms (chest pain, dizziness, acute injury) or eating-disorder signals, it recommends a specialist and alerts the coach.
- Safety boundaries: a minimum calorie floor, and refusal of extreme diets, steroid doses or stimulant abuse.
- Every suggestion for the coach requires approval before it reaches the trainee.
- Conversations and actions are written to the activity log.
- A daily usage cap per user, per plan, controls cost.
- Unnecessary identifiers are removed before data is sent to the model.
- The admin configures the name, tone, core instructions and knowledge base (the coach's philosophy and methods).