AI next-day planning
An AI coach with hard safety rails and receipts.
Skeptical of AI fitness advice? Good. So are we. That's why the parts of this product that keep you safe are deterministic code, the parts that plan your training are transparent rules over your own data, and the AI writes explanations, with citations, rather than making judgment calls.
01Next-day planning
Regenerated every night, from everything you did.
Tomorrow's session starts from a template for your goal and equipment, then gets adjusted by your readiness level, your workout feedback, a set-level snapshot of what you lifted, your recovery picture, and your nutrition, protein target included.
Citations on everything. Plans and chat answers cite an 855-chunk knowledge base built from a real strength & conditioning reference text. When the app says "back off," you can read why.
A coach you can question. Ask why the plan changed, what to do about a niggle, how to handle a missed week. The chat knows your goals, readiness, recent logs, and nutrition, and answers with sources.
Honest about the AI split. Readiness scoring and safety checks are deterministic: same inputs, same outputs, testable. The language model writes conversational replies, and if it ever fails, a deterministic answer steps in.
Provenance on every plan. Each plan is labeled with where it came from: an AI starter plan from onboarding, a self-guided AI plan, or one you adjusted. And when readiness is estimated from defaults instead of measured data, the plan says so rather than faking confidence.
Adjust it on demand. Not feeling tomorrow's session? Regenerate it whenever you like, optionally around a focus: a body area, lighter or harder, shorter or longer. If nothing needed to change, it tells you 'already up to date' rather than claiming success.
TomorrowLower body · readiness-adjusted
- Warm-upHip openers + ramp-up · 8 min
- MainSquat 3×5 @ RPE 7 · RDL 3×8 · Split squat 2×10
- CooldownBreathing + quad stretch · 5 min
Volume −15% · short sleep two nights running
Squat load held · HRV back at baseline
Why this plan · 3 citations
Illustrative product UI
02Safety
Five gates the AI can't be talked out of.
Before any AI writes a word, deterministic rules check the message first. They're sourced to AHA and NEDA guidance, covered by positive and negative tests, and a flagged message never reaches the model at all.
Emergency symptoms
Chest pain, fainting, or similar language stops the workout and points to emergency care. Always.
Pain & injury
A pain flag blocks progression on that movement pattern and swaps in technique and mobility work.
Eating-disorder risk
Restrictive or compensatory language is never echoed back. The app answers with support and a referral instead.
Supplements
Education only. The app never doses, prescribes, or sells.
Readiness downshift
When recovery data says back off, volume and intensity come down. There's no motivational override.
Safety decisions are deterministic code, not model output. The AI is never asked to judge whether chest pain matters.
Halden provides general fitness guidance, not medical advice. If something hurts, or a health question is on your mind, talk to a qualified professional.
Coaching you can cross-examine.
iPhone, iPad + web today, via early access. Android coming. Not yet on the App Store.