1. Overview

VoxFit today is a logging app: users speak their workouts and meals, AI structures the data, and the app shows history (streaks, weekly volume, macro tracking). What it doesn't do yet is close the loop in either direction:

• Forward — after logging weeks of sessions, the user still has to decide what to train next themselves. All that recorded history — goal, frequency, exercises, progression — sits unused for anything but display.
• Backward — the app captures mood, energy, and physical flags (things the user mentions mid-session, like "knee felt off") per session, but never synthesizes them across sessions. A pattern mentioned four times in six weeks is invisible unless someone manually scrolls history.

This PRD covers two features that close those loops:

1. AI Workout Plan Generator — proposes a forward-looking training plan grounded in the user's goal and actual recent training history.
2. AI Progress & Wellness Coach — reviews a recent window of activity and reflects back trends and recurring notes, in the same calm "coach" voice the app already uses.

Both are AI-generated, both are on-demand and single-shot (tap a button → one generation → review a result card → keep or discard), matching every AI feature already shipped in VoxFit (extract-workout, suggest-diet-meals, log-food). Neither introduces a chat interface.

2. Goals & Non-Goals

Goals

• Give users a personalized forward training plan derived from their real goal + real training history, not a generic template.
• Give users a calm, pattern-based reflection on how they're actually trending — training, nutrition, and recurring notes — without requiring them to piece it together themselves.
• Reuse the app's existing AI-call conventions, data, and visual "coach" voice rather than introducing a parallel pattern.

Non-goals (v1)

• No chat or conversational agent UI. Both features are single-shot generate-and-review, like everything else in the app today.
• No medical diagnosis, clinical advice, or "healthcare expert" framing (see §4 for why this matters).
• No automatic or scheduled generation — the app has no cron/scheduled-function infrastructure today, and adding it is out of scope for v1.
• No plan-vs-actual adherence tracking (matching logged sessions against a planned session) — real feature in itself, deferred.
• No editable plans (edit an individual day/exercise within a generated plan) — v1 plans are regenerate-whole-or-keep.