A coach that adapts to your reality.
A real training plan with a long-term shape and sessions detailed down to the interval. You turn up and run them. Runsei watches how they actually land — in your body and in your week — and keeps the plan correct.
Coming to iPhone, Android and the web · English, Danish and Spanish
Your HRV came in 13% below your own 30-day normal and your resting HR is up 5 bpm — together, that’s the clearest sign yesterday’s session hasn’t been absorbed yet. Let’s keep today easy and move the intervals to Thursday.
Easy 6 km today · intervals moved to Thu
Health data you connect is stored in the EU and is never used to train anyone’s models.
No connected account needed — Runsei takes whatever you’ve got.

Built by a runner who needed it.
I’m Andreas — an engineer and a runner, training around a full work and family life in Denmark. I was also born with a club foot, so adjusting has never been optional for me: the asymmetry brings injuries, small kids bring weeks of illness, and every serious plan I followed broke the same way — life intervened, and the plan didn’t care.
I’ve tried most of the running apps, and each fell short somewhere. A plain AI chat falls shortest: it can’t hold a plan’s structure or stay consistent from one answer to the next. Runsei is the coach I wanted instead — it keeps your plan as real structure, watches how the training actually goes, asks what happened, and reshapes it around reality. Conservative by default, honest when it matters. It proposes; you decide.
Andreas Schmidt · Founder, Improva Labs · sub-35 10K
A plan with a shape — and sessions you can just go and run.
The point of a coach is that you stop designing your own training. Runsei holds the long-term structure, writes the sessions in full, and it’s on you to turn up. Then it checks whether the plan is actually landing — and fixes the plan, not you.
Phases, not a pile of runs
Your block has a shape you agree up front — what each phase is for, how long it runs, where the race sits. Individual sessions are written about a week before their phase begins, so they’re built on the runner you are by then, not the one you were at signup.
How your plan is built →You always know what you’re in the middle of and what’s coming. Nothing is written so far ahead that it stops fitting you.
Detailed down to the interval
Not “tempo run, 45 min.” Every step, every target, every recovery — the same structure your watch can execute, because the workout file and the screen are built from one model. Turn up, press start, run it.
Sending a session to your watch →Check adherence. Runsei fixes the plan.
One word tells you whether the block is delivering what it was written to deliver — and opening it shows exactly which sessions moved it, worst first, each one reaching the run behind it. If the answer is “not really”, that’s a conversation about the plan, not a verdict about you.
How adherence is judged →The work is landing where it was meant to. A session or two drifted, which is normal.
One session was cut short; the other two were run too hard, not too easy — a sign the paces are pitched slightly low. So the coach raises them, rather than telling you to push more.
Every number leads to the method behind it.
Nothing here is a generic formula applied to you — it’s computed from your own runs, and each one opens onto how it was worked out.
The Execution Score
A session is graded on fidelity to what it was for — not on how fast you ran it. Hold threshold effort up a hill and the slower pace still scores as on-stimulus. It’s deterministic and published: the same run always scores the same, and when there isn’t enough data to grade it honestly, you get no score rather than an invented one.
How the score is built →The outlined bar — Thu, 6×1 km, hilly: pace 14 s/km slower than target, heart rate on target. Scored 94 · excellent. Slowing to hold effort on a climb is the correct response, not a miss.
Training load, read as a shape
Your 42-day Base is what your body is prepared for; your 7-day Recent is what you’ve asked of it lately. What matters isn’t a high number — it’s a sudden one. Load that rose week over week reads as a controlled build, not as ramping too fast.
How load is computed →Building steadily — Recent sits above Base, but it got there over ten weeks. No spike.
Aerobic efficiency
The honest test of aerobic fitness isn’t a hard day — it’s what an easy day costs you. Runsei tracks how much ground you cover per heartbeat on aerobic runs. Every easy run is a data point, so unlike a race-time estimate the line actually trends — smoothed weekly, so one hot or hilly run can’t move it.
How efficiency is tracked →Your aerobic engine is improving — the same effort now covers more ground per beat, about 11 s/km faster at the same heart rate than ten weeks ago.
Critical Speed, not a formula
Across your genuinely hard efforts, distance rises almost linearly with time. The slope is your Critical Speed — the fastest pace you can hold in a true steady state — and it predicts any distance. Every prediction carries a confidence, and anything beyond the efforts you’ve actually done is labelled as extrapolation rather than sold as a promise.
The critical-power model →| Distance | Predicted | Confidence |
|---|---|---|
| 5 km | 16:52 | High |
| 10 km | 34:22 | High |
| Half | 1:13:13 | Indicative · extrapolation |
| Marathon | 2:27:03 | Indicative · a ceiling, not a promise |
Are you on track for race day?
The graph you asked for: last race → today → target, with the projection band and a green / amber / red read on whether the current trajectory reaches the goal. It needs the trend-projection model and a target-race field before it can be honest, so the homepage holds the space and ships it when it’s real — never as decoration.
Measured against you, never a population.
Every recovery signal is compared with your own normal — a 30-day median that deliberately excludes the last week, so a hard block can’t quietly redefine what “normal” means for you. There is no such thing as a good HRV number; there is only your trend.
Down the day after a hard session is expected — that’s training working. Down for several days running is the load outweighing the recovery.
The most robust of the signals, and the one that often notices an illness a day before you do. Paired with a suppressed HRV, it’s the pattern worth acting on.
Duration isn’t the whole measure. How much your nights vary has its own effect, so Runsei names a wide spread even when the average looks fine.
Wear something overnight and these arrive on their own — the morning check-in then asks only the one thing no watch can measure: how you feel. Tell us you don’t, and it never asks you for a number again.
The parts that make it usable on a Tuesday.
Your plan
- Built around your goal, your days and your real paces
- Import a plan you already have — a photo or a PDF
- Phases you can see, and a goal you can ask to review
- Every change logged, traceable and reversible in one tap
On the run
- Structured sessions — warm-up, reps, targets — not paragraphs
- Export any session as a .fit workout your watch steps you through
- Subscribe to your plan from Apple, Google or Outlook Calendar
- Weather on the session, and factored into how the run is read
Talking to your coach
- A real conversation — it plans, explains and logs
- Talk instead of typing; send a photo or a PDF
- Injuries and constraints it remembers and trains around
- Notifications you control, one line at a time
Your data
- Stored in the EU, with an EU-only mode for the AI
- Never used to train anyone’s models
- Export everything Runsei holds, as one file
- Disconnect any source at any time — every connection is reversible
Grounded in published research — and honest about how strong it is.
Every method Runsei uses is explained in plain language and cites what it rests on. Where the evidence is modest, we say so rather than dressing it up.
Read the full methodology — 26 articles, the same ones your coach reads →
Join the waiting list.
Runsei is in private testing with a small group. Join the list to hear the moment we launch — and raise your hand if you’d like to be one of the first beta testers.