Completed workouts in STAS

STAS receives completed activities from Intervals.icu and saves them as workout records. These records support the website, reports, current condition, strategy reviews, and chats with ChatGPT or Claude.

A planned workout says what should happen. A completed workout says what actually happened. STAS compares those two only after the real activity arrives.

Open workout history from your profile. Each workout opens a detail page with metrics, intervals, zones, wellness context, and the report editor.

In your watch app and in Intervals.icu, you see the source activity data: route, time, distance, heart rate, pace, power, zones, and any other fields your device or platform provided.

In STAS, the workout appears in your profile and workout history as a prepared workout card: session details, intervals and metrics when available, plus a field or button for your post-workout report.

After STAS processes the workout, it sends an automatic Telegram message. From that message, you can quickly write a report without opening the detailed web page.

That report is saved to the STAS workout and to the Intervals.icu activity description. Telegram does not replace the workout page; it is a fast way to add your comment right after the session.

STAS keeps the important fields, cleans up noisy details, normalizes the metrics, and keeps the original Intervals activity as the source behind the workout record.

STAS also adds coaching context. It does not just store numbers; it helps explain what those numbers mean for training.

This context helps ChatGPT and Claude understand the workout as a training session, not just a list of values.

• session type: running sessions are labeled Easy, Threshold, Interval, Long, Recovery, or Unclassified when there is not enough data; cycling, swimming, and strength work usually keep the sport activity type;
• key parts of the session: for example, 4 x 8 minutes at threshold or 5 x 1000 m, with pace, heart rate, power, and recovery when the data is available;
• splits and zones: kilometer splits by pace, time in heart rate zones, and power zones for cycling;
• athlete report: what actually happened, how the load felt, and why the workout differed from the plan;
• link to the original Intervals.icu activity: STAS keeps the source behind the record, but shows the prepared record instead of the full raw data set.

This is the point of STAS: turn the activity into clear training facts before you ask for analysis, so ChatGPT or Claude can focus on the workout instead of decoding raw data.

• save the useful workout fields;
• clean and normalize values for comparison;
• mark meaningful work segments and intervals when the data supports it;
• attach the user report to the same workout;
• build short summaries for weeks, months, trends, load, and key sessions.

The response contains the id, date, activity type, session type, metrics, zones, splits, athlete report, and sport-specific fields. When a request uses full=1, STAS also includes intervals.

{
  "id": "trn_2026_04_06_run_threshold",
  "date": "2026-04-06T07:12:00.000Z",
  "workoutType": "Run",
  "sessionType": "Threshold",
  "activityName": "Morning threshold run",
  "distance": 12.4,
  "movingTime": 3660,
  "pace": 4.92,
  "avgHr": 151,
  "maxHr": 174,
  "fitness": 52.1,
  "fatigue": 61.4,
  "form": -9.3,
  "load": 84,
  "intensity": "moderate",
  "elevationGain": 86,
  "hrZoneTimes": [
    320,
    1180,
    1470,
    610,
    80
  ],
  "splitsKm": [
    {
      "km": 1,
      "pace": "5:10/km"
    },
    {
      "km": 2,
      "pace": "5:05/km"
    }
  ],
  "intervalSummary": {
    "work": "4 x 8 min",
    "recovery": "3 min jog"
  },
  "userReport": "Completed as planned. Last two reps felt hard but controlled.",
  "sportMetrics": {
    "gradedPace": 4.86,
    "cadence": 176,
    "verticalOscillation": 8.4
  },
  "intervals": [
    {
      "name": "Work 1",
      "moving_time": 480,
      "distance": 1640,
      "avg_hr": 164
    },
    {
      "name": "Work 2",
      "moving_time": 480,
      "distance": 1660,
      "avg_hr": 168
    }
  ]
}

The base record stays the same, but the extra details depend on the activity type: running has one set of fields, cycling adds power and FTP, swimming has its own details, and strength training may have fewer sport-specific fields.

{
  "Run": {
    "workoutType": "Run",
    "sessionType": "Interval",
    "sportMetrics": {
      "gradedPace": 4.18,
      "cadence": 180,
      "verticalOscillation": 8.1
    }
  },
  "Ride": {
    "workoutType": "Ride",
    "sessionType": "Ride",
    "sportMetrics": {
      "avgWatts": 188,
      "normalizedPower": 215,
      "ftp": 260,
      "intensityFactor": 0.83,
      "tss": 96,
      "powerZoneTimes": [
        780,
        2200,
        1800,
        900,
        240,
        0,
        0
      ],
      "cadence": 84,
      "decoupling": 3.2
    }
  },
  "Swim": {
    "workoutType": "Swim",
    "sessionType": "Swim",
    "pace": 1.82,
    "sportMetrics": {
      "swimCadenceSpm": 31
    }
  },
  "WeightTraining": {
    "workoutType": "WeightTraining",
    "sessionType": "WeightTraining",
    "movingTime": 2700,
    "load": 31,
    "sportMetrics": null
  }
}

In ChatGPT, this is a request to the STAS Actions endpoint. A workout review usually uses a date window and full=1 when detailed intervals are needed.

GET /trainings?oldest=2026-04-01&newest=2026-04-08&limit=30&full=1

In Claude, the same request goes through the get_trainings tool in the STAS connector.

{
  "name": "get_trainings",
  "arguments": {
    "oldest": "2026-04-01",
    "newest": "2026-04-08",
    "limit": 30,
    "full": true
  }
}

STAS instructions tell ChatGPT and Claude to work from fresh data, not from chat memory alone.

• read the athlete summary first to understand profile, goals, rules, condition, and compact history;
• for specific workout analysis, load full workouts directly because the recent-workout preview is only a short overview;
• for detailed analysis and exports, use full=1 to include intervals, and split the date window if the response is too large;
• compare plan versus fact only after reading planned events for the same period.

When you ask for a workout review, ChatGPT or Claude can look at the facts of that session and compare them with context:

• what was actually done: duration, distance, activity type, and session type;
• how the effort played out: pace or speed, heart rate, power, zones, and training load;
• which parts mattered most: intervals, work segments, recovery, and splits;
• what the athlete added in the report and what that day’s recovery data shows;
• how this workout fits the plan, current condition, and recent history.

ChatGPT or Claude do not need to decode every raw activity detail for every reply. STAS has already prepared the useful training context.

A regular message in ChatGPT or Claude is not saved as a workout report. Reports are saved through STAS features such as Telegram or the workout page.

They are the facts STAS works from. Completed workouts show what actually happened, so ChatGPT or Claude can review training and help plan from real sessions, not guesses.

No. STAS receives full activity data, then saves and prepares a workout record. ChatGPT or Claude receives the prepared context: useful fields, intervals, report, metrics, and history summaries.

The record stores the date, name, activity type, session type, distance, time, pace or speed, heart rate, load, Fitness/Fatigue/Form, and your report. Some values stay empty if the source did not provide them.

Power, cadence, power zones, swim-specific fields, wellness, intervals, and some evidence fields depend on what Intervals.icu received from the device or platform.

STAS builds compact summaries for recent workouts, weeks, months, trends, load, sport mix, and key sessions. That gives ChatGPT or Claude context beyond one activity.

No. STAS can help notice training signals and missing context, but it does not make medical diagnoses and does not replace a qualified professional.