Last updated: July 2026 · LiftWeight 1.0
LiftWeight’s plans, progression suggestions, and stall/revision flags come from Coach — a rules-based training system we built, not a machine-learning model and not a general-purpose AI. Give it the same inputs twice — same goal, same logged workouts, same equipment — and it produces the exact same plan or suggestion, every time. Every rule is code we can point to, version, and test; nothing is learned from a training corpus or hidden from us.
Natural-language AI shows up in three places, and none of them can change what’s actually in your plan. First, the short “why this plan” summary shown alongside a generated plan may occasionally be reworded by a language model purely so it reads more naturally as a sentence. That model is never shown your name, your raw workout history, or your body metrics — only the plan’s own already-decided summary and the list of constraints applied (for example, “Equipment available: barbell, dumbbell”) — and it is never allowed to add, remove, or change an exercise, a set, a rep, or a weight. If it’s unavailable, you see the same plain-English explanation the rules engine always produces, with nothing missing.
Second, once the rules engine has already built your plan, a separate “Coach check” pass — an independent sanity check from a stronger model — reads a compact summary of your goal, experience, and constraints and may surface a short advisory note worth a second look, for example a potential conflict between two things you told it. It never sees the generated plan itself, so it has no way to edit it; if it has nothing to flag, or the check isn’t available, you simply don’t see a note.
Third, after you finish a workout, a short “Coach’s note” may appear on your workout summary, generated from that workout’s own stats — sets completed, volume, any new bests. It’s written after the fact, from a workout that’s already done, so it can’t influence what you were asked to do.
None of these three calls ever decide what exercises, sets, reps, or weights are in your plan — that’s still 100% the deterministic rules described above — and none of them is used to make an automated decision that affects your training program.
Depending on which suggestion it’s making, Coach reads:
It does not read your name, email, progress photos, or your broader body-measurement history over time — that lives in Progress and never feeds a Coach decision.
It never changes a saved plan by itself — every revision it proposes is something you explicitly accept, edit, or dismiss. It isn’t a medical device and doesn’t diagnose anything; its suggestions are training heuristics, not clinical judgment.
You can export everything Coach has used about you — plans, sessions, sets — from Settings → Your data → Export. If you’d rather not have Coach generate a plan for you at all, you can start from a template, import a plan from text, or build one entirely by hand instead — Coach is one way to create a plan, not the only one.