An endurance athlete in dark training kit, lit softly against a dark blue-grey backdrop.

The Primed Story · 5 minute read

PrimedThe coach that learns you.

AI-powered personalised endurance coaching.

A persistent coaching relationship built from your training, recovery, goals, feedback and response over time.

Try Primed (opens the App Store in a new tab)See how it works

MediaPartnershipsOn iPhone, iPad and Mac. Android coming soon.

Athletes have more data than ever. They still have to coach themselves.

Watches, rings and apps each measure part of the picture. Turning all of it into today’s decision is still left to the athlete.

  • Sleep
  • HRV
  • Training load
  • Power
  • Recovery
  • Race goals
  • Feedback

What should I do next?

Everything is measured. Nothing decides.

A coach, not another dashboard.

Primed continuously turns history, recovery, training and feedback into the next decision.

TODAY What should I do?

Primed Today on Mac. The decision reads ‘Aerobic maintenance with recovery priority’, above today’s move, a shortened 65-minute Zone 2 Zwift endurance ride with a 145 bpm heart-rate ceiling and two stop conditions.
Today on Mac: the decision, the session and its boundaries.Full size screenshot (opens in a new tab)

COACH Why, and how did it go?

Primed Coach on iPhone. The athlete says they dropped the last part of a ride by 10% to keep heart rate in the 140s; Primed replies that this was a smart call and reads the ride’s power fade and 12.4% cardiac drift.
Coach on iPhone: the athlete explains; Primed reads the ride.Full size screenshot (opens in a new tab)

PLAN What changes next?

Primed Schedule showing week 10. Two sessions are marked Adjusted, with the original sessions struck through beneath the shortened replacements.
The plan: adjusted sessions, with the original still visible.Full size screenshot (opens in a new tab)

Real Primed screens from the founder’s own training account. Cropped, never edited.

One athlete. One evolving model.

The Athlete Digital Twin is Primed’s running model of you. It is not a body scan or an avatar. It is what Primed has learned about how you train, recover and respond, kept current and used for every decision.

  • PhysiologyHeart rate, HRV and the ranges that are normal for you.
  • Training historyWhat you have done, week after week.
  • RecoverySleep and readiness, read against your own baseline.
  • FatigueHow load builds up, and how you usually carry it.
  • Behavioural patternsWhen you train, when you skip, when you push through.
  • Subjective feedbackHow sessions felt, in your words.
  • Goals and eventsWhat you are training for, and when.
  • Coaching memoryWhat was recommended before, and why.
  • ConstraintsTime, niggles, travel and preferences.
  • Data qualityWhich sources to trust, and what to do when they disagree.

Many longitudinal signals resolve into One current athlete state

Every decision becomes context for the next one.

  1. UnderstandBring history, recovery and goals together.
  2. DecideChoose today’s session, and explain why.
  3. TrainGive clear targets and boundaries.
  4. ObserveRead what happened, and how it felt.
  5. AdaptChange what comes next.
  6. → Next decision

Illustrative example

  1. Planned

    A threshold session is on the plan.

  2. New evidence

    The athlete reports unusually heavy legs.

  3. Primed re-evaluates

    It weighs the evidence together:

    • Recovery signals
    • Recent load
    • Athlete feedback
    • Plan context
  4. The recommendation changes

    The session is reduced to steady aerobic work, and the reason is explained.

  5. Context for next time

    How the athlete responds becomes part of the next decision.

A simplified illustration of how Primed reasons. Not medical advice.

In the app

A Primed decision card labelled Steady and shortened: ‘Aerobic maintenance with recovery priority’, explaining that intensity is kept low and duration trimmed because of elevated heart-rate markers.
A real decision: intensity kept low, duration trimmed.Full size screenshot (opens in a new tab)
Primed Coach on iPhone: ‘You faded late in today’s 72-minute ride. With your training load rising, I’ve trimmed tomorrow’s tempo session to 77 minutes to ensure you absorb the work.’
Primed explaining a change before the athlete asks.Full size screenshot (opens in a new tab)

A data import can tell you what happened. It cannot recreate the coaching relationship.

Day-one data import

What happened

  • Activities
  • Power
  • Heart rate
  • Sleep
  • Readiness
  • GPS
  • Historical load

Primed over time

What was decided, what happened, and what it meant

  • What was prescribed
  • What was actually executed
  • How the athlete responded
  • Subjective effort
  • Relevant constraints
  • Prior decisions
  • What was changed
  • Which signals proved reliable
  • What worked
  • What failed
  • How the athlete responds to coaching

A competitor can import the athlete’s history. It cannot import the history of the coaching relationship.

