I recently wrote about my experiments with the on-device LLM Apple announced at WWDC. A few people were curious how I set up the testing rig allowing me to review model responses in batches.

My initial setup relied on me being able to edit the code of my test app directly, which was a little hard to share. So instead of explaining how that worked I made a macOS app with the bare minimum of UI that transforms my hardcoded test rig into something anyone can try.

Introducing Local Large Language Model Lab, or LLLML for short. Download it here.

The app requires macOS Tahoe 26 to run, and won’t work without Apple Intelligence. Upgrade to the latest developer beta at your own risk.

And a big caveat: I put the app together very hackily, and spent zero effort on designing the experience or the UI. I hope it lets you try the LLM, but be aware that I prioritised being able to share it quickly over anything else.