perf(record): keep one speech model loaded — 2948 ms → 77 ms to start recording - #13

Merged

meocong merged 3 commits into

feat/warm-models-and-compose

Aug 14, 2026

Commits on Aug 14, 2026

perf(record): keep one speech model loaded, so pressing record starts in 77 ms instead of 3 seconds
meocong
committed
80a22e1
fix(record): a build without recognition has nothing to warm, and that is not an error
meocong
committed
e818b5a
fix(record): do not ask a daemon that cannot warm anything
meocong
committed
ea9f73e