The model can change. The athlete relationship persists.

Primed keeps the persistent athlete system separate from the foundation-model layer. The part that knows you belongs to Primed. The model that reasons and writes can change as better ones arrive.

Foundation model layer

Replaceable · improving
  • Gemini
  • OpenAI
  • OpenRouter

Different work can use different models

  • Coach conversation
  • Summarisation
  • Embeddings
  • Other reasoning tasks

Persistent Primed system

Accumulating · yours
  • Athlete Digital Twin
  • Coach Memory
  • Constraints
  • Adherence
  • Decision history
  • Feedback
  • Outcomes

Better models increase the value of the context Primed already holds.

Under the hood
  • Model access runs through one Primed service layer, so a model can be switched for the whole product or chosen per task.
  • Current providers are Google Gemini, OpenAI and OpenRouter.
  • The Athlete Digital Twin and Coach Memory live in Primed’s own system, not inside any one model. Changing a model does not reset what Primed has learned about an athlete.

Facts first. State second. AI reasoning third.

Primed does not ask an LLM to invent the athlete’s physiology.

  1. Step 1: Observe + calculate

    Deterministic

    Known facts and deterministic calculations.

    Activities, recovery data and training load come from connected sources and are calculated in code.

  2. Step 2: Resolve + decide

    Committed state

    Committed athlete state, constraints and policy.

    Primed settles the current state and the rules that apply before a language model is involved.

  3. Step 3: Explain + interact

    AI reasoning

    Natural-language coaching.

    The model explains the decision and talks it through with the athlete.

Objective facts are resolved before generative reasoning.

This makes Primed more consistent, not infallible. Recommendations come with their reasoning, so an athlete can question them.

An AI coach has to behave like one coach.

The failure mode

Today
Recovery
Coach
Hard workout
Widget
Old session

This is split brain.

Primed’s approach

  1. Controlled athlete state
  2. Canonical coaching truth
    • Today
    • Coach
    • Plan
    • Widgets
    • iPhone
    • iPad
    • Mac
  4. Do they agree?

  • Facts

    Deterministic assertions

    The numbers and states each surface shows are checked against one truth.

  • Behaviour

    Synthetic athlete scenarios

    Whole athlete situations are replayed from start to finish.

  • Meaning

    Semantic audits

    Does the coaching say the same thing everywhere it appears?

Technical details
  • 15 golden regression cases across safety, divergence, decision, evidence and regression, all passing in the coaching pipeline gate.
  • 31 end-to-end athlete scenarios covering time, health, compliance, schedule, lifecycle and data.
  • 36 recorded schedule-incident patterns replayed as mutants, to prove the checks catch the failures they are meant to.

Already live. Already learning.

Primed is a production product on the App Store today, on iPhone, iPad and Mac, with connected training and recovery services.

Platforms

  • iPhoneLive
  • iPadLive
  • MacLive
  • AndroidComing soon

Connected services

  • GarminLive
  • WHOOPLive
  • OuraLive
  • Apple HealthLive
  • WithingsLive
  • WahooComing soon
Primed Schedule on Mac, showing the current training block and the week’s planned and adjusted sessions.
Schedule on Mac.Full size screenshot (opens in a new tab)

Early recruited test cohort

hardware-verified connected athletes
16
multi-day continuity
15 / 16
Coach conversation turns over 14 days
1,382
plan adjustment events over 14 days
1,000+
athletes providing subjective feedback
15

Founder-recruited testing cohort. Early engagement evidence, not product-market fit.

An illustrative runner standing after a session, faint trajectories behind him.

Illustrative athlete

Built for where AI is going.

The modelkeeps getting more capable×The Primed systemkeeps accumulating athlete context=Better coaching

Advances in AI improve two things at once:

  1. The intelligence coaching the athlete.
  2. The intelligence building Primed.

Primed can adopt better models without replacing the coaching architecture.

Primed is built with AI throughout its own engineering, so better models also shorten the time between an idea and an athlete using it.

PrimedThe coach that learns you.

A one-to-one coaching relationship for a world where intelligence scales as software.

Download the Primed Story (PDF · 12 pages) On Android? Join the waitlist on the home page.