No worries – your data will definitely not be exposed unless you explicitly choose to share it and configure this via the API.
Not speaking for the OP but from my personal experience.
The LLM is capable of looking at and parsing more data points than I am, and quicker. It can surface facts that I may not have considered. For example, it’s pretty good a noticing stuff such as decoupling. I know I can navigate to the web ui to view a graph that will show decoupling but that’s busy work when I can just ask the LLM. I can then start asking it more interesting questions and look for patterns through my historical data.
It’s not a coach, it’s a coach’s assistant. The LLM has an infinite amount of time in their day to parse my data. I could do all of this myself, but I would have to spend hours writing python scripts. That’s the value, it saves me time.
1 Like
Hi, you can find it over here: ChatGPT - Coach GPT for Intervals.icu
4 Likes
Why bother “chatting” with ChatGPT about my yearly plan?
-
Calendar sanity‑check
“I’m eyeing a gran fondo in June—does that still fit after my A‑race in May?”
The LLM quickly sketches how much taper + build time you really have, so you can decide whether to add or drop an event. -
Family‑first reshuffling
Whole‑week view with kids’ holidays ⇒ “Shift my overload block one week earlier, keep long ride ≤ 3 h on Sunday.”
It rewrites the plan in seconds instead of me juggling drag‑and‑drop. -
Smart rest blocks
Ask: “After four weeks at 8 h, 9 h, 10 h, 9 h, where should I insert a low‑stress week?”
It spots the logical deload slot and suggests mileage/power caps. -
4/7 sounding board
When doubts hit at 11 pm, I can type: “Missed Tuesday; double up Wednesday or let it go?” and get a reasoned answer right away.
TL;DR: It’s a fast, always‑awake planning assistant that helps me balance races, rest and family life—while I stay in charge of the final call.
2 Likes
Am I right in thinking that getting this working requires us to create our own GPTs, which in turn needs ChatGPT Plus?
Just found out this thread, which is super relevant now for me since I have, too, created app for using Claude with intervals.icu 
Its been really fun to chat with Claude about my training and it feels like the training plan for me is great.
I made the app with TypeScript and didn’t code one single line, all was generated with Claude. We first chat how I’m feeling and what I want to do, then it fetches my fitness data and all the other relevant metrics, creates a plan for the week and uploads workouts to icu. It has a 3 month training plan configured which it follows and adds any user request (mine) on top of that longer plan.
5 Likes
Would you mind sharing it?
1 Like
Yes I could make the repository public. I might have to clean up the git history first There was some athelte data earlier hard coded. On the other hand, who cares.
You need: npm and claude code installed via npm. I’m not sure if they have free tier at all.
Indeed claude code is not in the free tier 
little use then for me.
1 Like
We might be able to implement something in intervals. Stay tuned!
16 Likes
Please, don’t do this.
As @jamesmmc said:
[Having AI-powered analysis right inside intervals.icu would be] …a reason for me to immediately close my account and cease recommending the site to anyone.
3 Likes
That feature would be opt-in only of course. But currently, we don’t have any exact plans, we just know that training with AI is really helpful.
6 Likes
So is reading a book and writing your own training plan (as a bonus, you actually learn something useful in the process - as opposed to mindless consuming the LLM effusions) - but you don’t seem to consider integrating a bookstore.
There is an API already and everyone who wants to can present their (own) data to every tool they want.
To put it clearly:
If LLM are directly integrated into intervals (opt-in or not, “special” API or not, …) I’m out.
This is my vote against the waste of energy, the questionable (at best) training data collection and the irresponsible use of this technology.
4 Likes
Not everyone has time to read a book and write their own plan…hence why coaches exist. Not everyone can afford to hire a coach and get 3rd party training review, hence why people use LLMs. There’s reason to be cautious and a late adopter to new technology, but blindly opposing it isn’t reasonable.
7 Likes
As a free user I tried pulling my activities, wellness and food macros (from another app) for the last week in, as well as my schedule for the following week, as json files and uploading them into ChatGPT.
It was just comically bad. Fabricating sleep scores and HRV values. Even when I’d corrected it, it was still doing analysis on the values it had previously hallucinated.
Gemini was even worse.
5 Likes
I tried creating a custom GPT a month or so ago but gave up ’cause sometimes it didn’t accept the auth and it really didn’t want to download more than a few activities at a time. Since I’m not good at the scriptIng I gave up. But I again tested with @Jonasdlm code and it works like a charm! Thanks!
But I’m more interested in getting help with analysing my finished activities than my plan forward. Does anyone have a good schema for Activities to add to @Jonasdlm ’s excellent script?
As @Paul_Grave wrote, ChatGPT is great with noticing decoupling and patterns that I can use as a start for my analysis. But as many also noted, you must also take hallucinations and other issues with LLMs into account.
Buenos días. Tengo una duda desde mi poco conocimiento. Estoy intentando utilizar con clave API_Key para ChatGPT y cuando llego a ese apartado el pantallazo que has
puesto le copio la K que encuentro en ajustes de Intervals y me da error dice que la clave api debe de empezar por eyxxxxxxxxxxxx. Me podrías ayudar ? Gracias
Le estas intentando mandarle un token, como dice en la docu
tienes que mandarle donde pone el APIKEY el base64 de este string username:password .
1 Like
Buenas tardes Daniel. Si lo he echo y no hay manera. No tengo mucha experiencia y al final creo que después de intentarlo 100 veces voy a desistir. Intentaba poder usar un Chatgpt personalizado para enviar entrenamientos y luego poder analizarlos y no ve la forma .Muchas gracias por la ayuda
Si me das un poco mas de información te intento ayudar, pero es un poco como comenta vas a
y si tienes dudas para generar el apikey le dices a chatgpt que te haga esto
genera un base64 del siguiente texto API_KEY:password y en password le pasas tu apikey, el apikey esta en ajustes y es algo como
2m8npppppv7awwwoku y lo planchas donde pone [HIDEN], después empiezas una conversación y le pasas el athleteid.
yo demento me funciona lo de arriba pero no me genera bien los entrenamientos, estoy creando mi propio action a ver..