MyTrainPal - AI Assistant connected to Intervals

[Krzysztof]

You raised a really interesting point about what an app can offer alongside the Compare Activities tool in Intervals.icu.

The core philosophy behind how I designed MyTrainPal was to bring the interaction as close as possible to talking to a real coach. Imagine you’re reviewing two rides with a human coach — you’d look at the charts together, you’d point at a specific interval or time window and ask a question, and the coach would explain why your metrics look like that and what it reveals about your physiology.

This is exactly what you can do now with MyTrainPal. Ask it to compare any two activities in natural language — it pulls the data and shows you comparison charts. Then you select a specific window of interest and click “Send to coach.” The AI coach sees exactly what you’re looking at and can reason about that specific data in detail.

[David_Bannister]

This looks excellent though I am unsure about the value necessarily of comparing two very different rides for efficiency particlualrly. But it will allow sets of intervals to compared very well for sure. And with the W’ & duration curve exposed to the AI it should also allow a view on whether the user performed well or more poorly than would be expected.

That could then lead to a suggestion about what ability area to focus on for the next training block perhaps. These are a couple of sites that aim to show that strength/deficiency focus, examples below with my own metrics inserted.

I am currently using Ride Cave as my workout execution platform (as it can show live ventilation, Pwr:Hr/decoupling & live W’ Bal set against one another) & a variety of AI, including yours, for analysis & suggestions. What I am hoping is that rather than having a standard periodised plan e.g. 12wks leading to a race, I can get the AI to do me a short term plan that gives me a suggested workout to enhance the areas where I have the best chance of improvement. So hopefully it can give me a suggested route that is avaialble on RC for example (though it could be Zwift, Rouvy or MyWhoosh if using those platforms (with an appropriate intensity goal), a structured workout (again any platform) or a track race, currently Ride Cave only I think (with appropriate difficulty level, & lapnumber/elimination settings) so that I can have a choice of types of experiences that produce the same physiological result but keep training fresher than just e.g. a structured ride. So like Athletica’s Workout Wizard but different workout types as well as structures. So for example it might say that I could do a VO2 Max interval workout (& parse the suitable structure), ride Xterra-type track Weston Park (lots of short inclines minimnicking VO2 intervals) aiming for a avHR of xx bpm or do an Elimination Race set to Medium (so bots are chosen around my FTP) with riders eliminated every 3 lap (so for me that is roughly 30s hard, 60s easy for example on a 250m track or 60s hard/120s easy on a 500m track).

I will try out MyTrainPal & see what it suggests from the images below.

Many thanks
Dave

[Krzysztof]

Thanks, this is really helpful feedback.

I think you’re right that I probably didn’t choose the best example to show that feature. Comparing two very different rides for efficiency isn’t always the clearest use case.

What the above screenshot shows better is the part I was actually trying to highlight: whether you’re looking at one activity or comparing several, you can select the exact interval, rep set, or chart window you care about and send only that section to the AI. That way it can focus on the specific point where execution changed, rather than giving a generic opinion on the whole ride.

In terms of how this fits the workflow you described, the app can use power-duration curve, W′ / W′bal, and interval execution data to guide what to do next, not just explain what happened. By default it will usually suggest a structured workout, but because it’s AI you can also tell it to always suggest alternatives for the same objective, for example a suitable Zwift/Rouvy/MyWhoosh route or a race/simulation-style option, and include those alternatives in the workout description.

So the default output is structured, but you can shape the style of recommendation quite a lot depending on how you prefer to train.

I hope that it lines up with what you had in mind.

[David_Bannister]

That sounds great. In terms of the variety of platforms then suggesting a specific one would be a big ask indeed but also relatively little leg work really to match a structured workout to another’s execution platform’s own choices or even a course/hill profile of a route to see doing if the structured workout on that route is feasible.

