tri · ai

Forensic analytics for serious triathletes.

Forensic analysis of every swim, bike, run and race, projections, and fuel timelines. Plus a coach that reads how your training actually goes and proposes changes you approve, never behind your back. Built on the data your watch already captures. No hidden math, no vibes.

SwimBikeRun

Closed beta · Garmin or TrainingPeaks required · your data stays yours

Already invited? Sign in
tri-ai.app/program
70.3 Pontevedra
13 Sep · 70.3 · 15-week build
15 weeks out
peak 68 CTL · race-day form +8
Base
Build
Peak
Taper
Weekly loadprogressive ramp, recovery weeks dropped
Week 9 · Build9.2h · 118 km · 600 TSS
Mon
rest
Tue
KEY
Threshold swim
3.2 km · 60 TSS
Wed
KEY
VO₂ intervals
1h15 · 95 TSS
Thu
Tempo run
13 km · 78 TSS
Fri
Technique
2.4 km · 30 TSS
Sat
KEY
Long endurance
3h · 165 TSS
Sun
Brick run
18 km · 120 TSS

Adaptive coaching · coming to beta

A coach that adapts to your training and leaves you in control.

It reads how every key session actually went, and when the plan needs adjusting it proposes the smallest change that fixes it, with the reasoning, for you to approve. Your plan is never rewritten without you.

It never changes your plan behind your back

Other adaptive apps silently rewrite your week overnight, the #1 reason athletes quit them. tri-ai proposes; you review and approve every change. Nothing goes to your calendar without you.

Every change comes with the reason

Nothing unexplained. Each suggestion carries a one-line physiological rationale: move Thursday's intervals, ease Saturday's ride, and why. The thing athletes leave other platforms to find.

It reacts to what you actually did

Every interval judged against your prescription, not "assumed complete" like the calendar apps. A blown or missed key session is seen, and the coach responds to it.

What you get

Lap-by-lap forensics

Every interval, every drift, every dropped pace flagged against your prescribed plan. No more 'felt good': see the truth.

Per-workout AI analysis

Claude reviews each session and writes a forensic note across 5 sections: what worked, what didn't, what to adjust. Pick the voice that fits: balanced, conservative (Friel-leaning), aggressive (FasCat-leaning), or race-week taper.

Race Scout: who am I racing?

Paste in the race's public start-list link and Race Scout works through the field, looks up how each competitor has been ranking, and tells you who is ahead of you and who is behind on paper. Pre-race intel, not guesswork.

Race-day forensics

After every race, drop the chip-time URL and Race Analyser builds a per-leg debrief: where you executed, where you bled time, where you sat in your age-group. Per-discipline lap charts, body battery + stamina drift, power curve on the bike, all benchmarked against YOUR training fitness, not generic age-group numbers.

Ask Coach

A conversational AI scoped to your training data alone. "Why has my run pace plateaued the last six weeks?" → grounded answer with actual numbers, not generic advice.

Form & fitness curve

CTL · ATL · TSB tracked daily so taper and peak hit when you need them, not when you feel them.

Brick-derived race pacing

Race-day projections that calibrate from YOUR brick workouts, not a generic +8s/km penalty.

Daily readiness

Body battery, sleep score, HRV, and the day's planned intensity feed a single GO / MODIFY / SKIP recommendation. No more 'I feel OK' guesswork: the data tells you when to back off.

Race-day fuel timeline

Pre-during-post carbs scaled to your weight, ride duration, and goal intensity. Minute-by-minute intake schedule for race morning, when to take the gel, when to drink, when to top up.

Race Analyser · Olympic-distance debrief

Finished the race. Now see what actually happened.

Upload the .fit + the chip-time URL. tri-ai stitches the watch data with the official splits, ranks you against your age group, and writes the per-leg debrief, grounded in YOUR training fitness, not generic age-group benchmarks.

