Why I built Athmex

The problem was not a lack of information. It was the constant re-explaining. A new chat meant typing my current training paces, PBs, target race, goals, injuries, equipment and preferences before I could ask a useful question.

ChatGPT seemed like a natural fit, and early on it genuinely was. I could get useful answers, but only after rebuilding the same athlete briefing in the chat.

It also forgot selectively. One conversation lost my race plan; the next remembered an old injury or used a stale PB as if it were current. I could not tell which version of me it was working from, so I stopped trusting the advice even when it was probably right.

The frustrating part was that the answers were often excellent right after I had pasted everything in. The missing piece was a reliable, athlete-controlled memory.

So I built the thing I actually needed: a place to keep my athlete profile, paces, PBs, goals, races, injuries, equipment and preferences so ChatGPT and Claude can read them instead of me retyping them. That’s Athmex. Activity connectors are optional and remain separate.

Athmex does not produce the advice itself. It gives the AI you are already using the context to make that advice more relevant, and lets you update or revoke what it can access.

I’ve used Athmex on my own training for the last six months: every session, every question to Claude, built on my actual data. If you want to see the training behind it, you can follow me on Strava.

This started as a hobby project for my own training. It is now an open beta for runners, cyclists and triathletes who want their AI to remember the details that make their training personal. I’m building it carefully and would like to hear what would make it more useful.