Is the AI powerful enough yet or the Intervals API accommodating enough to allow it to see the power and/or pace curves for previous years or just the current one. So if I have selected power curves for successive years, like in my example below, could I ask it to comment on an activity or group of activities from this year comparing it/them to what I was able to do based on a previous year’s power curve? Or would it have to be comparing it to the best effort activity from that previous year, which likely seeded the power/pace curve anyway? So perhaps little difference in reality. But I am thinking that this best effort is probably easier to find in Intervals (blue writing & a direct link to the activity if clicked on) than in other analysis/recording platforms perhaps e.g Training Peaks. If it is not yet able, then is this something that you might consider for your roadmap in terms of the ease of highlighting year by year improvement or deterioration?

[Krzysztof]

Yes, the system is capable of doing it (example in screenshot). I’ve set arbitrary limit of max 180 days allowed to import from Intervals, but I can change it any moment and multi year queries will work as you described.

[David_Bannister]

I think that 180 days is likely sufficient to import probably. It is more about getting the previous power/pace duration curve probably that would be useful. And that would hopefully not burn anywhere near the numbers of tokens that importing individual activities would require. So whilst it might be great to compare a 2023 marathon to a 2026 one there might be less benfit than comparing the 2026 one to the pace curve from the 2023 which would contain the 3-4hr best effort tail of the marathon probably. Unless they bonked/hit the wall at least.

The cheaper alternative, especially if people are using intervals anyway, is to import a screenshot from their Compare page, for MyTrainPal to analyse. That way it would pick up years or seasons or quarters to compare against. Whatever the individual’s referecne durations are. And whatever metrics they wanted to compare against e.g. the current workout etc

That is where comparison becomes difficult or time consuming. And so where AI coaching can probably add a lot of value.

[Krzysztof]

It’s more a limitation driven by database cost, AI gets precomputed values regardless if you ask about a year or a week. Token cost will be the same. But it’s not really relevant as the plan is to remove that import limit in the future and AI will be able to do that multi-year analysis for you.

[Krzysztof]

Hi everyone,

It’s been quiet here for a while, which is actually a great sign, new users are signing up every day and don’t feel the need to report any issues

But I’m back with a big announcement. After thorough testing, the Android version of MyTrainPal has been officially released and is available to everyone! MyTrainPal on Google Play

To celebrate, MyTrainPal also received a new landing page, easier on the eyes and friendlier for more casual users. I invite you to check it out.

MyTrainPal.app

But it doesn’t end there. MyTrainPal now offers a 7-day free trial of the PLUS subscription. The app offers so much that I wanted to give you room to keep exploring even if the initial planning session exhausted your free-tier credits.

Few notable improvements that were shipped over last few weeks:

• Attach photos: Chat window allows now to attach photos to the conversation.
• GPX upload: Create workout tailored for your route.
• Scheduled prompts: very powerful tool, create scheduled prompt executions with any interval you want (daily, weekly, monthly etc.). Scheduled prompts trigger push notification when they are finished. You can configure your coach to give you daily reports, adjust weekly schedule, send morning readiness briefing. Possibilities are endless.
• Send any interval to coach: just select any part of any stream chart (power, cadence, hr, W’bal etc.) and click “Send to coach”. AI will focus its analysis on this particular interval, then you can ask follow up questions.
• Wahoo direct integration: Improved handling of ramp steps compared to integration through Intervals.icu.
• Improved integration with Intervals.icu: Daily Wellness notes, Activity Notes, Activity RPE are synced both ways to enhance even further the context of the AI coach.

Next on the roadmap:

• Garmin direct integration: for improved export of the strength workouts.
• Polar direct integration: So users don’t have to open Intervals.icu to trigger refresh.

[David_Bannister]

@Krzysztof Many thanks for adding Tymewear & Dfaa1 analysis to MyTrainPal. I will post an analysis of a couple of runs based on Tyemwear thresholds from a recent Ramp Test with the analysis from MyTrainPal in case anybody is interested in what the platform can now do with the extra metrics. It was very useful advice & much quicker than I managed doing it all manually.

