sextant
Bring your own AI · the clean pipe

You screenshot your run into ChatGPT. It only sees the averages.

To get your run to your AI today, you screenshot a chart and paste it. That bridge is lossy, slow, and error-prone — your AI reasons on a thumbnail of your run. Sextant is the clean pipe: the whole structured run, straight to the AI you already pay for.

No screenshot. No typing. Free for a run.

Get started →
Same run · same question

One AI got a screenshot. The other got the run.

You finished a tempo set — five times one kilometre at 3:55/km — and you ask your AI the same thing either way. The only difference is what reached it.

You → your AI
Was I actually faster tonight, or just working harder than last winter?
What you do nowa screenshot
Fitness · Splits5.0 km
KMPaceAvg HR
13:56151
23:55153
33:55154
43:55155
53:56156
Avg3:55154
ChatGPT
From this summary, your five reps are even — 3:55/km throughout — with average heart rate edging up to 156 on the last one. Looks like a solid, controlled tempo. I can't really tell if you're fitter than last winter from this, though — it's just tonight's averages.
a picture of the averages → it can describe tonight. Nothing more reached it.
With Sextantthe run, structured
sextant · 5×1K tempofull run + context
set 5×1000m · 3:55/km · HR/rep ▂▃▄▅▆ streams hr·pace·cadence per sec · src gps+stream same effort, last winter HR 162 → now 154 rep 5 decoupling +4% · load 3 hard days
Your AI · same run, read in full
Faster — not just working harder. Same 3:55/km reps as last winter, but your heart held 154 instead of 162 — eight fewer beats for the identical pace. The engine grew, the effort didn't. One honest note — rep 5 drifted, HR climbing while pace held, which lines up with three hard days this week. A little fatigue on the last rep, not a slowdown.
the whole run, with its sources and context → it answers, and shows its work.

Same run. Same question. The screenshot carried the averages; Sextant carried the run — every stream, its sources, your history as context. The data existed either way. Only one of them reached your AI.

Why the screenshot fails

A screenshot is a broken pipe.

Your run already exists in full — every second of it. A screenshot throws almost all of it away on the way to your AI, three times over.

01 · Lossy
A screenshot

Only the averages on the screen. The per-second heart rate, pace, and cadence — the shape inside every rep — never make it. Your AI gets a thumbnail of the run.

Sextant

Every stream at full resolution, with the sources it came from. Your AI reads inside the run, not a summary of it.

02 · Slow
A screenshot

Crop a chart, paste it, then re-type your context — your threshold, the weather, that it was a tempo — every single run.

Sextant

One tap to export, or connect once and just ask. Your thresholds and history ride along automatically.

03 · Error-prone
A screenshot

Wrong crop, a misread number, a detail you forgot to mention. The AI reasons confidently on the wrong inputs.

Sextant

Exact values, nothing transcribed by hand. What the watch recorded is what the AI reads.

Working today · in Claude

Connect once. Then just ask.

This is Claude on a phone with the Sextant connector. It pulls the runs, the plan, and your thresholds through read-only tools, then answers with the numbers. No screenshot anywhere.

Claude's settings for the Sextant connector, listing its read-only tools: get achievements, get athlete metrics, get best efforts, get coach thread, and more.
The connector's tools. All read-only: it can't edit or delete anything.
A runner asks Claude how training is going and whether to run today. Claude answers from Sextant: yes, 35 minutes easy per the plan; 8 runs in 2.5 weeks; week 2 at 167 minutes against 145 planned; Monday at 11:18 per mile and 150 bpm versus 11:26 and 154 bpm on 9/23.
A real question, and the start of Claude's answer from real runs.
The part that surprises people

The clean pipe is free.

Hand any run to your own AI. Free.

No new coaching subscription, no second bot. Sextant structures the run; the AI you already pay for does the reading. The thing that kills the screenshot costs nothing. Every tool, and a coach read on every run, are where the paid plans begin.

  • Every run in Apple Health, back to the first one
  • Coach reads on your first 10 runs
  • All-time PRs from your whole history
  • Connect ChatGPT or Claude: core tools, 100 calls a day
Stop screenshotting

Give your AI a run it can read.

Get started →
Free iPhone beta. Works with Apple Watch and any watch that syncs to Apple Health.
We never post anything, and your runs stay private — never sold.