AI Coach¶

PeakLine ships an AI Coach that generates a short narrative summary of each activity — written in plain language, grounded in the same numbers the rest of the analysis shows.

Under the hood, it's Google Gemini 1.5 Flash (a fast, low-cost model from Google's Gemini family) given a structured prompt that includes your zones, climbs, weather, and PeakLine Score.

What it looks like¶

On any activity page, you'll find an AI Coach section near the top. A typical output:

Solid Z2 endurance ride with a couple of harder pushes on the long climb in the second half. Your wind impact was −9% (headwind on the way out), which partly explains the slightly slower outbound pace. You spent 68% in endurance, 18% in tempo, and 12% in threshold — a clean polarized session. Recovery should be straightforward; consider a Z1 spin tomorrow before another quality day.

It's not a replacement for a real coach. It's a summary that translates the numbers into a sentence.

What it has access to¶

The prompt sent to Gemini includes:

Activity type, distance, duration, elevation gain.
Zone distribution (% time in each power/HR zone).
PeakLine Score and tier.
Stress Points and load category.
Climbs — count, total elevation, longest climb, best VAM.
Weather — temp, wind speed/direction, AQI category.
Comparison to your previous 10 same-type activities (speed delta, SP delta).

It does not see your raw GPS coordinates, exact route, address, photo data, or anything from other activities.

When it runs¶

Automatically the first time you open an activity page, if you have AI Coach enabled in Settings → AI features.
On demand — click the ↻ Regenerate button to ask for a fresh take. Useful if the first one feels off.

The cost per summary is fractions of a cent. You can use it on every activity without thinking about it.

When it's not useful¶

Very short activities (< 15 minutes) — there isn't enough structure to summarize.
Indoor rides with no zones set — without a zone breakdown, the model just paraphrases distance and duration.
Manual-entry activities — without sensor data, the output is generic.

In those cases the coach card stays empty rather than serving a bad summary.

Privacy¶

If you'd rather not send any data to Google, turn it off in Settings → AI features. With AI Coach off:

No data ever leaves PeakLine for AI processing.
The activity page renders normally, just without the coach card.

When AI Coach is on, only the structured per-activity summary above is sent. Google's terms for the Gemini API do not allow them to use your data to train future models.

Roadmap¶

A few things planned (not yet shipped):

Weekly summary — Sunday-night rollup of the week's training load, biggest workouts, and what to focus on next week.
"Why did this ride feel hard?" prompt — feed your subjective rating along with the numbers and get a structured explanation.
Coach memory — let the model see your last 4 weeks of summaries instead of just the current activity, so it can pick up on patterns.

If you have ideas, the feature ideas page is the right place.