Prompt
Can you compare the run that I did today with the ramp Test done on 8/3/26 & a similar 10K run done on 29/3/26 please. I am looking to see if you feel that the likely Tymewear estimated AeT (HR Endurance Zone <110bpm & HR VT1<123bpm) from the Ramp test corroborates with the decoupling of 4.1% for today’s run, (but 0% for the longer 10K run on 29/3), with the Tymewear & DFAa1 & RR/a1 data from today’s run. It felt very close to AeT in terms of RPE today & perhaps a little over it by the end of the run, moderate rather than easy probably. I could have easily carried on but was starting to fatigue a little, though the fatigue was probably more neuromuscular (legs) rather than central. Tymewear has my AeT for power at 138W which seems too low based on RPE & an equivalent HR for 138W of power. But it would somewhat back up the slight neuromuscular fatigue after a relatively short easy run. Though the longer 10K run had no decoupling & so it might just be extra fatigue for today’s run. Also running at 138W my running gait would likely be negatively impacted (it would fall between a fast walk & a jog, so likely unnatural) & my HR way below my HR AeT/VT1 of 123bpm & so I feel that I would be gaining little aerobic benefit really. I do know that runners always tend to overestimate how fast an easy run should be but 138W would just be too slow for me to maintain comfortably in terms of gait & engagement I think.

Answer - Thought for 31s
To analyze your progress and validate those AeT estimates, I’ve compared your Run Ramp Test (Tymewear) on March 8, the 10K run on March 29, and your Long aerobic run from today (April 4).

Comparison Overview

1. AeT Validation: HR vs. Power

The Tymewear estimate of 138W for AeT does seem significantly conservative based on your recent performance.

• March 29 (10K): You held 167W at 113 bpm for 71 minutes with essentially zero decoupling (0.05%). This is the hallmark of a “true” aerobic steady state. If 138W were your limit, we would have seen significant HR drift at 167W over that duration.
• Today (April 4): You averaged 160W at 107 bpm. Even with the 4.1% decoupling, your efficiency factor (Power/HR) was actually higher than the 10K run (1.49 vs 1.47), meaning you were technically more aerobically efficient today despite the drift.

2. DFA a1 & RR/a1 Analysis (Today’s Run)

Looking at the raw autonomic data from today’s session, we see a clear transition:

• 0–20 mins: DFA a1 stayed well above 0.75 (averaging ~1.1–1.3), indicating very low stress.
• ~30 mins mark: DFA a1 crossed below the 0.75 threshold for the first time as your HR hit 110 bpm.
• 30–60 mins: It hovered between 0.69 and 0.85, while HR stabilized around 111–114 bpm.

This strongly suggests that your metabolic VT1 (AeT) is likely right around 110–115 bpm, which aligns perfectly with the lower end of the Tymewear HR endurance zone (<110 bpm) and is well below the 123 bpm VT1 estimate.

3. The “Decoupling Mystery”: Intensity or Fatigue?

The fact that you had 0% decoupling at a higher power/HR on March 29, but 4.1% today at a lower intensity, confirms your RPE assessment: this was likely neuromuscular or “accumulated” fatigue rather than an aerobic breach.

When you are “freshly” fatigued (neuromuscularly), your gait efficiency drops. Your heart has to work slightly harder to maintain the same power output because your muscles are less “springy” (lower vertical stiffness), leading to that 4.1% drift even at a low heart rate.

Coaching Takeaway

• Power Target: Ignore the 138W AeT estimate for now. Your gait and engagement concerns are valid—running at 138W would likely be mechanically detrimental. Your “real-world” aerobic floor for training is closer to 160–165W.
• HR Ceiling: Your cap for “easy” runs should stay strictly under 123 bpm (VT1), but for pure aerobic recovery, staying near that 110–112 bpm mark (where DFA a1 was ~0.75 today) is the sweet spot.
• Fatigue Check: The decoupling today was a “fatigue signal.” Since the 10K was rock solid, your aerobic base is fine; you just need a bit more recovery to bring that neuromuscular snap back.

This is the Compare Activites Chart from Intervals & then the Tymewear analysis Tab for the Ramp test for perspective as to my usual manual method of comparison. It took me significantly longer than 31s & so a good addition to MyTrainpal. Again many thanks..

