Reference
Harlen docs
What Harlen reads, what it produces, how a read is made, and what it will not do. For the short version, start on the home page. For the questions trainers ask most, see the FAQ.
Overview
Harlen is an AI assistant for personal trainers. It holds a context model for every client on a roster, reads their data against their own baseline, and does the work between sessions: the overnight scan, the block, the note, the answer, the calendar. Nothing reaches a client until the coach sends it.
Athletes connect their wearables through the Harlen iOS app, and coaches work from any modern browser.
Status
- Stage
- Beta, coaching its first rosters
- Access
- Starts with a short founder-led demo
- Pricing
- Not published. The model is not final and is discussed on the call.
- Coach surface
- Web, any modern browser
- Athlete surface
- Harlen for iOS
Product, pricing and packaging are still changing, so this page describes where Harlen is today.
What Harlen does
Five jobs. You hand one over the way you would hand it to an assistant, and it takes the job start to finish. All five are live in beta today.
| Job | What it does | Input | Output |
|---|---|---|---|
| Programming | Drafts an individualised block from the client’s own physiology and history, then exports to a spreadsheet. | Goals, injury history, last cycle, wearable data | Program draft, coach edits it |
| Monitoring | Reads every client against their own baseline overnight and raises the two or three who need a look. | Wearables, check-ins, injuries, medical history | Ranked brief, nothing sent without approval |
| Notes | Sorts what you said after a session into events and durable facts, folded into that client’s context model. | Typed or spoken note | Filed facts, carried into every later read |
| Research | Answers a training question with the conclusion first and the peer-reviewed work underneath. | Plain-language question, optionally scoped to a client | Answer with citations |
| Calendar | Places sessions, calls and reassessments, and moves them when a client does. | Roster schedule, client changes | Calendar entries |
Data sources
If a client wears a device that writes to Apple Health, Harlen can generally read it, whether or not the brand is listed here.
| Source | Kind | Connection | Status |
|---|---|---|---|
| Apple Watch | Wearable | Apple Health | Active |
| Garmin | Wearable | Direct | Active |
| Polar | Wearable | Direct | Active |
| Whoop | Wearable | Apple Health | Active |
| Coros | Wearable | Direct | Beta |
| Oura | Wearable | Direct | Beta |
| Fitbit | Wearable | Google Health | Beta |
| Withings | Scale and body composition | Direct | Active |
| MyFitnessPal | Nutrition log | Direct | Active |
| PubMed | Research library | Direct | Active |
| Your library | Coaching manuals and method standards you upload | Upload | Beta |
Beta means the connection is built and still under test.
Signals read
Every signal is measured against the client’s own history, and the reference column shows what each value is compared to.
| Signal | Covers | Read against |
|---|---|---|
| Sleep | Duration and stages | Client’s rolling baseline |
| HRV | Heart rate variability | Client’s rolling baseline |
| Resting heart rate | Morning resting HR | Client’s rolling baseline |
| Training load | Session load and accumulation | Client’s recent history |
| Workouts | Completed sessions and activity | Prescribed block |
| Nutrition | Logged intake | Client’s own pattern |
| Body metrics | Weight, composition, waist and limb measurements | Client’s own trend |
| Check-ins | Soreness, mood, session RPE, free-text notes | Client’s own pattern |
How a read is made
- Everything is measured against the client. Each marker is compared to that client’s own rolling normal. An HRV of 70 on its own says very little. An HRV of 70 for someone who has been steady at 92 for two months says a lot.
- Movement is what gets raised. A reading that drops below a client’s own baseline comes to your attention. A reading that looks low on paper but is ordinary for that person stays quiet.
- Context is applied. The client’s notes, injuries, travel, check-ins and programming are part of the same model, so the read accounts for what you already told Harlen.
- The work is shown. Every flag carries the underlying data and the reasoning behind it, and every research answer carries its peer-reviewed sources. You can always see what Harlen saw.
- You approve. Drafts wait on the coach. Nothing reaches a client until it is sent.
We wrote about the reasoning at length on the blog.
Security and data
- Inference
- Athlete data is anonymised before it reaches the AI inference layer.
- Retention
- Zero-data-retention endpoints with model providers. Athlete data is not stored, logged, or used to train anyone’s models.
- In transit
- Encrypted.
- Sale of data
- Never. Client data powers the reads inside that coach’s own account and nothing else.
- Isolation
- Each athlete has their own context model. One client’s data never colours another’s read.
Limits
What Harlen does not do, stated plainly so it does not have to come up on the call.
- Nothing reaches a client without you
- Harlen drafts messages, deloads and blocks. Sending is a coach action, every time.
- Not a medical device
- Harlen surfaces patterns in training and recovery data. It does not provide medical advice or diagnosis. Refer clients to qualified medical professionals for health concerns.
- Not a HIPAA-covered entity
- Wearable data is not PHI. The data is handled with that level of care regardless.
- It does not make the coaching decision
- Harlen does the reading and shows the work. What to change is the coach’s call.
- No published pricing
- Harlen is in beta and pricing is not final. It is discussed on the demo call once we understand the roster.