Southport Triathlon
2:23:05chip time54.6 km · avg HR 158
vs M30-34 field
14 / 73 (top 19%)
  • 🏊
    Swim25:58

    1.5 km · 1:43 / 100m · HR drift +14 bpm rep 1→12 · fitness-bound

    B+
  • 🔁
    T12:39

    chip time · 50 s slower than your 3-brick median

    B
  • 🚴
    Bike1:11:32

    NP 209 W · 89% race-day FTP · 7% fade over second half (220→200 W)

    A−
  • 🔁
    T22:04

    chip time · in line with median

    B+
  • 🏃
    Run41:22

    4:01 / km · Z3/Z4 throughout · off-bike penalty +4 s/km vs standalone

    A

Cross-cutting takeaway

The fade is on the bike, not the run, bike-leg durability is the limiter. Next block: 3× sub-threshold (88-93% FTP) rides per week for 4 weeks, no all-out efforts on the bike until race week.

Example output, your debrief uses your own thresholds, your own brick history, and your actual results page.

Common questions

▸What does it integrate with?
Garmin Connect or TrainingPeaks for activities and wellness data, Anthropic Claude for AI coach analyses, and Open-Meteo for race-day weather. See the About page for the full list.
▸Is my data safe? Where is it stored?
Athlete data is stored in the EU (London region, eu-west-2), encrypted in transit and at rest. We don't sell or share your data. We use privacy-first product analytics (PostHog EU, no IP collection, no session replay) to understand which features get used. No ad-tech, no third-party trackers. Full details in the Privacy Notice.
▸Do I need to ditch my current platform or my coach?
No, and you're in control either way. Use tri-ai as your adaptive coach (AI plan-building is coming to the beta), or keep your existing plan delivery and human coach and let tri-ai add forensic post-session analysis, race-day projections, and fuel timelines on top. Whatever you choose, tri-ai never rewrites your plan without your say-so.
▸What's different from Garmin's own Connect app?
Garmin Connect tells you what happened. tri-ai tells you whether each interval held against your prescription, where pace drifted, what off-bike penalty to plan for, and what to adjust next time, at lap-level resolution, with an AI coach narrative on demand. It pulls from the same Garmin data you already have. And there's no chatbot in Garmin Connect that can answer cross-session questions about your training history.
▸What's different from Strava or Intervals.icu?
Strava is social-first (segments, kudos, route discovery). Intervals.icu is workout-builder-first with strong analytics, but no per-session AI narrative and no conversational layer. tri-ai sits in a different lane: forensic analysis of what you did, race-day projection of what you'll do, and a conversational AI that ties them together. We don't try to replace your plan delivery or your social feed, we make the data underneath actually readable.
▸What can Ask Coach actually answer?
Anything Coach can reason about from your training data, trend questions ("has my pace improved over 12 weeks?"), comparison questions ("am I stronger now than two months ago?"), recovery questions ("am I overtrained?"), compliance questions ("am I actually hitting my prescribed sessions?"). Every number it gives you is looked up from your own training history rather than guessed at, and if it cannot find the answer in your data it says so instead of inventing one. Three questions a day during the beta.
▸Is my data used to train AI models?
No. We are on a zero-retention agreement with Anthropic, so nothing we send is kept once your answer comes back, and none of it is ever used to train their models or anyone else's. What gets sent is only the session you asked about, or the two or three same-day sessions in a brick, along with your training thresholds. Not your name, not your email. Full sub-processor list with regions and data scope at /privacy.
▸How much does it cost?
Free during beta. Paid plans launching once we exit beta, existing users get at least 30 days' notice and a loyalty discount. See Pricing for the four written commitments.
▸Do I need a Garmin watch?
No. tri-ai reads from Garmin Connect or TrainingPeaks, and most of our beta athletes are on TrainingPeaks. Either one gives us your activities with lap-by-lap detail, your daily wellness (sleep, HRV, resting heart rate) and your thresholds. You do need one of the two: without a connected source there is nothing for tri-ai to read, so it won't be useful to you yet. We're evaluating Wahoo, Polar and Apple Watch as the user base grows.
Garmin or TrainingPeaks
official source data
Encrypted
in transit + at rest
EU hosted
London / eu-west-2
Zero-retention AI
Anthropic ZDR tier
No ad-tech
no data sold, no pixels