[Krzysztof]

David, thank you for taking the time to write this up. This is exactly the kind of use case I built the tool for, i.e. complex, multi-session analysis that would take long to do manually but where the insights really matter.

Should you need MyTrainPal to handle more data streams, please let me know. I’m always happy to extend its capabilities provided there’s a real-value use case. Like the Tymewear and DFA a1 you suggested, which I think are great enhancements, and I’m excited to see how you’re able to leverage them in the analysis of your runs.

[David_Bannister]

I don’t use them myself but Moxy & Core seem to have a following & momentum as potential useful sensors. Perhaps anybody reading this who has such a sensor might make a comment so as to the gauage the value of an integration. Plus you pobably need someone with the device to enable an integration in the first place?

Also a lot of people are asking for nutrition featues both in Intervals & the various AI Coaches, so perhaps explore if there any glucose sensors/meters that people might be using that ideally could be covered. Supersapiens is one that elite triathletes use (& SS are sponsoring some to use it). It costs about $150-200 per month & is not yet approved in the US & so it is one to watch perhaps rather than aiming to integrate it now. If it proves popular then the price will likely come down. And those using it might be able to save money on gel usage which aren’t exactly cheap if CGM monitoring actually proves it unnecessary. Although with Supersapiens they are struggling to adequately monetise it I think, perhaps hence the triathlon sponsorship. Triathletes often seem willing to spend a lot of money to shave aero or a handful of grams with expensive equipment/trisuits etc & so boost their watts. There is also the Dexcom G7 which is similar but without the athletics hype I think.

More generally though enabling the tracking & anlysis of blood glucose via manual import/entry might be worth thinking about especially if subscribers are going to be trying to finely tune their food intake against exercise demands. David at Intervals might be able to give an idea of how many users are entering their glucose levels into Intervals Wellness for example? If it is lots, then there is probably a demand for automated/enhanced analysis provision in AI apps.

[Krzysztof]

Great suggestions David, and this ties into something I’ve been thinking about.

I’d like to expand MyTrainPal’s supported data streams to cover muscle oxygenation, core temperature, and potentially glucose monitoring. The architecture makes adding new streams straightforward, the bigger challenge is validating the analysis against real-world data from someone who actually trains with these sensors.

So here’s an open offer: if anyone reading this uses Moxy, Core, or a CGM sensor and has their data flowing into Intervals.icu, I’d love to work with you to test the integration. In return, I’ll set you up with 2 free months of the Plus plan.

Just DM me or email [email protected] with what sensor you use and I’ll get you started.

[Krzysztof]

MyTrainPal now offers a free hosted MCP server for your Intervals.icu, Strava, and other training data

If you already have a Claude, OpenAI, or other AI subscription and want easy access to all your training data without hosting anything locally, this new feature makes it simple.

Just open the app’s Settings and you’ll see the new MCP Access section. There you’ll find your personal API key plus clear instructions for connecting it to Claude and Codex.

Once connected, you instantly get a rich set of training analytics, including:

• Athlete context: profile, thresholds, HR/power/pace zones, body metrics, and all your connected services
• Plan context: active training plan, current phase and week intent, targets, key sessions, compliance, and load state
• Activity queries: session lists, single-activity details, best efforts, and threshold history
• Deep activity analysis: HR, power, pace, elevation, raw time-series data, interval/lap breakdowns, and more
• Activity comparisons: overlay multiple workouts by time, distance, lap, or interval
• Training metrics: CTL/ATL/TSB trends, weekly load breakdown, volume trends, zone distribution, and power/pace curves
• Recovery metrics: daily readiness numbers, HRV, sleep, resting HR, weight, and body battery
• Visualization: it can render clean charts directly from any of the datasets above

[Krzysztof]

Polar Direct Integration

Polar integration is now available in MyTrainPal! Connect directly to your Polar device to sync activities and wellness data.

Intervals.icu users can keep integrating Polar data through Intervals.icu connection and skip this new integration entirely.

