Ask Koutsi

Everywhere else in openkoutsi, Koutsi decides what to tell you: the daily card on your dashboard, the analysis under a ride, the guidance on a goal. Ask Koutsi turns that around. You get a text box, and Koutsi answers by going and looking at your actual training data.

That last part is the point. Koutsi can read your activities, your plan, your goals, your zones and your fitness numbers, so it can answer things a general chatbot cannot:

• "I'm three weeks out from my event and my form is still negative — should I be worried?"
• "Work is going to eat next week. What should I cut?"
• "My threshold intervals keep falling apart in the last rep. What do I do about that?"
• "Should I be doing more Z2 or more intensity right now?"

Turning it on

Ask Koutsi appears in the sidebar once you switch on Profile → Analysis → Let Koutsi look things up.

It shares that switch with the agentic daily card rather than having its own, because the two need the same thing: permission for Koutsi to look data up instead of being handed a fixed summary. A chat that cannot see your training would just be a general-purpose chatbot in a cycling jersey.

Your AI model has to support tool calling

Looking things up means calling tools, and not every model can. If the model you have selected cannot, the page will tell you so and point you at your AI settings rather than letting you write a question that was never going to be answered. See Using your own AI model.

What Koutsi will and won't discuss

Koutsi is a cycling coach, and it stays one. Four kinds of question get four different treatments:

You ask about	What Koutsi does
Coaching — intervals, periodisation, fitness and form, pacing, adapting your plan	Answers fully. This is the job.
Around the edges — fuelling for a long ride, sleep, strength work, bike fit, race tactics	Answers as a coach would: practical and general, without claiming to be a specialist.
Medical — symptoms, pain, injury, illness, medication, or whether something your body is doing is dangerous	Does not answer. Points you at a doctor.
Unrelated — anything that isn't your training	Declines briefly and steers back.

The medical line

This is worth stating plainly, because openkoutsi holds heart rate, weight and effort data, and that makes it very easy for an AI to sound authoritative about your health.

Koutsi will not tell you whether a symptom is serious, will not guess at a diagnosis, and will not tell you to train through something. If you ask about chest pain, a joint that keeps swelling, a resting heart rate that worries you, or how far to cut your food before a race, it will say that this is outside what it can help with and that you should speak to a doctor or another qualified clinician.

That is not Koutsi being unhelpful. It genuinely does not know, and neither does the language model behind it — it has your power files, not your medical history.

If something feels wrong, ask a person

A chat window is the wrong place for a health question, however convenient it is. Talk to a doctor.

The other half matters too: asking what to eat on a four-hour ride is a question for your coach, and Koutsi should answer it. If it refuses something that is plainly a coaching question, that is a bug worth reporting.

Having a conversation

Type a question and press Enter. Koutsi may take a moment — it is often making several lookups before it writes anything, and it tells you what it is doing while it works ("Koutsi is checking your power curve…").

Each lookup stays in the conversation as a short line — "Your power curve", "Your recent sessions" — in the order Koutsi made them, just above the answer they fed. They appear one at a time as it works, so you can see how far along it is rather than watching a spinner, and they stay afterwards as a record of what the answer was actually based on.

The lookups are names, not data

A step tells you that Koutsi read your power curve, never what it found. The answer is where the findings are.

Answers stream in as they are written, and they are saved as they go. You can reload the page, close the tab, or pick the conversation up on your phone; the answer carries on and is there when you come back.

Follow-ups work as you would expect — Koutsi remembers what you have both said in this conversation. What it does not keep is the data it looked up: each turn it goes and reads your training fresh. That is deliberate, and it means an answer tomorrow reflects tomorrow's rides rather than a stale copy.

Conversations

Every conversation is kept in the sidebar, named after your opening question, and stays until you delete it. Deleting one removes everything in it, including what you wrote, and cannot be undone.

There is a limit on how long a single conversation can get. When you reach it, start a new one — you lose nothing by doing so, because Koutsi looks your training up fresh each time anyway.

Koutsi offers; you decide

Koutsi can offer you a training plan, or a change to one, and nothing more.

Ask it for one — "could you build me eight weeks for a hilly gran fondo in October?", "push the whole thing back a week", "make Thursday easier" — and it drafts what it is suggesting and shows it to you as a card under its reply, with Yes and No. Nothing has happened to your training at that point. Your Yes is the only thing that changes anything, and if you never answer, nothing ever does.

A few things worth knowing about the card:

• It tells you what accepting would archive. A new plan files away any plan you are following whose dates overlap it, so the card names those plans before you click rather than after. Nothing is lost — you can bring an archived plan back from the plan page — but you should know it is going to happen.
• Drafting takes a moment. Koutsi writes the weeks properly rather than sketching them, so expect a few seconds. Occasionally it cannot reach the AI model for that part and openkoutsi's own planner writes the weeks instead; the card says so when that happens.
• An offer stands for a day. Sleep on it if you want. After 24 hours it lapses and you ask again.
• You can carry on talking first. Asking a follow-up question does not cancel an offer — you can go back and accept it afterwards. A new offer replaces the old one.
• The answer is a button, never typing "yes". That is deliberate: the decision has to sit outside the conversation, so nothing said inside it can be mistaken for your consent.

If your training has moved on since Koutsi drafted something — you created another plan, the start date has passed, the session it was going to change has been done — accepting is refused rather than applied, and Koutsi will draft a fresh one when you ask.

Everything else is still advice. Koutsi cannot mark a workout done, cannot link a ride to a planned session, and cannot rebuild the rest of a plan's weeks. When an answer is about your plan there is a link to open it, so you can make any other change yourself.

Limits

Ask Koutsi is the one AI feature you can trigger as often as you like, and each question costs several AI calls rather than one. So there is a daily limit on questions. You will not normally notice it; when you are close, the page starts showing how many you have left.

On a server where AI features need a subscription, chat is covered by the same subscription as everything else — see AI features, subscriptions and BYOK.

If something goes wrong

Because Koutsi has to look things up to answer, a chat question cannot fall back to a simpler answer the way the daily card can. So when a question fails, it says why:

• Koutsi was busy — your server was working on other AI jobs and did not get to yours in time. Try again in a moment.
• Your model can't use tools — a permanent limitation of the model you have selected, not a temporary glitch. Choose a different one.
• The AI server had a problem or couldn't be reached — worth another try; if it keeps happening, check the server in your AI settings.

Anything that can sensibly be retried offers a Try again button.

Where your words go

What you type is sent to whichever AI model your server is configured to use, or to your own if you are running one. Your conversations are stored in your own database alongside the rest of your data, are included in your data export as chat.json, and are deleted with your account.

If you use your own model (BYOK), what that model does with your messages is between you and it — see Your data & AI and Privacy & consent.