Good questions.

El-co-uh. It was called Velcora until October 2026, and the app's current version still uses that name.

No, and it doesn't pretend to be. Those run models far larger than any laptop can hold, on data-center hardware. LCOA is for the large share of everyday work a smaller model does well: drafting, explaining, small coding jobs, sorting files. It works offline, costs nothing per question, and keeps your conversations on your computer.

When a task is beyond the model you picked, it says so, and can pass it to a larger one you have installed.

Leave it on Triangle Auto and it picks per question, within your computer's memory. If you'd rather choose, the benchmarks page shows how every model did on the same tests, and the models page says how much memory each needs.

A 64-bit Windows 10 or 11 PC. 8 GB of memory runs the small models, 16 GB runs the everyday ones, and the largest need 32 GB or more. No graphics card is needed: every benchmark on this site was measured on a processor alone.

Yes, once a model is downloaded. Web search, downloads, update checks and a connected Google account need the internet. Nothing else does.

They are exactly what came out, and every answer is kept. The samples are small, so each score carries a range that is often 15 points or more either side. The benchmarks page has the method, the raw counts and the commands to run them again.

Only what its Reach level allows, and you set that. At Watch it only looks. Tidy, Do, Drive and Run each add a little more, and every action above Watch is shown to you before it happens. Folders need your approval too, one at a time. More on privacy and control.

So people with a Mac can try it and say what breaks. It builds and passes its checks, but nobody has run it on real Mac hardware yet, and the download page says so.