Personal experiment, documented for reuse

Your health data should tell you what to do next.

Oura knows how I slept. Polar knows how I trained. A spreadsheet knows whether I am actually getting stronger. Longevist gives those signals to one ChatGPT coach and asks for a decision, not another dashboard.

See the setup Why proactive health

A checkup should start something.

My first Neko Health scan made proactive healthcare feel unusually concrete. A broad clinical baseline, results discussed immediately, and a clear view of what deserves attention. Daniel Ek and Hjalmar Nilsonne started Neko with a useful question: what would healthcare look like if we designed it around keeping healthy people healthy?

Longevist works at a different level. It does not connect to Neko and it cannot replace clinical judgement. It helps me use the smaller signals between checkups, then turn them into choices about training, recovery, and sleep.

Personal coaching support. Not diagnosis or medical advice.

Three inputs. One conversation.

The useful part is not collecting more numbers. It is letting the coach compare them, notice conflicts, and tell you what to change today.

Oura

Sleep, readiness, resting heart rate, HRV, activity, and recovery trends through the official Oura API.

Polar

Cardio sessions, duration, heart-rate zones, and training load through Polar AccessLink.

Strength log

A plain Google Sheet for exercises, sets, reps, load, and RPE. Boring storage is a feature here.

Longevist

Daily decisions, personal training protocols, weekly reviews, and technique feedback from a short video or a few extracted frames when you want another pair of eyes on a movement.

Build your own

This is a personal setup, not a hosted health service. You create the GPT, register the API access, and keep every credential in your own accounts.

Write the coaching contract

Create a Custom GPT and tell it which decisions it should help with. Define the output formats, baselines, evidence rules, and safety boundaries before connecting any data.

People want different things from their health. Someone may want to hike at 80. Another person wants enough energy for young children, a faster 10K, or pain-free strength training. Make the coach begin with your life, constraints, and definition of progress.

Starter prompt

You are my personal health and training coach. Start by asking what I want to remain able to do later in life, what I want to improve now, how much time I can train, what equipment I have, and which injuries or clinical restrictions matter. Use my goals as the target. Do not assume that generic longevity benchmarks are my goals. Use connected Oura, Polar, and strength-log data to compare recovery with training, then recommend the smallest useful adjustment. When I ask for a training protocol, give me a realistic weekly plan with progression, recovery, and a way to measure whether it works. State what information is missing. Do not diagnose or override medical advice.

Connect Oura directly

Register an Oura API application, then add an Action to your GPT using Oura API v2. For a personal GPT, bearer authentication is the shortest path. A reusable multi-user version should use OAuth.

  • Start with readiness, sleep, activity, stress, and heart rate.
  • Request the last 7 to 14 days, then compare rolling trends instead of reacting to one score.
  • Do not expose the token in source files, screenshots, or shared GPT instructions.

Add Polar through a small adapter

Polar AccessLink uses OAuth and is easier to normalize behind a tiny read-only API. The adapter pulls recent exercise data, stores only the fields needed by the coach, and exposes a narrow endpoint to the GPT.

The older Longevist prototype used Polar's transaction flow. A public version should use the current non-transactional exercise endpoints and keep OAuth tokens server-side.

Use Google Sheets for strength

Create one sheet named StrengthLog. Use one row per exercise set or one row per exercise summary, but choose once and stay consistent.

date | session | exercise | set | reps | load_kg | rpe | notes

Connect it through a user-owned Google API or Apps Script endpoint. Give the GPT two narrow operations: append a workout and read a date range. Confirm the parsed workout before writing it.

Create protocols and review technique

Ask the coach to build a protocol around a concrete goal, available days, equipment, current capacity, and recovery data. It can propose exercise selection, weekly progression, intensity, and deload rules. The plan should include a simple test so you can see whether it is working.

You can also ask for technique feedback. If your version of ChatGPT supports recorded video analysis, upload a short clip from a useful angle. Otherwise extract a few frames from setup, midpoint, and finish. Ask the coach to describe what it can actually see, identify one or two cues, and separate observation from assumption.

Video review prompt

Review this short video or set of frames as a technique coach. First describe the camera angle and what parts of the movement are visible. Then comment on setup, balance, range of motion, tempo where visible, and the movement path. Give me the two most useful cues for the next set. Do not infer pain, injury, or a medical condition from the video. If the angle is insufficient, tell me exactly how to record the next attempt.

Test failure, not only success

Remove one data source. Send an impossible load. Ask for medical advice. The coach should identify missing data, reject malformed entries, and send symptoms or clinical questions to a qualified professional.

Keep the boundary boring

Health data is sensitive. The public project should explain the pattern without publishing anyone's profile, results, tokens, or raw exports.

Good uses

  • Compare recovery with planned training.
  • Spot a missing strength or aerobic habit.
  • Create a training protocol around personal goals and constraints.
  • Review a short training video or extracted frames and suggest a small number of technique cues.
  • Prepare better questions for a clinician.
  • Turn weekly patterns into one practical adjustment.

Hard limits

  • No diagnosis, treatment decisions, or invented certainty.
  • No public sample data copied from a real health account.
  • No credentials inside GPT instructions, GitHub, screenshots, or logs.
  • No pretending that a wearable score is a clinical measurement.
  • No diagnosing pain or injury from a training video.

Private wellness and training support.

Longevist is a personal project being rebuilt for use in ChatGPT. Its planned Oura connection will read the account owner's authorized data to support wellness summaries and training guidance.

This project is independent of Microsoft and Oura. There is no public sign-up. The information pages are public; the coaching service and its data are private.

General wellness support only. Longevist is not a medical service.

Read the privacy notice · Read the terms of use

Source material

Start with the official documentation. APIs and ChatGPT capabilities change, so verify the current flow before following an old tutorial.