Mobile apps are now pending review on App Stores, I hope they will be done soon and Polar will also appear on Android and iOS versions of MyTrainPal.

[Ignasi_Vila_Martinez]

I’ve been playing around with the app all week and I love it. I have no issues following the suggested cycling workouts; those are great. However, the gym training part is a weak point of the app. Let me explain:

When the app creates the training plan, it gives you a title for the strength workout but doesn’t include the recommended exercises for the gym. I’ve had to ask the chat for them, and it did recommend them then, but it would be much better if the recommended exercises were already included in the text from the start.

On the other hand, the tracking for the strength sessions is based on heart rate. This is quite complicated considering that with neuromuscular training, the heart rate only goes up once you’ve finished the reps, and it only rises slightly because you’re doing just 4-6 reps with long rests between sets. I think there’s room for improvement there. In the end, I’ve loaded the strength workout into my Garmin with each specific exercise and weight, but it seems the app is only interested in my heart rate during the session.

[Krzysztof]

I completely understand what you mean — so far the app tried to squeeze strength sessions into the same format as endurance sessions. The results weren’t great, but it’s a broader problem, and even intervals.icu doesn’t really solve it.

That said, it’s been on my mind for a while and I think I’ve finally figured out how to handle strength workouts properly in MyTrainPal, in a way that stays consistent with how endurance training already works.

The AI coach now always includes sets and reps for strength workouts. These can be tracked during the workout inside the app with a dedicated live session mode.

The workout runs in the background, so you can hop over to the chat and ask questions about the exercises mid-session.

During rest periods, a timer is displayed, and the app sends a notification when rest is over so you can come back and start the next set.

After the workout is done, the AI coach can use this data to evaluate your execution and adjust future workouts accordingly — it sees how many reps you did, at what load, and how the session compared to what was planned.

These changes are now available in Web version, it will take 1-2 days for the update to be available on iOS and Android.

Let me know please if that will allow you to track strength sessions better.

[Krzysztof]

Hi,

Anthropic recently announced managed agents, which generated a fair amount of buzz and some bold claims. If you’re curious how to actually put them to use, I spent some time evaluating the feature and put together a short walkthrough.

1. Create a Vault

Go to https://platform.claude.com/workspaces/default/vaults and click “+ New Vault”. Name it something like “Athlete MCP Vault”.

2. Add a Credential

Once the vault is created, click “+ Add credential”. You’ll need the address of a publicly available MCP server that hosts your data. You can set one up for free with MyTrainPal — just log in, connect your Intervals account, and copy the MCP Endpoint and API Key.

Paste those into the Add Credential form and press “Add credential”.

If everything went right, it should look like this:

3. Generate an Agent

Head to Claude Platform. Here you can describe in plain English what you want your agent to do, and it will generate a specification for you. Make sure to mention https://mytrainpal.app/api/mcp in your prompt — this ensures the generated spec includes the MCP server block:

mcp_servers:
  - name: mytrainpal
    url: https://mytrainpal.app/api/mcp
    type: url

Once you’re happy with the result, click “Create this agent”.

4. Connect the Vault and Run

During environment setup, you’ll be asked which vault to use — select the Athlete MCP Vault you created earlier.

From there, you can create a test run. I asked it to generate a report for today:

What’s next?

At this point the agent has access to your training data in a structured, queryable way. You can extend it however you like — have it send reports via email, Telegram, or WhatsApp, build a custom readiness evaluation algorithm, and so on. Managed agents are flexible, though how far you take it will depend on the specifics of your workflow.

[Ignasi_Vila_Martinez]

I have already created the training session for next week thanks. I will use them and see how it goes. Thanks

[Ignasi_Vila_Martinez]

Strength training at the gym with the plan has been going well. The recovery countdown worked for all exercises except for the mobility at the start and the main exercise, which was back squats. For the rest of the exercises, it worked perfectly."

[Krzysztof]

Thank you so much for trying it out and sharing your experience!

I’ll look into the issues you mentioned and make sure they are fixed. But I’m glad to hear you liked this new feature overall.

If you feel it can be still improved in any way, please let me know.