Migrating from SugarWOD to Hevy with AI

I have been tracking workouts in SugarWOD for years. It let me record my lifts and worked well for my Olympic lifting programming, but I couldn’t add custom movements. SugarWOD doesn’t know every lift, so anything I was doing outside the library became a note or a gap in my history. As I have been shifting toward more varied workouts and weightlifting programming, that started to matter more. I kept losing the history of what weight I used for some odd dumbbell movement a few weeks ago, usually right when I wanted to match it.

I did some research and found that Hevy could work since it had bigger library of movements, but more importantly, it supported custom exercises. But Hevy can’t import data from SugarWOD directly, it only reads data in the Strong app CSV format, and there was no built-in converter.

A few years ago, this is where I would have stopped. I would either eat the loss and start clean in Hevy, or stay on SugarWOD out of inertia. Writing a one-off CSV converter for a single export of my own data was never worth an evening. With AI, it became small enough to try.

Using AI outside work

This was never a technically hard migration. I knew that I probably could spend a few hours researching and learning SugarWOD’s workouts.csv format and figuring out how to map it to Strong’s per-set columns in an afternoon. I just never wanted to spend that time after a full day of work, on top of after doing my workout, and then coming back to my computer to do more of the same.

So instead I just described the problem to an AI and in a matter of minutes I had a stdlib python script. It parsed SugarWOD data, emitted Strong-shaped rows, and I could run it locally to get the new file to import into Hevy. The cost of “just build it” dropped from a weekend project to something I could do between normal Slack messages.

But not everything was perfect. The script only got correct because I kept running a validation loop manually. I exported and converted my data and when imported into Hevy I had to spot-check a workout I knew. I kept finding mistakes, so I would then tell the AI what was wrong and had it update the script. Writing this now, I probably could have automated more of that if I had used the websites instead of my phone, or if either app had an API available to free tier customers.

Migrating bugs

If anyone else wants to know what those things are, they were small, specific things I didn’t find documented anywhere obvious:

• Weights are kg in Strong CSV
  • SugarWOD logs lbs. If I copied 320 straight through, Hevy read it as 320 kg (~705 lbs displayed). I had to write kg values that round-tripped back to my actual plate weight.
• Precision matters
  • After conversion, 190 lbs can display as 189.99 depending on decimal places. The converter tries 2–5 kg decimals until the round-trip matches, since workout weights usually only increment by 2.5 or 5 lbs.
• Duplicate export rows
  • Some workouts on the same date and title had different set_details. Rather than asking AI to analyze every case, I kept it simple and deduped down to the richest row.
• Unmapped names become new custom exercises
  • Back Squat and Squat (Barbell) split my history unless I mapped them together, so I had to match those to new names in Hevy.

None of that is hard once you see it. I only saw it because I kept importing real workouts and eventually noticed my 320 lb squat now claimed I had lifted a small car 10 times over.

Now a free tool

The converter is open source: github.com/smyrick/sugarwod-to-hevy.

If you are moving your own data I would instead just ask your own AI tools to help you do it by pointing at the link above. I tried documenting everything so that non technical users (and agents) could take advantage of it, but if you are comfortable running a python script, there are instructions in the repo.

Learnings

I didn’t set out to write a tool or some unicorn app, and I didn’t need one to already exist. I just needed the bar for “just build it” to be low enough that keeping years of training history was worth ten minutes instead of a lost weekend. That’s what AI changed for me here. It didn’t make the impossible possible so much as it made the not-worth-it worth doing.

PRs welcome for exercise names the mapper doesn’t know yet.